swift/test/unit/obj/test_ssync_sender.py

1659 lines
67 KiB
Python
Raw Normal View History

Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
# Copyright (c) 2013 OpenStack Foundation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import io
import os
import time
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
import unittest
import eventlet
import mock
import six
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
from swift.common import exceptions, utils
from swift.common.storage_policy import POLICIES
from swift.common.utils import Timestamp
from swift.obj import ssync_sender, diskfile, ssync_receiver
2016-10-30 22:24:18 -07:00
from swift.obj.replicator import ObjectReplicator
2016-10-30 22:24:18 -07:00
from test.unit.obj.common import BaseTest
Add checksum to object extended attributes Currently, our integrity checking for objects is pretty weak when it comes to object metadata. If the extended attributes on a .data or .meta file get corrupted in such a way that we can still unpickle it, we don't have anything that detects that. This could be especially bad with encrypted etags; if the encrypted etag (X-Object-Sysmeta-Crypto-Etag or whatever it is) gets some bits flipped, then we'll cheerfully decrypt the cipherjunk into plainjunk, then send it to the client. Net effect is that the client sees a GET response with an ETag that doesn't match the MD5 of the object *and* Swift has no way of detecting and quarantining this object. Note that, with an unencrypted object, if the ETag metadatum gets mangled, then the object will be quarantined by the object server or auditor, whichever notices first. As part of this commit, I also ripped out some mocking of getxattr/setxattr in tests. It appears to be there to allow unit tests to run on systems where /tmp doesn't support xattrs. However, since the mock is keyed off of inode number and inode numbers get re-used, there's lots of leakage between different test runs. On a real FS, unlinking a file and then creating a new one of the same name will also reset the xattrs; this isn't the case with the mock. The mock was pretty old; Ubuntu 12.04 and up all support xattrs in /tmp, and recent Red Hat / CentOS releases do too. The xattr mock was added in 2011; maybe it was to support Ubuntu Lucid Lynx? Bonus: now you can pause a test with the debugger, inspect its files in /tmp, and actually see the xattrs along with the data. Since this patch now uses a real filesystem for testing filesystem operations, tests are skipped if the underlying filesystem does not support setting xattrs (eg tmpfs or more than 4k of xattrs on ext4). References to "/tmp" have been replaced with calls to tempfile.gettempdir(). This will allow setting the TMPDIR envvar in test setup and getting an XFS filesystem instead of ext4 or tmpfs. THIS PATCH SIGNIFICANTLY CHANGES TESTING ENVIRONMENTS With this patch, every test environment will require TMPDIR to be using a filesystem that supports at least 4k of extended attributes. Neither ext4 nor tempfs support this. XFS is recommended. So why all the SkipTests? Why not simply raise an error? We still need the tests to run on the base image for OpenStack's CI system. Since we were previously mocking out xattr, there wasn't a problem, but we also weren't actually testing anything. This patch adds functionality to validate xattr data, so we need to drop the mock. `test.unit.skip_if_no_xattrs()` is also imported into `test.functional` so that functional tests can import it from the functional test namespace. The related OpenStack CI infrastructure changes are made in https://review.openstack.org/#/c/394600/. Co-Authored-By: John Dickinson <me@not.mn> Change-Id: I98a37c0d451f4960b7a12f648e4405c6c6716808
2016-06-30 16:52:58 -07:00
from test.unit import patch_policies, make_timestamp_iter, skip_if_no_xattrs, \
debug_logger
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
class NullBufferedHTTPConnection(object):
def __init__(*args, **kwargs):
pass
def putrequest(*args, **kwargs):
pass
def putheader(*args, **kwargs):
pass
def endheaders(*args, **kwargs):
pass
def getresponse(*args, **kwargs):
pass
def close(*args, **kwargs):
pass
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
class FakeResponse(ssync_sender.SsyncBufferedHTTPResponse):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def __init__(self, chunk_body=''):
self.status = 200
self.close_called = False
if not six.PY2:
chunk_body = chunk_body.encode('ascii')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
if chunk_body:
self.fp = io.BytesIO(
b'%x\r\n%s\r\n0\r\n\r\n' % (len(chunk_body), chunk_body))
self.ssync_response_buffer = b''
self.ssync_response_chunk_left = 0
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def read(self, *args, **kwargs):
return b''
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def close(self):
self.close_called = True
class FakeConnection(object):
def __init__(self):
self.sent = []
self.closed = False
def send(self, data):
self.sent.append(data)
def close(self):
self.closed = True
@patch_policies()
class TestSender(BaseTest):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def setUp(self):
Add checksum to object extended attributes Currently, our integrity checking for objects is pretty weak when it comes to object metadata. If the extended attributes on a .data or .meta file get corrupted in such a way that we can still unpickle it, we don't have anything that detects that. This could be especially bad with encrypted etags; if the encrypted etag (X-Object-Sysmeta-Crypto-Etag or whatever it is) gets some bits flipped, then we'll cheerfully decrypt the cipherjunk into plainjunk, then send it to the client. Net effect is that the client sees a GET response with an ETag that doesn't match the MD5 of the object *and* Swift has no way of detecting and quarantining this object. Note that, with an unencrypted object, if the ETag metadatum gets mangled, then the object will be quarantined by the object server or auditor, whichever notices first. As part of this commit, I also ripped out some mocking of getxattr/setxattr in tests. It appears to be there to allow unit tests to run on systems where /tmp doesn't support xattrs. However, since the mock is keyed off of inode number and inode numbers get re-used, there's lots of leakage between different test runs. On a real FS, unlinking a file and then creating a new one of the same name will also reset the xattrs; this isn't the case with the mock. The mock was pretty old; Ubuntu 12.04 and up all support xattrs in /tmp, and recent Red Hat / CentOS releases do too. The xattr mock was added in 2011; maybe it was to support Ubuntu Lucid Lynx? Bonus: now you can pause a test with the debugger, inspect its files in /tmp, and actually see the xattrs along with the data. Since this patch now uses a real filesystem for testing filesystem operations, tests are skipped if the underlying filesystem does not support setting xattrs (eg tmpfs or more than 4k of xattrs on ext4). References to "/tmp" have been replaced with calls to tempfile.gettempdir(). This will allow setting the TMPDIR envvar in test setup and getting an XFS filesystem instead of ext4 or tmpfs. THIS PATCH SIGNIFICANTLY CHANGES TESTING ENVIRONMENTS With this patch, every test environment will require TMPDIR to be using a filesystem that supports at least 4k of extended attributes. Neither ext4 nor tempfs support this. XFS is recommended. So why all the SkipTests? Why not simply raise an error? We still need the tests to run on the base image for OpenStack's CI system. Since we were previously mocking out xattr, there wasn't a problem, but we also weren't actually testing anything. This patch adds functionality to validate xattr data, so we need to drop the mock. `test.unit.skip_if_no_xattrs()` is also imported into `test.functional` so that functional tests can import it from the functional test namespace. The related OpenStack CI infrastructure changes are made in https://review.openstack.org/#/c/394600/. Co-Authored-By: John Dickinson <me@not.mn> Change-Id: I98a37c0d451f4960b7a12f648e4405c6c6716808
2016-06-30 16:52:58 -07:00
skip_if_no_xattrs()
super(TestSender, self).setUp()
self.daemon_logger = debug_logger('test-ssync-sender')
2016-10-30 22:24:18 -07:00
self.daemon = ObjectReplicator(self.daemon_conf,
self.daemon_logger)
2016-10-30 22:24:18 -07:00
job = {'policy': POLICIES.legacy} # sufficient for Sender.__init__
self.sender = ssync_sender.Sender(self.daemon, None, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_call_catches_MessageTimeout(self):
def connect(self):
exc = exceptions.MessageTimeout(1, 'test connect')
# Cancels Eventlet's raising of this since we're about to do it.
exc.cancel()
raise exc
with mock.patch.object(ssync_sender.Sender, 'connect', connect):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES.legacy)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
self.assertEqual(1, len(error_lines))
self.assertEqual('1.2.3.4:5678/sda1/9 1 second: test connect',
error_lines[0])
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_call_catches_ReplicationException(self):
def connect(self):
raise exceptions.ReplicationException('test connect')
with mock.patch.object(ssync_sender.Sender, 'connect', connect):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES.legacy)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
self.assertEqual(1, len(error_lines))
self.assertEqual('1.2.3.4:5678/sda1/9 test connect',
error_lines[0])
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_call_catches_other_exceptions(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES.legacy)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
self.sender.connect = 'cause exception'
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
for line in error_lines:
self.assertTrue(line.startswith(
'1.2.3.4:5678/sda1/9 EXCEPTION in ssync.Sender: '))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_call_catches_exception_handling_exception(self):
2016-10-30 22:24:18 -07:00
self.sender.node = None # Will cause inside exception handler to fail
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
self.sender.connect = 'cause exception'
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
for line in error_lines:
self.assertTrue(line.startswith(
Prevent ssync writing bad fragment data to diskfile Previously, if a reconstructor sync type job failed to provide sufficient bytes from a reconstructed fragment body iterator to match the content-length that the ssync sender had already sent to the ssync receiver, the sender would still proceed to send the next subrequest. The ssync receiver might then write the start of the next subrequest to the partially complete diskfile for the previous subrequest (including writing subrequest headers to that diskfile) until it has received content-length bytes. Since a reconstructor ssync job does not send an ETag header (it cannot because it does not know the ETag of a reconstructed fragment until it has been sent) then the receiving object server does not detect the "bad" data written to the fragment diskfile, and worse, will label it with an ETag that matches the md5 sum of the bad data. The bad fragment file will therefore appear good to the auditor. There is no easy way for the ssync sender to communicate a lack of source data to the receiver other than by disconnecting the session. So this patch adds a check in the ssync sender that the sent byte count is equal to the sent Content-Length header value for each subrequest, and disconnect if a mismatch is detected. The disconnect prevents the receiver finalizing the bad diskfile, but also prevents subsequent fragments in the ssync job being sync'd until the next cycle. Closes-Bug: #1631144 Co-Authored-By: Kota Tsuyuzaki <tsuyuzaki.kota@lab.ntt.co.jp> Change-Id: I54068906efdb9cd58fcdc6eae7c2163ea92afb9d
2016-10-12 20:00:13 +01:00
'EXCEPTION in ssync.Sender'))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_call_calls_others(self):
connection = FakeConnection()
response = FakeResponse()
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
self.sender.connect = mock.MagicMock(return_value=(connection,
response))
self.sender.missing_check = mock.MagicMock(return_value=({}, {}))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.updates = mock.MagicMock()
self.sender.disconnect = mock.MagicMock()
success, candidates = self.sender()
self.assertTrue(success)
self.assertEqual(candidates, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.connect.assert_called_once_with()
self.sender.missing_check.assert_called_once_with(connection, response)
self.sender.updates.assert_called_once_with(connection, response, {})
self.sender.disconnect.assert_called_once_with(connection)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_connect(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1', backend_index=0)
job = dict(partition='9', policy=POLICIES[1])
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 1),
mock.call('X-Backend-Ssync-Frag-Index', 0),
mock.call('X-Backend-Ssync-Node-Index', 0),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_connect_handoff(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES[1], frag_index=9)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 1),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_connect_handoff_no_frag(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES[0])
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 0),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_connect_handoff_none_frag(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES[1], frag_index=None)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 1),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_connect_handoff_none_frag_to_primary(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1', backend_index=42)
job = dict(partition='9', policy=POLICIES[1], frag_index=None)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 1),
mock.call('X-Backend-Ssync-Frag-Index', 42),
mock.call('X-Backend-Ssync-Node-Index', 42),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_connect_handoff_replicated(self):
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
# no frag_index in rsync job
job = dict(partition='9', policy=POLICIES[1])
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
self.sender.suffixes = ['abc']
with mock.patch(
'swift.obj.ssync_sender.SsyncBufferedHTTPConnection'
) as mock_conn_class:
mock_conn = mock_conn_class.return_value
mock_resp = mock.MagicMock()
mock_resp.status = 200
mock_conn.getresponse.return_value = mock_resp
self.sender.connect()
mock_conn_class.assert_called_once_with('1.2.3.4:5678')
expectations = {
'putrequest': [
mock.call('SSYNC', '/sda1/9'),
],
'putheader': [
mock.call('Transfer-Encoding', 'chunked'),
mock.call('X-Backend-Storage-Policy-Index', 1),
],
'endheaders': [mock.call()],
}
for method_name, expected_calls in expectations.items():
mock_method = getattr(mock_conn, method_name)
self.assertEqual(expected_calls, mock_method.mock_calls,
'connection method "%s" got %r not %r' % (
method_name, mock_method.mock_calls,
expected_calls))
def test_call(self):
def patch_sender(sender, available_map, send_map):
connection = FakeConnection()
response = FakeResponse()
sender.connect = mock.MagicMock(return_value=(connection,
response))
sender.missing_check = mock.MagicMock()
sender.missing_check = mock.MagicMock(return_value=(available_map,
send_map))
sender.updates = mock.MagicMock()
sender.disconnect = mock.MagicMock()
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
'frag_index': 0,
}
available_map = dict([('9d41d8cd98f00b204e9800998ecf0abc',
'1380144470.00000'),
('9d41d8cd98f00b204e9800998ecf0def',
'1380144472.22222'),
('9d41d8cd98f00b204e9800998ecf1def',
'1380144474.44444')])
# no suffixes -> no work done
sender = ssync_sender.Sender(
self.daemon, node, job, [], remote_check_objs=None)
patch_sender(sender, available_map, {})
success, candidates = sender()
self.assertTrue(success)
self.assertEqual({}, candidates)
# all objs in sync
sender = ssync_sender.Sender(
self.daemon, node, job, ['ignored'], remote_check_objs=None)
patch_sender(sender, available_map, {})
success, candidates = sender()
self.assertTrue(success)
self.assertEqual(available_map, candidates)
# one obj not in sync, sync'ing faked, all objs should be in return set
wanted = '9d41d8cd98f00b204e9800998ecf0def'
sender = ssync_sender.Sender(
self.daemon, node, job, ['ignored'],
remote_check_objs=None)
patch_sender(sender, available_map, {wanted: []})
success, candidates = sender()
self.assertTrue(success)
self.assertEqual(available_map, candidates)
# one obj not in sync, remote check only so that obj is not sync'd
# and should not be in the return set
wanted = '9d41d8cd98f00b204e9800998ecf0def'
remote_check_objs = set(available_map.keys())
sender = ssync_sender.Sender(
self.daemon, node, job, ['ignored'],
remote_check_objs=remote_check_objs)
patch_sender(sender, available_map, {wanted: []})
success, candidates = sender()
self.assertTrue(success)
expected_map = dict([('9d41d8cd98f00b204e9800998ecf0abc',
'1380144470.00000'),
('9d41d8cd98f00b204e9800998ecf1def',
'1380144474.44444')])
self.assertEqual(expected_map, candidates)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_call_and_missing_check_metadata_legacy_response(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if device == 'dev' and partition == '9' and suffixes == ['abc'] \
and policy == POLICIES.legacy:
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000),
'ts_meta': Timestamp(1380155570.00005)})
else:
raise Exception(
'No match for %r %r %r' % (device, partition, suffixes))
connection = FakeConnection()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.node = {}
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender.suffixes = ['abc']
response = FakeResponse(
chunk_body=(
':MISSING_CHECK: START\r\n'
'9d41d8cd98f00b204e9800998ecf0abc\r\n'
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
':MISSING_CHECK: END\r\n'
':UPDATES: START\r\n'
':UPDATES: END\r\n'
))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
self.sender.connect = mock.MagicMock(return_value=(connection,
response))
Prevent ssync writing bad fragment data to diskfile Previously, if a reconstructor sync type job failed to provide sufficient bytes from a reconstructed fragment body iterator to match the content-length that the ssync sender had already sent to the ssync receiver, the sender would still proceed to send the next subrequest. The ssync receiver might then write the start of the next subrequest to the partially complete diskfile for the previous subrequest (including writing subrequest headers to that diskfile) until it has received content-length bytes. Since a reconstructor ssync job does not send an ETag header (it cannot because it does not know the ETag of a reconstructed fragment until it has been sent) then the receiving object server does not detect the "bad" data written to the fragment diskfile, and worse, will label it with an ETag that matches the md5 sum of the bad data. The bad fragment file will therefore appear good to the auditor. There is no easy way for the ssync sender to communicate a lack of source data to the receiver other than by disconnecting the session. So this patch adds a check in the ssync sender that the sent byte count is equal to the sent Content-Length header value for each subrequest, and disconnect if a mismatch is detected. The disconnect prevents the receiver finalizing the bad diskfile, but also prevents subsequent fragments in the ssync job being sync'd until the next cycle. Closes-Bug: #1631144 Co-Authored-By: Kota Tsuyuzaki <tsuyuzaki.kota@lab.ntt.co.jp> Change-Id: I54068906efdb9cd58fcdc6eae7c2163ea92afb9d
2016-10-12 20:00:13 +01:00
df = mock.MagicMock()
df.content_length = 0
self.sender.df_mgr.get_diskfile_from_hash = mock.MagicMock(
return_value=df)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.disconnect = mock.MagicMock()
success, candidates = self.sender()
self.assertTrue(success)
found_post = found_put = False
for chunk in connection.sent:
if b'POST' in chunk:
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
found_post = True
if b'PUT' in chunk:
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
found_put = True
self.assertFalse(found_post)
self.assertTrue(found_put)
def test_call_and_missing_check(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if device == 'dev' and partition == '9' and suffixes == ['abc'] \
and policy == POLICIES.legacy:
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
{'ts_data': Timestamp(1380144470.00000)})
else:
raise Exception(
'No match for %r %r %r' % (device, partition, suffixes))
connection = FakeConnection()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.node = {}
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender.suffixes = ['abc']
response = FakeResponse(
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
chunk_body=(
':MISSING_CHECK: START\r\n'
'9d41d8cd98f00b204e9800998ecf0abc d\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
self.sender.connect = mock.MagicMock(return_value=(connection,
response))
self.sender.updates = mock.MagicMock()
self.sender.disconnect = mock.MagicMock()
success, candidates = self.sender()
self.assertTrue(success)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(candidates,
dict([('9d41d8cd98f00b204e9800998ecf0abc',
{'ts_data': Timestamp(1380144470.00000)})]))
def test_call_and_missing_check_with_obj_list(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if device == 'dev' and partition == '9' and suffixes == ['abc'] \
and policy == POLICIES.legacy:
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
else:
raise Exception(
'No match for %r %r %r' % (device, partition, suffixes))
job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender = ssync_sender.Sender(self.daemon, None, job, ['abc'],
['9d41d8cd98f00b204e9800998ecf0abc'])
connection = FakeConnection()
response = FakeResponse(
chunk_body=(
':MISSING_CHECK: START\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
self.sender.connect = mock.MagicMock(return_value=(connection,
response))
self.sender.updates = mock.MagicMock()
self.sender.disconnect = mock.MagicMock()
success, candidates = self.sender()
self.assertTrue(success)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(candidates,
dict([('9d41d8cd98f00b204e9800998ecf0abc',
{'ts_data': Timestamp(1380144470.00000)})]))
def test_call_and_missing_check_with_obj_list_but_required(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if device == 'dev' and partition == '9' and suffixes == ['abc'] \
and policy == POLICIES.legacy:
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
else:
raise Exception(
'No match for %r %r %r' % (device, partition, suffixes))
job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
'frag_index': 0,
}
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender = ssync_sender.Sender(self.daemon, {}, job, ['abc'],
['9d41d8cd98f00b204e9800998ecf0abc'])
connection = FakeConnection()
response = FakeResponse(
chunk_body=(
':MISSING_CHECK: START\r\n'
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
'9d41d8cd98f00b204e9800998ecf0abc d\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
self.sender.connect = mock.MagicMock(return_value=(connection,
response))
self.sender.updates = mock.MagicMock()
self.sender.disconnect = mock.MagicMock()
success, candidates = self.sender()
self.assertTrue(success)
self.assertEqual(candidates, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_connect_send_timeout(self):
self.daemon.node_timeout = 0.01 # make disconnect fail fast
self.daemon.conn_timeout = 0.01
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1')
job = dict(partition='9', policy=POLICIES.legacy)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
def putrequest(*args, **kwargs):
eventlet.sleep(0.1)
with mock.patch.object(
ssync_sender.bufferedhttp.BufferedHTTPConnection,
'putrequest', putrequest):
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
for line in error_lines:
self.assertTrue(line.startswith(
'1.2.3.4:5678/sda1/9 0.01 seconds: connect send'))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_connect_receive_timeout(self):
self.daemon.node_timeout = 0.02
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1', index=0)
job = dict(partition='9', policy=POLICIES.legacy)
self.sender = ssync_sender.Sender(self.daemon, node, job, None)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
class FakeBufferedHTTPConnection(NullBufferedHTTPConnection):
def getresponse(*args, **kwargs):
eventlet.sleep(0.1)
with mock.patch.object(
ssync_sender, 'SsyncBufferedHTTPConnection',
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
FakeBufferedHTTPConnection):
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
for line in error_lines:
self.assertTrue(line.startswith(
'1.2.3.4:5678/sda1/9 0.02 seconds: connect receive'))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_connect_bad_status(self):
self.daemon.node_timeout = 0.02
node = dict(replication_ip='1.2.3.4', replication_port=5678,
device='sda1', index=0)
job = dict(partition='9', policy=POLICIES.legacy)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
class FakeBufferedHTTPConnection(NullBufferedHTTPConnection):
def getresponse(*args, **kwargs):
response = FakeResponse()
response.status = 503
response.read = lambda: 'an error message'
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
return response
Make SSYNC receiver return a reponse when initial checks fail The ssync Receiver performs some checks on request parameters in initialize_request() before starting the exchange of missing hashes and updates e.g. the destination device must be available; the policy must be valid. Currently if any of these checks fails then the receiver just closes the connection, so the Sender gets no useful response code and noise is generated in logs by httplib and wsgi Exceptions. This change moves the request parameter checks to the Receiver constructor so that the HTTPExceptions raised are actually sent as responses. (The 'connection close' exception handling still applies once the 'missing_check' and 'updates' handshakes are in progress.) Moving initialize_request() revealed the following lurking bug: * initialize_request() sets req.environ['eventlet.minimum_write_chunk_size'] = 0 * this was previously ineffective because the Response environ had already been copied from Request environ before this value was set, so the Response never used the value :/ * Now that it is effective (a good thing) it causes the empty string yielded by the receiver when there are no missing hashes in missing_checks() to be sent to the sender immediately. This makes the Sender.readline() think there has been an early disconnect and raise an Exception (a bad thing), as revealed by test/unit/obj/test_ssync_sender.py:TestSsync.test_nothing_to_sync The fix for this is to simply make the receiver skip sending the empty string if there are no missing object_hashes. Change-Id: I036a6919fead6e970505dccbb0da7bfbdf8cecc3
2015-04-27 09:17:46 +01:00
missing_check_fn = 'swift.obj.ssync_sender.Sender.missing_check'
with mock.patch(missing_check_fn) as mock_missing_check:
with mock.patch.object(
ssync_sender, 'SsyncBufferedHTTPConnection',
Make SSYNC receiver return a reponse when initial checks fail The ssync Receiver performs some checks on request parameters in initialize_request() before starting the exchange of missing hashes and updates e.g. the destination device must be available; the policy must be valid. Currently if any of these checks fails then the receiver just closes the connection, so the Sender gets no useful response code and noise is generated in logs by httplib and wsgi Exceptions. This change moves the request parameter checks to the Receiver constructor so that the HTTPExceptions raised are actually sent as responses. (The 'connection close' exception handling still applies once the 'missing_check' and 'updates' handshakes are in progress.) Moving initialize_request() revealed the following lurking bug: * initialize_request() sets req.environ['eventlet.minimum_write_chunk_size'] = 0 * this was previously ineffective because the Response environ had already been copied from Request environ before this value was set, so the Response never used the value :/ * Now that it is effective (a good thing) it causes the empty string yielded by the receiver when there are no missing hashes in missing_checks() to be sent to the sender immediately. This makes the Sender.readline() think there has been an early disconnect and raise an Exception (a bad thing), as revealed by test/unit/obj/test_ssync_sender.py:TestSsync.test_nothing_to_sync The fix for this is to simply make the receiver skip sending the empty string if there are no missing object_hashes. Change-Id: I036a6919fead6e970505dccbb0da7bfbdf8cecc3
2015-04-27 09:17:46 +01:00
FakeBufferedHTTPConnection):
self.sender = ssync_sender.Sender(
self.daemon, node, job, ['abc'])
success, candidates = self.sender()
self.assertFalse(success)
self.assertEqual(candidates, {})
error_lines = self.daemon_logger.get_lines_for_level('error')
for line in error_lines:
self.assertTrue(line.startswith(
'1.2.3.4:5678/sda1/9 Expected status 200; got 503'))
self.assertIn('an error message', line)
Make SSYNC receiver return a reponse when initial checks fail The ssync Receiver performs some checks on request parameters in initialize_request() before starting the exchange of missing hashes and updates e.g. the destination device must be available; the policy must be valid. Currently if any of these checks fails then the receiver just closes the connection, so the Sender gets no useful response code and noise is generated in logs by httplib and wsgi Exceptions. This change moves the request parameter checks to the Receiver constructor so that the HTTPExceptions raised are actually sent as responses. (The 'connection close' exception handling still applies once the 'missing_check' and 'updates' handshakes are in progress.) Moving initialize_request() revealed the following lurking bug: * initialize_request() sets req.environ['eventlet.minimum_write_chunk_size'] = 0 * this was previously ineffective because the Response environ had already been copied from Request environ before this value was set, so the Response never used the value :/ * Now that it is effective (a good thing) it causes the empty string yielded by the receiver when there are no missing hashes in missing_checks() to be sent to the sender immediately. This makes the Sender.readline() think there has been an early disconnect and raise an Exception (a bad thing), as revealed by test/unit/obj/test_ssync_sender.py:TestSsync.test_nothing_to_sync The fix for this is to simply make the receiver skip sending the empty string if there are no missing object_hashes. Change-Id: I036a6919fead6e970505dccbb0da7bfbdf8cecc3
2015-04-27 09:17:46 +01:00
# sanity check that Sender did not proceed to missing_check exchange
self.assertFalse(mock_missing_check.called)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_newline_in_buffer(self):
response = FakeResponse()
response.ssync_response_buffer = b'Has a newline already.\r\nOkay.'
self.assertEqual(response.readline(), b'Has a newline already.\r\n')
self.assertEqual(response.ssync_response_buffer, b'Okay.')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_buffer_exceeds_network_chunk_size_somehow(self):
response = FakeResponse()
response.ssync_response_buffer = b'1234567890'
self.assertEqual(response.readline(size=2), b'1234567890')
self.assertEqual(response.ssync_response_buffer, b'')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_at_start_of_chunk(self):
response = FakeResponse()
response.fp = io.BytesIO(b'2\r\nx\n\r\n')
self.assertEqual(response.readline(), b'x\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_chunk_with_extension(self):
response = FakeResponse()
response.fp = io.BytesIO(
b'2 ; chunk=extension\r\nx\n\r\n')
self.assertEqual(response.readline(), b'x\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_broken_chunk(self):
response = FakeResponse()
response.fp = io.BytesIO(b'q\r\nx\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertRaises(
exceptions.ReplicationException, response.readline)
self.assertTrue(response.close_called)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_terminated_chunk(self):
response = FakeResponse()
response.fp = io.BytesIO(b'b\r\nnot enough')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertRaises(
exceptions.ReplicationException, response.readline)
self.assertTrue(response.close_called)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_all(self):
response = FakeResponse()
response.fp = io.BytesIO(b'2\r\nx\n\r\n0\r\n\r\n')
self.assertEqual(response.readline(), b'x\n')
self.assertEqual(response.readline(), b'')
self.assertEqual(response.readline(), b'')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_readline_all_trailing_not_newline_termed(self):
response = FakeResponse()
response.fp = io.BytesIO(
b'2\r\nx\n\r\n3\r\n123\r\n0\r\n\r\n')
self.assertEqual(response.readline(), b'x\n')
self.assertEqual(response.readline(), b'123')
self.assertEqual(response.readline(), b'')
self.assertEqual(response.readline(), b'')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_timeout(self):
connection = FakeConnection()
connection.send = lambda d: eventlet.sleep(1)
response = FakeResponse()
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
self.assertRaises(exceptions.MessageTimeout, self.sender.missing_check,
connection, response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_has_empty_suffixes(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device != 'dev' or partition != '9' or
policy != POLICIES.legacy or
suffixes != ['abc', 'def']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield # Just here to make this a generator
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc', 'def']
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':MISSING_CHECK: START\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
available_map, send_map = self.sender.missing_check(connection,
response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
self.assertEqual(send_map, {})
self.assertEqual(available_map, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_has_suffixes(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc', 'def']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0def',
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
{'ts_data': Timestamp(1380144472.22222),
'ts_meta': Timestamp(1380144473.22222)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf1def',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144474.44444),
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
'ts_ctype': Timestamp(1380144474.44448),
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
'ts_meta': Timestamp(1380144475.44444)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc', 'def']
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':MISSING_CHECK: START\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
available_map, send_map = self.sender.missing_check(connection,
response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'33\r\n9d41d8cd98f00b204e9800998ecf0abc 1380144470.00000\r\n\r\n'
b'3b\r\n9d41d8cd98f00b204e9800998ecf0def 1380144472.22222 '
b'm:186a0\r\n\r\n'
b'3f\r\n9d41d8cd98f00b204e9800998ecf1def 1380144474.44444 '
b'm:186a0,t:4\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
self.assertEqual(send_map, {})
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
candidates = [('9d41d8cd98f00b204e9800998ecf0abc',
dict(ts_data=Timestamp(1380144470.00000))),
('9d41d8cd98f00b204e9800998ecf0def',
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
dict(ts_data=Timestamp(1380144472.22222),
ts_meta=Timestamp(1380144473.22222))),
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
('9d41d8cd98f00b204e9800998ecf1def',
dict(ts_data=Timestamp(1380144474.44444),
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
ts_meta=Timestamp(1380144475.44444),
ts_ctype=Timestamp(1380144474.44448)))]
self.assertEqual(available_map, dict(candidates))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_far_end_disconnect(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
response = FakeResponse(chunk_body='\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
exc = None
try:
self.sender.missing_check(connection, response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), 'Early disconnect')
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'33\r\n9d41d8cd98f00b204e9800998ecf0abc 1380144470.00000\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_far_end_disconnect2(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=':MISSING_CHECK: START\r\n')
exc = None
try:
self.sender.missing_check(connection, response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), 'Early disconnect')
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'33\r\n9d41d8cd98f00b204e9800998ecf0abc 1380144470.00000\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_missing_check_far_end_unexpected(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
response = FakeResponse(chunk_body='OH HAI\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
exc = None
try:
self.sender.missing_check(connection, response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), "Unexpected response: 'OH HAI'")
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'33\r\n9d41d8cd98f00b204e9800998ecf0abc 1380144470.00000\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_missing_check_send_map(self):
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc']):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.suffixes = ['abc']
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':MISSING_CHECK: START\r\n'
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
'0123abc dm\r\n'
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
available_map, send_map = self.sender.missing_check(connection,
response)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(
b''.join(connection.sent),
b'17\r\n:MISSING_CHECK: START\r\n\r\n'
b'33\r\n9d41d8cd98f00b204e9800998ecf0abc 1380144470.00000\r\n\r\n'
b'15\r\n:MISSING_CHECK: END\r\n\r\n')
self.assertEqual(send_map, {'0123abc': {'data': True, 'meta': True}})
self.assertEqual(available_map,
dict([('9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})]))
def test_missing_check_extra_line_parts(self):
# check that sender tolerates extra parts in missing check
# line responses to allow for protocol upgrades
def yield_hashes(device, partition, policy, suffixes=None, **kwargs):
if (device == 'dev' and partition == '9' and
policy == POLICIES.legacy and
suffixes == ['abc']):
yield (
'9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})
else:
raise Exception(
'No match for %r %r %r %r' % (device, partition,
policy, suffixes))
connection = FakeConnection()
self.sender.job = {
'device': 'dev',
'partition': '9',
'policy': POLICIES.legacy,
}
self.sender.suffixes = ['abc']
response = FakeResponse(
chunk_body=(
':MISSING_CHECK: START\r\n'
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
'0123abc d extra response parts\r\n'
':MISSING_CHECK: END\r\n'))
2016-10-30 22:24:18 -07:00
self.sender.df_mgr.yield_hashes = yield_hashes
available_map, send_map = self.sender.missing_check(connection,
response)
self.assertEqual(send_map, {'0123abc': {'data': True}})
self.assertEqual(available_map,
dict([('9d41d8cd98f00b204e9800998ecf0abc',
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
{'ts_data': Timestamp(1380144470.00000)})]))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_timeout(self):
connection = FakeConnection()
connection.send = lambda d: eventlet.sleep(1)
response = FakeResponse()
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
self.assertRaises(exceptions.MessageTimeout, self.sender.updates,
connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_updates_empty_send_map(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_unexpected_response_lines1(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
'abc\r\n'
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), "Unexpected response: 'abc'")
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_unexpected_response_lines2(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
'abc\r\n'
':UPDATES: END\r\n'))
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), "Unexpected response: 'abc'")
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_is_deleted(self):
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
df = self._make_open_diskfile(device, part, *object_parts)
object_hash = utils.hash_path(*object_parts)
delete_timestamp = utils.normalize_timestamp(time.time())
df.delete(delete_timestamp)
connection = FakeConnection()
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES.legacy,
'frag_index': 0,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.node = {}
send_map = {object_hash: {'data': True}}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.send_delete = mock.MagicMock()
self.sender.send_put = mock.MagicMock()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.send_delete.assert_called_once_with(
connection, '/a/c/o', delete_timestamp)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(self.sender.send_put.mock_calls, [])
# note that the delete line isn't actually sent since we mock
# send_delete; send_delete is tested separately.
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_update_send_delete(self):
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
df = self._make_open_diskfile(device, part, *object_parts)
object_hash = utils.hash_path(*object_parts)
delete_timestamp = utils.normalize_timestamp(time.time())
df.delete(delete_timestamp)
connection = FakeConnection()
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender.node = {}
send_map = {object_hash: {'data': True}}
response = FakeResponse(
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'30\r\n'
b'DELETE /a/c/o\r\n'
b'X-Timestamp: %s\r\n\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n'
% delete_timestamp.encode('ascii')
)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_put(self):
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# sender has data file and meta file
ts_iter = make_timestamp_iter()
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
t1 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
df = self._make_open_diskfile(
device, part, *object_parts, timestamp=t1)
t2 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
metadata = {'X-Timestamp': t2.internal, 'X-Object-Meta-Fruit': 'kiwi'}
df.write_metadata(metadata)
object_hash = utils.hash_path(*object_parts)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
df.open()
expected = df.get_metadata()
connection = FakeConnection()
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES.legacy,
'frag_index': 0,
}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.node = {}
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# receiver requested data only
send_map = {object_hash: {'data': True}}
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.send_delete = mock.MagicMock()
self.sender.send_put = mock.MagicMock()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.send_post = mock.MagicMock()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(self.sender.send_delete.mock_calls, [])
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(self.sender.send_post.mock_calls, [])
self.assertEqual(1, len(self.sender.send_put.mock_calls))
args, _kwargs = self.sender.send_put.call_args
connection, path, df = args
self.assertEqual(path, '/a/c/o')
self.assertTrue(isinstance(df, diskfile.DiskFile))
self.assertEqual(expected, df.get_metadata())
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
# note that the put line isn't actually sent since we mock send_put;
# send_put is tested separately.
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_updates_post(self):
ts_iter = make_timestamp_iter()
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
t1 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
df = self._make_open_diskfile(
device, part, *object_parts, timestamp=t1)
t2 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
metadata = {'X-Timestamp': t2.internal, 'X-Object-Meta-Fruit': 'kiwi'}
df.write_metadata(metadata)
object_hash = utils.hash_path(*object_parts)
df.open()
expected = df.get_metadata()
connection = FakeConnection()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender.node = {}
# receiver requested only meta
send_map = {object_hash: {'meta': True}}
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.send_delete = mock.MagicMock()
self.sender.send_put = mock.MagicMock()
self.sender.send_post = mock.MagicMock()
response = FakeResponse(
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(self.sender.send_delete.mock_calls, [])
self.assertEqual(self.sender.send_put.mock_calls, [])
self.assertEqual(1, len(self.sender.send_post.mock_calls))
args, _kwargs = self.sender.send_post.call_args
connection, path, df = args
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(path, '/a/c/o')
self.assertIsInstance(df, diskfile.DiskFile)
self.assertEqual(expected, df.get_metadata())
# note that the post line isn't actually sent since we mock send_post;
# send_post is tested separately.
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_updates_put_and_post(self):
ts_iter = make_timestamp_iter()
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
t1 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
df = self._make_open_diskfile(
device, part, *object_parts, timestamp=t1)
t2 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
metadata = {'X-Timestamp': t2.internal, 'X-Object-Meta-Fruit': 'kiwi'}
df.write_metadata(metadata)
object_hash = utils.hash_path(*object_parts)
df.open()
expected = df.get_metadata()
connection = FakeConnection()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES.legacy,
'frag_index': 0,
}
self.sender.node = {}
# receiver requested data and meta
send_map = {object_hash: {'meta': True, 'data': True}}
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.sender.send_delete = mock.MagicMock()
self.sender.send_put = mock.MagicMock()
self.sender.send_post = mock.MagicMock()
response = FakeResponse(
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(self.sender.send_delete.mock_calls, [])
self.assertEqual(1, len(self.sender.send_put.mock_calls))
self.assertEqual(1, len(self.sender.send_post.mock_calls))
args, _kwargs = self.sender.send_put.call_args
connection, path, df = args
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(path, '/a/c/o')
self.assertIsInstance(df, diskfile.DiskFile)
self.assertEqual(expected, df.get_metadata())
args, _kwargs = self.sender.send_post.call_args
connection, path, df = args
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(path, '/a/c/o')
self.assertIsInstance(df, diskfile.DiskFile)
self.assertEqual(expected, df.get_metadata())
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
def test_updates_storage_policy_index(self):
device = 'dev'
part = '9'
object_parts = ('a', 'c', 'o')
df = self._make_open_diskfile(device, part, *object_parts,
policy=POLICIES[0])
object_hash = utils.hash_path(*object_parts)
expected = df.get_metadata()
connection = FakeConnection()
self.sender.job = {
'device': device,
'partition': part,
'policy': POLICIES[0],
'frag_index': 0}
self.sender.node = {}
send_map = {object_hash: {'data': True}}
self.sender.send_delete = mock.MagicMock()
self.sender.send_put = mock.MagicMock()
response = FakeResponse(
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
self.sender.updates(connection, response, send_map)
args, _kwargs = self.sender.send_put.call_args
connection, path, df = args
self.assertEqual(path, '/a/c/o')
self.assertTrue(isinstance(df, diskfile.DiskFile))
self.assertEqual(expected, df.get_metadata())
2016-10-30 22:24:18 -07:00
self.assertEqual(os.path.join(self.tx_testdir, 'dev/objects/9/',
object_hash[-3:], object_hash),
df._datadir)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_timeout_start(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
orig_readline = response.readline
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def delayed_readline(*args, **kwargs):
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
eventlet.sleep(1)
return orig_readline(*args, **kwargs)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
response.readline = delayed_readline
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.http_timeout = 0.01
self.assertRaises(exceptions.MessageTimeout, self.sender.updates,
connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_disconnect_start(self):
connection = FakeConnection()
response = FakeResponse(chunk_body='\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), 'Early disconnect')
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_unexp_start(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
'anything else\r\n'
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), "Unexpected response: 'anything else'")
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_timeout_end(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
':UPDATES: END\r\n'))
orig_readline = response.readline
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def delayed_readline(*args, **kwargs):
rv = orig_readline(*args, **kwargs)
if rv == b':UPDATES: END\r\n':
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
eventlet.sleep(1)
return rv
response.readline = delayed_readline
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.http_timeout = 0.01
self.assertRaises(exceptions.MessageTimeout, self.sender.updates,
connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_disconnect_end(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
'\r\n'))
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), 'Early disconnect')
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_updates_read_response_unexp_end(self):
connection = FakeConnection()
response = FakeResponse(
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
chunk_body=(
':UPDATES: START\r\n'
'anything else\r\n'
':UPDATES: END\r\n'))
exc = None
try:
self.sender.updates(connection, response, {})
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.ReplicationException as err:
exc = err
self.assertEqual(str(exc), "Unexpected response: 'anything else'")
self.assertEqual(
b''.join(connection.sent),
b'11\r\n:UPDATES: START\r\n\r\n'
b'f\r\n:UPDATES: END\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_send_delete_timeout(self):
connection = FakeConnection()
connection.send = lambda d: eventlet.sleep(1)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
exc = None
try:
self.sender.send_delete(connection, '/a/c/o',
utils.Timestamp('1381679759.90941'))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.MessageTimeout as err:
exc = err
self.assertEqual(str(exc), '0.01 seconds: send_delete')
def test_send_delete(self):
connection = FakeConnection()
self.sender.send_delete(connection, '/a/c/o',
utils.Timestamp('1381679759.90941'))
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.assertEqual(
b''.join(connection.sent),
b'30\r\n'
b'DELETE /a/c/o\r\n'
b'X-Timestamp: 1381679759.90941\r\n'
b'\r\n\r\n')
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_send_put_initial_timeout(self):
df = self._make_open_diskfile()
df._disk_chunk_size = 2
connection = FakeConnection()
connection.send = lambda d: eventlet.sleep(1)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
exc = None
try:
self.sender.send_put(connection, '/a/c/o', df)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.MessageTimeout as err:
exc = err
self.assertEqual(str(exc), '0.01 seconds: send_put')
def test_send_put_chunk_timeout(self):
df = self._make_open_diskfile()
connection = FakeConnection()
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
one_shot = [None]
def mock_send(data):
try:
one_shot.pop()
except IndexError:
eventlet.sleep(1)
connection.send = mock_send
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
exc = None
try:
self.sender.send_put(connection, '/a/c/o', df)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
except exceptions.MessageTimeout as err:
exc = err
self.assertEqual(str(exc), '0.01 seconds: send_put chunk')
def _check_send_put(self, obj_name, meta_value):
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
ts_iter = make_timestamp_iter()
t1 = next(ts_iter)
body = b'test'
extra_metadata = {'Some-Other-Header': 'value',
u'Unicode-Meta-Name': meta_value}
df = self._make_open_diskfile(obj=obj_name, body=body,
timestamp=t1,
extra_metadata=extra_metadata)
expected = dict(df.get_metadata())
expected['body'] = body if six.PY2 else body.decode('ascii')
expected['chunk_size'] = len(body)
expected['meta'] = meta_value
wire_meta = meta_value if six.PY2 else meta_value.encode('utf8')
path = six.moves.urllib.parse.quote(expected['name'])
expected['path'] = path
expected['length'] = format(145 + len(path) + len(wire_meta), 'x')
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# .meta file metadata is not included in expected for data only PUT
t2 = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
metadata = {'X-Timestamp': t2.internal, 'X-Object-Meta-Fruit': 'kiwi'}
df.write_metadata(metadata)
df.open()
connection = FakeConnection()
self.sender.send_put(connection, path, df)
expected = (
'%(length)s\r\n'
'PUT %(path)s\r\n'
'Content-Length: %(Content-Length)s\r\n'
'ETag: %(ETag)s\r\n'
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
'Some-Other-Header: value\r\n'
'Unicode-Meta-Name: %(meta)s\r\n'
'X-Timestamp: %(X-Timestamp)s\r\n'
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
'\r\n'
'\r\n'
'%(chunk_size)s\r\n'
'%(body)s\r\n' % expected)
if not six.PY2:
expected = expected.encode('utf8')
self.assertEqual(b''.join(connection.sent), expected)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_send_put(self):
self._check_send_put('o', 'meta')
def test_send_put_unicode(self):
if six.PY2:
self._check_send_put(
'o_with_caract\xc3\xa8res_like_in_french', 'm\xc3\xa8ta')
else:
self._check_send_put(
'o_with_caract\u00e8res_like_in_french', 'm\u00e8ta')
def _check_send_post(self, obj_name, meta_value):
ts_iter = make_timestamp_iter()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# create .data file
extra_metadata = {'X-Object-Meta-Foo': 'old_value',
'X-Object-Sysmeta-Test': 'test_sysmeta',
'Content-Type': 'test_content_type'}
ts_0 = next(ts_iter)
df = self._make_open_diskfile(obj=obj_name,
extra_metadata=extra_metadata,
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
timestamp=ts_0)
# create .meta file
ts_1 = next(ts_iter)
newer_metadata = {u'X-Object-Meta-Foo': meta_value,
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
'X-Timestamp': ts_1.internal}
df.write_metadata(newer_metadata)
path = six.moves.urllib.parse.quote(df.read_metadata()['name'])
wire_meta = meta_value if six.PY2 else meta_value.encode('utf8')
length = format(61 + len(path) + len(wire_meta), 'x')
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
connection = FakeConnection()
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
with df.open():
self.sender.send_post(connection, path, df)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
self.assertEqual(
b''.join(connection.sent),
b'%s\r\n'
b'POST %s\r\n'
b'X-Object-Meta-Foo: %s\r\n'
b'X-Timestamp: %s\r\n'
b'\r\n'
b'\r\n' % (length.encode('ascii'), path.encode('ascii'),
wire_meta,
ts_1.internal.encode('ascii')))
def test_send_post(self):
self._check_send_post('o', 'meta')
def test_send_post_unicode(self):
if six.PY2:
self._check_send_post(
'o_with_caract\xc3\xa8res_like_in_french', 'm\xc3\xa8ta')
else:
self._check_send_post(
'o_with_caract\u00e8res_like_in_french', 'm\u00e8ta')
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_disconnect_timeout(self):
connection = FakeConnection()
connection.send = lambda d: eventlet.sleep(1)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
self.sender.daemon.node_timeout = 0.01
self.sender.disconnect(connection)
self.assertEqual(b''.join(connection.sent), b'')
self.assertTrue(connection.closed)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
def test_disconnect(self):
connection = FakeConnection()
self.sender.disconnect(connection)
self.assertEqual(b''.join(connection.sent), b'0\r\n\r\n')
self.assertTrue(connection.closed)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
class TestModuleMethods(unittest.TestCase):
def test_encode_missing(self):
object_hash = '9d41d8cd98f00b204e9800998ecf0abc'
ts_iter = make_timestamp_iter()
t_data = next(ts_iter)
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
t_type = next(ts_iter)
t_meta = next(ts_iter)
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
d_meta_data = t_meta.raw - t_data.raw
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
d_type_data = t_type.raw - t_data.raw
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# equal data and meta timestamps -> legacy single timestamp string
expected = '%s %s' % (object_hash, t_data.internal)
self.assertEqual(
expected.encode('ascii'),
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
ssync_sender.encode_missing(object_hash, t_data, ts_meta=t_data))
# newer meta timestamp -> hex data delta encoded as extra message part
expected = '%s %s m:%x' % (object_hash, t_data.internal, d_meta_data)
self.assertEqual(
expected.encode('ascii'),
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
ssync_sender.encode_missing(object_hash, t_data, ts_meta=t_meta))
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
# newer meta timestamp -> hex data delta encoded as extra message part
# content type timestamp equals data timestamp -> no delta
expected = '%s %s m:%x' % (object_hash, t_data.internal, d_meta_data)
self.assertEqual(
expected.encode('ascii'),
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
ssync_sender.encode_missing(object_hash, t_data, t_meta, t_data))
# content type timestamp newer data timestamp -> delta encoded
expected = ('%s %s m:%x,t:%x'
% (object_hash, t_data.internal, d_meta_data, d_type_data))
self.assertEqual(
expected.encode('ascii'),
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
ssync_sender.encode_missing(object_hash, t_data, t_meta, t_type))
# content type timestamp equal to meta timestamp -> delta encoded
expected = ('%s %s m:%x,t:%x'
% (object_hash, t_data.internal, d_meta_data, d_type_data))
self.assertEqual(
expected.encode('ascii'),
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
ssync_sender.encode_missing(object_hash, t_data, t_meta, t_type))
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
# test encode and decode functions invert
expected = {'object_hash': object_hash, 'ts_meta': t_meta,
Update container on fast-POST This patch makes a number of changes to enable content-type metadata to be updated when using the fast-POST mode of operation, as proposed in the associated spec [1]. * the object server and diskfile are modified to allow content-type to be updated by a POST and the updated value to be stored in .meta files. * the object server accepts PUTs and DELETEs with older timestamps than existing .meta files. This is to be consistent with replication that will leave a later .meta file in place when replicating a .data file. * the diskfile interface is modified to provide accessor methods for the content-type and its timestamp. * the naming of .meta files is modified to encode two timestamps when the .meta file contains a content-type value that was set prior to the latest metadata update; this enables consistency to be achieved when rsync is used for replication. * ssync is modified to sync meta files when content-type differs between local and remote copies of objects. * the object server issues container updates when handling POST requests, notifying the container server of the current immutable metadata (etag, size, hash, swift_bytes), content-type with their respective timestamps, and the mutable metadata timestamp. * the container server maintains the most recently reported values for immutable metadata, content-type and mutable metadata, each with their respective timestamps, in a single db row. * new probe tests verify that replication achieves eventual consistency of containers and objects after discrete updates to content-type and mutable metadata, and that container-sync sync's objects after fast-post updates. [1] spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Change-Id: Ia597cd460bb5fd40aa92e886e3e18a7542603d01
2015-08-10 10:30:10 -05:00
'ts_data': t_data, 'ts_ctype': t_type}
msg = ssync_sender.encode_missing(**expected)
actual = ssync_receiver.decode_missing(msg)
self.assertEqual(expected, actual)
expected = {'object_hash': object_hash, 'ts_meta': t_meta,
'ts_data': t_meta, 'ts_ctype': t_meta}
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
msg = ssync_sender.encode_missing(**expected)
actual = ssync_receiver.decode_missing(msg)
self.assertEqual(expected, actual)
def test_decode_wanted(self):
parts = ['d']
expected = {'data': True}
self.assertEqual(ssync_sender.decode_wanted(parts), expected)
parts = ['m']
expected = {'meta': True}
self.assertEqual(ssync_sender.decode_wanted(parts), expected)
parts = ['dm']
expected = {'data': True, 'meta': True}
self.assertEqual(ssync_sender.decode_wanted(parts), expected)
# you don't really expect these next few...
Add POST capability to ssync for .meta files ssync currently does the wrong thing when replicating object dirs containing both a .data and a .meta file. The ssync sender uses a single PUT to send both object content and metadata to the receiver, using the metadata (.meta file) timestamp. This results in the object content timestamp being advanced to the metadata timestamp, potentially overwriting newer object data on the receiver and causing an inconsistency with the container server record for the object. For example, replicating an object dir with {t0.data(etag=x), t2.meta} to a receiver with t1.data(etag=y) will result in the creation of t2.data(etag=x) on the receiver. However, the container server will continue to list the object as t1(etag=y). This patch modifies ssync to replicate the content of .data and .meta separately using a PUT request for the data (no change) and a POST request for the metadata. In effect, ssync replication replicates the client operations that generated the .data and .meta files so that the result of replication is the same as if the original client requests had persisted on all object servers. Apart from maintaining correct timestamps across sync'd nodes, this has the added benefit of not needing to PUT objects when only the metadata has changed and a POST will suffice. Taking the same example, ssync sender will no longer PUT t0.data but will POST t2.meta resulting in the receiver having t1.data and t2.meta. The changes are backwards compatible: an upgraded sender will only sync data files to a legacy receiver and will not sync meta files (fixing the erroneous behavior described above); a legacy sender will operate as before when sync'ing to an upgraded receiver. Changes: - diskfile API provides methods to get the data file timestamp as distinct from the diskfile timestamp. - diskfile yield_hashes return tuple now passes a dict mapping data and meta (if any) timestamps to their respective values in the timestamp field. - ssync_sender will encode data and meta timestamps in the (hash_path, timestamp) tuple sent to the receiver during missing_checks. - ssync_receiver compares sender's data and meta timestamps to any local diskfile and may specify that only data or meta parts are sent during updates phase by appending a qualifier to the hash returned in its 'wanted' list. - ssync_sender now sends POST subrequests when a meta file exists and its content needs to be replicated. - ssync_sender may send *only* a POST if the receiver indicates that is the only part required to be sync'd. - object server will allow PUT and DELETE with earlier timestamp than a POST - Fixed TODO related to replicated objects with fast-POST and ssync Related spec change-id: I60688efc3df692d3a39557114dca8c5490f7837e Co-Authored-By: Clay Gerrard <clay.gerrard@gmail.com> Closes-Bug: 1501528 Change-Id: I97552d194e5cc342b0a3f4b9800de8aa6b9cb85b
2015-04-22 12:56:50 +01:00
parts = ['md']
expected = {'data': True, 'meta': True}
self.assertEqual(ssync_sender.decode_wanted(parts), expected)
parts = ['xcy', 'funny', {'business': True}]
expected = {'data': True}
self.assertEqual(ssync_sender.decode_wanted(parts), expected)
Object replication ssync (an rsync alternative) For this commit, ssync is just a direct replacement for how we use rsync. Assuming we switch over to ssync completely someday and drop rsync, we will then be able to improve the algorithms even further (removing local objects as we successfully transfer each one rather than waiting for whole partitions, using an index.db with hash-trees, etc., etc.) For easier review, this commit can be thought of in distinct parts: 1) New global_conf_callback functionality for allowing services to perform setup code before workers, etc. are launched. (This is then used by ssync in the object server to create a cross-worker semaphore to restrict concurrent incoming replication.) 2) A bit of shifting of items up from object server and replicator to diskfile or DEFAULT conf sections for better sharing of the same settings. conn_timeout, node_timeout, client_timeout, network_chunk_size, disk_chunk_size. 3) Modifications to the object server and replicator to optionally use ssync in place of rsync. This is done in a generic enough way that switching to FutureSync should be easy someday. 4) The biggest part, and (at least for now) completely optional part, are the new ssync_sender and ssync_receiver files. Nice and isolated for easier testing and visibility into test coverage, etc. All the usual logging, statsd, recon, etc. instrumentation is still there when using ssync, just as it is when using rsync. Beyond the essential error and exceptional condition logging, I have not added any additional instrumentation at this time. Unless there is something someone finds super pressing to have added to the logging, I think such additions would be better as separate change reviews. FOR NOW, IT IS NOT RECOMMENDED TO USE SSYNC ON PRODUCTION CLUSTERS. Some of us will be in a limited fashion to look for any subtle issues, tuning, etc. but generally ssync is an experimental feature. In its current implementation it is probably going to be a bit slower than rsync, but if all goes according to plan it will end up much faster. There are no comparisions yet between ssync and rsync other than some raw virtual machine testing I've done to show it should compete well enough once we can put it in use in the real world. If you Tweet, Google+, or whatever, be sure to indicate it's experimental. It'd be best to keep it out of deployment guides, howtos, etc. until we all figure out if we like it, find it to be stable, etc. Change-Id: If003dcc6f4109e2d2a42f4873a0779110fff16d6
2013-08-28 16:10:43 +00:00
if __name__ == '__main__':
unittest.main()