swift/test/probe/test_reconstructor_revert.py

374 lines
14 KiB
Python
Raw Normal View History

#!/usr/bin/python -u
# Copyright (c) 2010-2012 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.
from hashlib import md5
import itertools
import unittest
import uuid
import random
import shutil
from collections import defaultdict
from test.probe.common import ECProbeTest, Body
from swift.common import direct_client
from swift.common.storage_policy import EC_POLICY
from swift.common.manager import Manager
from swift.obj import reconstructor
from swiftclient import client
class TestReconstructorRevert(ECProbeTest):
def setUp(self):
super(TestReconstructorRevert, self).setUp()
self.container_name = 'container-%s' % uuid.uuid4()
self.object_name = 'object-%s' % uuid.uuid4()
# sanity
self.assertEqual(self.policy.policy_type, EC_POLICY)
self.reconstructor = Manager(["object-reconstructor"])
def proxy_get(self):
# GET object
headers, body = client.get_object(self.url, self.token,
self.container_name,
self.object_name,
resp_chunk_size=64 * 2 ** 10)
resp_checksum = md5()
for chunk in body:
resp_checksum.update(chunk)
return resp_checksum.hexdigest()
def direct_get(self, node, part):
req_headers = {'X-Backend-Storage-Policy-Index': int(self.policy)}
headers, data = direct_client.direct_get_object(
node, part, self.account, self.container_name,
self.object_name, headers=req_headers,
resp_chunk_size=64 * 2 ** 20)
hasher = md5()
for chunk in data:
hasher.update(chunk)
return hasher.hexdigest()
def test_revert_object(self):
# create EC container
headers = {'X-Storage-Policy': self.policy.name}
client.put_container(self.url, self.token, self.container_name,
headers=headers)
# get our node lists
opart, onodes = self.object_ring.get_nodes(
self.account, self.container_name, self.object_name)
hnodes = self.object_ring.get_more_nodes(opart)
# kill 2 a parity count number of primary nodes so we can
# force data onto handoffs, we do that by renaming dev dirs
# to induce 507
p_dev1 = self.device_dir('object', onodes[0])
p_dev2 = self.device_dir('object', onodes[1])
self.kill_drive(p_dev1)
self.kill_drive(p_dev2)
# PUT object
contents = Body()
headers = {'x-object-meta-foo': 'meta-foo'}
headers_post = {'x-object-meta-bar': 'meta-bar'}
client.put_object(self.url, self.token, self.container_name,
self.object_name, contents=contents,
headers=headers)
client.post_object(self.url, self.token, self.container_name,
self.object_name, headers=headers_post)
# (Some versions of?) swiftclient will mutate the headers dict on post
headers_post.pop('X-Auth-Token', None)
# these primaries can't serve the data any more, we expect 507
# here and not 404 because we're using mount_check to kill nodes
for onode in (onodes[0], onodes[1]):
try:
self.direct_get(onode, opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 507)
else:
self.fail('Node data on %r was not fully destroyed!' %
(onode,))
# now take out another primary
p_dev3 = self.device_dir('object', onodes[2])
self.kill_drive(p_dev3)
# this node can't servce the data any more
try:
self.direct_get(onodes[2], opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 507)
else:
self.fail('Node data on %r was not fully destroyed!' %
(onode,))
# make sure we can still GET the object and its correct
# we're now pulling from handoffs and reconstructing
etag = self.proxy_get()
self.assertEqual(etag, contents.etag)
# rename the dev dirs so they don't 507 anymore
self.revive_drive(p_dev1)
self.revive_drive(p_dev2)
self.revive_drive(p_dev3)
# fire up reconstructor on handoff nodes only
for hnode in hnodes:
hnode_id = (hnode['port'] - 6000) / 10
self.reconstructor.once(number=hnode_id)
# first three primaries have data again
for onode in (onodes[0], onodes[2]):
self.direct_get(onode, opart)
# check meta
meta = client.head_object(self.url, self.token,
self.container_name,
self.object_name)
for key in headers_post:
self.assertIn(key, meta)
self.assertEqual(meta[key], headers_post[key])
# handoffs are empty
for hnode in hnodes:
try:
self.direct_get(hnode, opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 404)
else:
self.fail('Node data on %r was not fully destroyed!' %
(hnode,))
def test_delete_propagate(self):
# create EC container
headers = {'X-Storage-Policy': self.policy.name}
client.put_container(self.url, self.token, self.container_name,
headers=headers)
# get our node lists
opart, onodes = self.object_ring.get_nodes(
self.account, self.container_name, self.object_name)
hnodes = list(itertools.islice(
self.object_ring.get_more_nodes(opart), 2))
# PUT object
contents = Body()
client.put_object(self.url, self.token, self.container_name,
self.object_name, contents=contents)
# now lets shut down a couple of primaries
failed_nodes = random.sample(onodes, 2)
for node in failed_nodes:
self.kill_drive(self.device_dir('object', node))
# Write tombstones over the nodes that are still online
client.delete_object(self.url, self.token,
self.container_name,
self.object_name)
# spot check the primary nodes that are still online
delete_timestamp = None
for node in onodes:
if node in failed_nodes:
continue
try:
self.direct_get(node, opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 404)
delete_timestamp = err.http_headers['X-Backend-Timestamp']
else:
self.fail('Node data on %r was not fully destroyed!' %
(node,))
# run the reconstructor on the handoff node multiple times until
# tombstone is pushed out - each handoff node syncs to a few
# primaries each time
iterations = 0
while iterations < 52:
self.reconstructor.once(number=self.config_number(hnodes[0]))
iterations += 1
# see if the tombstone is reverted
try:
self.direct_get(hnodes[0], opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 404)
if 'X-Backend-Timestamp' not in err.http_headers:
# this means the tombstone is *gone* so it's reverted
break
else:
self.fail('Still found tombstone on %r after %s iterations' % (
hnodes[0], iterations))
# tombstone is still on the *second* handoff
try:
self.direct_get(hnodes[1], opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 404)
self.assertEqual(err.http_headers['X-Backend-Timestamp'],
delete_timestamp)
else:
self.fail('Found obj data on %r' % hnodes[1])
# repair the primaries
self.revive_drive(self.device_dir('object', failed_nodes[0]))
self.revive_drive(self.device_dir('object', failed_nodes[1]))
# run reconstructor on second handoff
self.reconstructor.once(number=self.config_number(hnodes[1]))
# verify tombstone is reverted on the first pass
try:
self.direct_get(hnodes[1], opart)
except direct_client.DirectClientException as err:
self.assertEqual(err.http_status, 404)
self.assertNotIn('X-Backend-Timestamp', err.http_headers)
else:
self.fail('Found obj data on %r' % hnodes[1])
# sanity make sure proxy get can't find it
try:
self.proxy_get()
except Exception as err:
self.assertEqual(err.http_status, 404)
else:
self.fail('Node data on %r was not fully destroyed!' %
(onodes[0]))
def test_reconstruct_from_reverted_fragment_archive(self):
headers = {'X-Storage-Policy': self.policy.name}
client.put_container(self.url, self.token, self.container_name,
headers=headers)
# get our node lists
opart, onodes = self.object_ring.get_nodes(
self.account, self.container_name, self.object_name)
# find a primary server that only has one of it's devices in the
# primary node list
group_nodes_by_config = defaultdict(list)
for n in onodes:
group_nodes_by_config[self.config_number(n)].append(n)
for config_number, node_list in group_nodes_by_config.items():
if len(node_list) == 1:
break
else:
self.fail('ring balancing did not use all available nodes')
primary_node = node_list[0]
# ... and 507 it's device
primary_device = self.device_dir('object', primary_node)
self.kill_drive(primary_device)
# PUT object
contents = Body()
etag = client.put_object(self.url, self.token, self.container_name,
self.object_name, contents=contents)
self.assertEqual(contents.etag, etag)
# fix the primary device and sanity GET
self.revive_drive(primary_device)
self.assertEqual(etag, self.proxy_get())
# find a handoff holding the fragment
for hnode in self.object_ring.get_more_nodes(opart):
try:
reverted_fragment_etag = self.direct_get(hnode, opart)
except direct_client.DirectClientException as err:
if err.http_status != 404:
raise
else:
break
else:
self.fail('Unable to find handoff fragment!')
# we'll force the handoff device to revert instead of potentially
# racing with rebuild by deleting any other fragments that may be on
# the same server
handoff_fragment_etag = None
for node in onodes:
Allow 1+ object-servers-per-disk deployment Enabled by a new > 0 integer config value, "servers_per_port" in the [DEFAULT] config section for object-server and/or replication server configs. The setting's integer value determines how many different object-server workers handle requests for any single unique local port in the ring. In this mode, the parent swift-object-server process continues to run as the original user (i.e. root if low-port binding is required), binds to all ports as defined in the ring, and forks off the specified number of workers per listen socket. The child, per-port servers drop privileges and behave pretty much how object-server workers always have, except that because the ring has unique ports per disk, the object-servers will only be handling requests for a single disk. The parent process detects dead servers and restarts them (with the correct listen socket), starts missing servers when an updated ring file is found with a device on the server with a new port, and kills extraneous servers when their port is found to no longer be in the ring. The ring files are stat'ed at most every "ring_check_interval" seconds, as configured in the object-server config (same default of 15s). Immediately stopping all swift-object-worker processes still works by sending the parent a SIGTERM. Likewise, a SIGHUP to the parent process still causes the parent process to close all listen sockets and exit, allowing existing children to finish serving their existing requests. The drop_privileges helper function now has an optional param to suppress the setsid() call, which otherwise screws up the child workers' process management. The class method RingData.load() can be told to only load the ring metadata (i.e. everything except replica2part2dev_id) with the optional kwarg, header_only=True. This is used to keep the parent and all forked off workers from unnecessarily having full copies of all storage policy rings in memory. A new helper class, swift.common.storage_policy.BindPortsCache, provides a method to return a set of all device ports in all rings for the server on which it is instantiated (identified by its set of IP addresses). The BindPortsCache instance will track mtimes of ring files, so they are not opened more frequently than necessary. This patch includes enhancements to the probe tests and object-replicator/object-reconstructor config plumbing to allow the probe tests to work correctly both in the "normal" config (same IP but unique ports for each SAIO "server") and a server-per-port setup where each SAIO "server" must have a unique IP address and unique port per disk within each "server". The main probe tests only work with 4 servers and 4 disks, but you can see the difference in the rings for the EC probe tests where there are 2 disks per server for a total of 8 disks. Specifically, swift.common.ring.utils.is_local_device() will ignore the ports when the "my_port" argument is None. Then, object-replicator and object-reconstructor both set self.bind_port to None if server_per_port is enabled. Bonus improvement for IPv6 addresses in is_local_device(). This PR for vagrant-swift-all-in-one will aid in testing this patch: https://github.com/swiftstack/vagrant-swift-all-in-one/pull/16/ Also allow SAIO to answer is_local_device() better; common SAIO setups have multiple "servers" all on the same host with different ports for the different "servers" (which happen to match the IPs specified in the rings for the devices on each of those "servers"). However, you can configure the SAIO to have different localhost IP addresses (e.g. 127.0.0.1, 127.0.0.2, etc.) in the ring and in the servers' config files' bind_ip setting. This new whataremyips() implementation combined with a little plumbing allows is_local_device() to accurately answer, even on an SAIO. In the default case (an unspecified bind_ip defaults to '0.0.0.0') as well as an explict "bind to everything" like '0.0.0.0' or '::', whataremyips() behaves as it always has, returning all IP addresses for the server. Also updated probe tests to handle each "server" in the SAIO having a unique IP address. For some (noisy) benchmarks that show servers_per_port=X is at least as good as the same number of "normal" workers: https://gist.github.com/dbishop/c214f89ca708a6b1624a#file-summary-md Benchmarks showing the benefits of I/O isolation with a small number of slow disks: https://gist.github.com/dbishop/fd0ab067babdecfb07ca#file-results-md If you were wondering what the overhead of threads_per_disk looks like: https://gist.github.com/dbishop/1d14755fedc86a161718#file-tabular_results-md DocImpact Change-Id: I2239a4000b41a7e7cc53465ce794af49d44796c6
2015-05-14 22:14:15 -07:00
if self.is_local_to(node, hnode):
# we'll keep track of the etag of this fragment we're removing
# in case we need it later (queue forshadowing music)...
try:
handoff_fragment_etag = self.direct_get(node, opart)
except direct_client.DirectClientException as err:
if err.http_status != 404:
raise
# this just means our handoff device was on the same
# machine as the primary!
continue
# use the primary nodes device - not the hnode device
part_dir = self.storage_dir('object', node, part=opart)
shutil.rmtree(part_dir, True)
# revert from handoff device with reconstructor
self.reconstructor.once(number=self.config_number(hnode))
# verify fragment reverted to primary server
self.assertEqual(reverted_fragment_etag,
self.direct_get(primary_node, opart))
# now we'll remove some data on one of the primary node's partners
partner = random.choice(reconstructor._get_partners(
primary_node['index'], onodes))
try:
rebuilt_fragment_etag = self.direct_get(partner, opart)
except direct_client.DirectClientException as err:
if err.http_status != 404:
raise
# partner already had it's fragment removed
if (handoff_fragment_etag is not None and
Allow 1+ object-servers-per-disk deployment Enabled by a new > 0 integer config value, "servers_per_port" in the [DEFAULT] config section for object-server and/or replication server configs. The setting's integer value determines how many different object-server workers handle requests for any single unique local port in the ring. In this mode, the parent swift-object-server process continues to run as the original user (i.e. root if low-port binding is required), binds to all ports as defined in the ring, and forks off the specified number of workers per listen socket. The child, per-port servers drop privileges and behave pretty much how object-server workers always have, except that because the ring has unique ports per disk, the object-servers will only be handling requests for a single disk. The parent process detects dead servers and restarts them (with the correct listen socket), starts missing servers when an updated ring file is found with a device on the server with a new port, and kills extraneous servers when their port is found to no longer be in the ring. The ring files are stat'ed at most every "ring_check_interval" seconds, as configured in the object-server config (same default of 15s). Immediately stopping all swift-object-worker processes still works by sending the parent a SIGTERM. Likewise, a SIGHUP to the parent process still causes the parent process to close all listen sockets and exit, allowing existing children to finish serving their existing requests. The drop_privileges helper function now has an optional param to suppress the setsid() call, which otherwise screws up the child workers' process management. The class method RingData.load() can be told to only load the ring metadata (i.e. everything except replica2part2dev_id) with the optional kwarg, header_only=True. This is used to keep the parent and all forked off workers from unnecessarily having full copies of all storage policy rings in memory. A new helper class, swift.common.storage_policy.BindPortsCache, provides a method to return a set of all device ports in all rings for the server on which it is instantiated (identified by its set of IP addresses). The BindPortsCache instance will track mtimes of ring files, so they are not opened more frequently than necessary. This patch includes enhancements to the probe tests and object-replicator/object-reconstructor config plumbing to allow the probe tests to work correctly both in the "normal" config (same IP but unique ports for each SAIO "server") and a server-per-port setup where each SAIO "server" must have a unique IP address and unique port per disk within each "server". The main probe tests only work with 4 servers and 4 disks, but you can see the difference in the rings for the EC probe tests where there are 2 disks per server for a total of 8 disks. Specifically, swift.common.ring.utils.is_local_device() will ignore the ports when the "my_port" argument is None. Then, object-replicator and object-reconstructor both set self.bind_port to None if server_per_port is enabled. Bonus improvement for IPv6 addresses in is_local_device(). This PR for vagrant-swift-all-in-one will aid in testing this patch: https://github.com/swiftstack/vagrant-swift-all-in-one/pull/16/ Also allow SAIO to answer is_local_device() better; common SAIO setups have multiple "servers" all on the same host with different ports for the different "servers" (which happen to match the IPs specified in the rings for the devices on each of those "servers"). However, you can configure the SAIO to have different localhost IP addresses (e.g. 127.0.0.1, 127.0.0.2, etc.) in the ring and in the servers' config files' bind_ip setting. This new whataremyips() implementation combined with a little plumbing allows is_local_device() to accurately answer, even on an SAIO. In the default case (an unspecified bind_ip defaults to '0.0.0.0') as well as an explict "bind to everything" like '0.0.0.0' or '::', whataremyips() behaves as it always has, returning all IP addresses for the server. Also updated probe tests to handle each "server" in the SAIO having a unique IP address. For some (noisy) benchmarks that show servers_per_port=X is at least as good as the same number of "normal" workers: https://gist.github.com/dbishop/c214f89ca708a6b1624a#file-summary-md Benchmarks showing the benefits of I/O isolation with a small number of slow disks: https://gist.github.com/dbishop/fd0ab067babdecfb07ca#file-results-md If you were wondering what the overhead of threads_per_disk looks like: https://gist.github.com/dbishop/1d14755fedc86a161718#file-tabular_results-md DocImpact Change-Id: I2239a4000b41a7e7cc53465ce794af49d44796c6
2015-05-14 22:14:15 -07:00
self.is_local_to(hnode, partner)):
# oh, well that makes sense then...
rebuilt_fragment_etag = handoff_fragment_etag
else:
# I wonder what happened?
self.fail('Partner inexplicably missing fragment!')
part_dir = self.storage_dir('object', partner, part=opart)
shutil.rmtree(part_dir, True)
# sanity, it's gone
try:
self.direct_get(partner, opart)
except direct_client.DirectClientException as err:
if err.http_status != 404:
raise
else:
self.fail('successful GET of removed partner fragment archive!?')
# and force the primary node to do a rebuild
self.reconstructor.once(number=self.config_number(primary_node))
# and validate the partners rebuilt_fragment_etag
try:
self.assertEqual(rebuilt_fragment_etag,
self.direct_get(partner, opart))
except direct_client.DirectClientException as err:
if err.http_status != 404:
raise
else:
self.fail('Did not find rebuilt fragment on partner node')
if __name__ == "__main__":
unittest.main()