python-swiftclient/swiftclient/service.py
Tim Burke 0ee7c8272e Make proper functions instead of assigning lambdas
Change-Id: I89255f6923c649c7b9d3d36e96c09f8bc4f51a3c
2019-02-21 09:26:32 -08:00

2956 lines
112 KiB
Python

# Copyright (c) 2010-2013 OpenStack, LLC.
#
# 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 __future__ import unicode_literals
import logging
import os
from concurrent.futures import as_completed, CancelledError, TimeoutError
from copy import deepcopy
from errno import EEXIST, ENOENT
from hashlib import md5
from os import environ, makedirs, stat, utime
from os.path import (
basename, dirname, getmtime, getsize, isdir, join, sep as os_path_sep
)
from posixpath import join as urljoin
from random import shuffle
from time import time
from threading import Thread
from six import Iterator, StringIO, string_types, text_type
from six.moves.queue import Queue
from six.moves.queue import Empty as QueueEmpty
from six.moves.urllib.parse import quote
import json
from swiftclient import Connection
from swiftclient.command_helpers import (
stat_account, stat_container, stat_object
)
from swiftclient.utils import (
config_true_value, ReadableToIterable, LengthWrapper, EMPTY_ETAG,
parse_api_response, report_traceback, n_groups, split_request_headers,
n_at_a_time
)
from swiftclient.exceptions import ClientException
from swiftclient.multithreading import MultiThreadingManager
DISK_BUFFER = 2 ** 16
logger = logging.getLogger("swiftclient.service")
class ResultsIterator(Iterator):
def __init__(self, futures):
self.futures = interruptable_as_completed(futures)
def __iter__(self):
return self
def __next__(self):
next_completed_future = next(self.futures)
return next_completed_future.result()
class SwiftError(Exception):
def __init__(self, value, container=None, obj=None,
segment=None, exc=None):
self.value = value
self.container = container
self.obj = obj
self.segment = segment
self.exception = exc
def __str__(self):
value = repr(self.value)
if self.container is not None:
value += " container:%s" % self.container
if self.obj is not None:
value += " object:%s" % self.obj
if self.segment is not None:
value += " segment:%s" % self.segment
return value
def process_options(options):
# tolerate sloppy auth_version
if options.get('auth_version') == '3.0':
options['auth_version'] = '3'
elif options.get('auth_version') == '2':
options['auth_version'] = '2.0'
if options.get('auth_version') not in ('2.0', '3') and not all(
options.get(key) for key in ('auth', 'user', 'key')):
# Use keystone auth if any of the new-style args are present
if any(options.get(k) for k in (
'os_user_domain_id',
'os_user_domain_name',
'os_project_domain_id',
'os_project_domain_name')):
# Use v3 if there's any reference to domains
options['auth_version'] = '3'
else:
options['auth_version'] = '2.0'
# Use new-style args if old ones not present
if not options['auth'] and options['os_auth_url']:
options['auth'] = options['os_auth_url']
if not options['user'] and options['os_username']:
options['user'] = options['os_username']
if not options['key'] and options['os_password']:
options['key'] = options['os_password']
# Specific OpenStack options
options['os_options'] = {
'user_id': options['os_user_id'],
'user_domain_id': options['os_user_domain_id'],
'user_domain_name': options['os_user_domain_name'],
'tenant_id': options['os_tenant_id'],
'tenant_name': options['os_tenant_name'],
'project_id': options['os_project_id'],
'project_name': options['os_project_name'],
'project_domain_id': options['os_project_domain_id'],
'project_domain_name': options['os_project_domain_name'],
'service_type': options['os_service_type'],
'endpoint_type': options['os_endpoint_type'],
'auth_token': options['os_auth_token'],
'object_storage_url': options['os_storage_url'],
'region_name': options['os_region_name'],
}
def _build_default_global_options():
return {
"snet": False,
"verbose": 1,
"debug": False,
"info": False,
"auth": environ.get('ST_AUTH'),
"auth_version": environ.get('ST_AUTH_VERSION', '1.0'),
"user": environ.get('ST_USER'),
"key": environ.get('ST_KEY'),
"retries": 5,
"force_auth_retry": False,
"os_username": environ.get('OS_USERNAME'),
"os_user_id": environ.get('OS_USER_ID'),
"os_user_domain_name": environ.get('OS_USER_DOMAIN_NAME'),
"os_user_domain_id": environ.get('OS_USER_DOMAIN_ID'),
"os_password": environ.get('OS_PASSWORD'),
"os_tenant_id": environ.get('OS_TENANT_ID'),
"os_tenant_name": environ.get('OS_TENANT_NAME'),
"os_project_name": environ.get('OS_PROJECT_NAME'),
"os_project_id": environ.get('OS_PROJECT_ID'),
"os_project_domain_name": environ.get('OS_PROJECT_DOMAIN_NAME'),
"os_project_domain_id": environ.get('OS_PROJECT_DOMAIN_ID'),
"os_auth_url": environ.get('OS_AUTH_URL'),
"os_auth_token": environ.get('OS_AUTH_TOKEN'),
"os_storage_url": environ.get('OS_STORAGE_URL'),
"os_region_name": environ.get('OS_REGION_NAME'),
"os_service_type": environ.get('OS_SERVICE_TYPE'),
"os_endpoint_type": environ.get('OS_ENDPOINT_TYPE'),
"os_cacert": environ.get('OS_CACERT'),
"os_cert": environ.get('OS_CERT'),
"os_key": environ.get('OS_KEY'),
"insecure": config_true_value(environ.get('SWIFTCLIENT_INSECURE')),
"ssl_compression": False,
'segment_threads': 10,
'object_dd_threads': 10,
'object_uu_threads': 10,
'container_threads': 10
}
_default_global_options = _build_default_global_options()
_default_local_options = {
'sync_to': None,
'sync_key': None,
'use_slo': False,
'segment_size': None,
'segment_container': None,
'leave_segments': False,
'changed': None,
'skip_identical': False,
'yes_all': False,
'read_acl': None,
'write_acl': None,
'out_file': None,
'out_directory': None,
'remove_prefix': False,
'no_download': False,
'long': False,
'totals': False,
'marker': '',
'header': [],
'meta': [],
'prefix': None,
'delimiter': None,
'fail_fast': False,
'human': False,
'dir_marker': False,
'checksum': True,
'shuffle': False,
'destination': None,
'fresh_metadata': False,
'ignore_mtime': False,
}
POLICY = 'X-Storage-Policy'
KNOWN_DIR_MARKERS = (
'application/directory', # Preferred
'text/directory', # Historically relevant
)
def get_from_queue(q, timeout=864000):
while True:
try:
item = q.get(timeout=timeout)
return item
except QueueEmpty:
# Do nothing here, we only have a timeout to allow interruption
pass
def get_future_result(f, timeout=86400):
while True:
try:
res = f.result(timeout=timeout)
return res
except TimeoutError:
# Do nothing here, we only have a timeout to allow interruption
pass
def interruptable_as_completed(fs, timeout=86400):
while True:
try:
for f in as_completed(fs, timeout=timeout):
fs.remove(f)
yield f
return
except TimeoutError:
# Do nothing here, we only have a timeout to allow interruption
pass
def get_conn(options):
"""
Return a connection building it from the options.
"""
return Connection(options['auth'],
options['user'],
options['key'],
options['retries'],
auth_version=options['auth_version'],
os_options=options['os_options'],
snet=options['snet'],
cacert=options['os_cacert'],
insecure=options['insecure'],
cert=options['os_cert'],
cert_key=options['os_key'],
ssl_compression=options['ssl_compression'],
force_auth_retry=options['force_auth_retry'])
def mkdirs(path):
try:
makedirs(path)
except OSError as err:
if err.errno != EEXIST:
raise
def split_headers(options, prefix=''):
"""
Splits 'Key: Value' strings and returns them as a dictionary.
:param options: Must be one of:
* an iterable of 'Key: Value' strings
* an iterable of ('Key', 'Value') pairs
* a dict of {'Key': 'Value'} pairs
:param prefix: String to prepend to all of the keys in the dictionary.
reporting.
"""
headers = {}
try:
headers = split_request_headers(options, prefix)
except ValueError as e:
raise SwiftError(e)
return headers
class SwiftUploadObject(object):
"""
Class for specifying an object upload, allowing the object source, name and
options to be specified separately for each individual object.
"""
def __init__(self, source, object_name=None, options=None):
if isinstance(source, string_types):
self.object_name = object_name or source
elif source is None or hasattr(source, 'read'):
if not object_name or not isinstance(object_name, string_types):
raise SwiftError('Object names must be specified as '
'strings for uploads from None or file '
'like objects.')
self.object_name = object_name
else:
raise SwiftError('Unexpected source type for '
'SwiftUploadObject: {0}'.format(type(source)))
if not self.object_name:
raise SwiftError('Object names must not be empty strings')
self.object_name = self.object_name.lstrip('/')
self.options = options
self.source = source
class SwiftPostObject(object):
"""
Class for specifying an object post, allowing the headers/metadata to be
specified separately for each individual object.
"""
def __init__(self, object_name, options=None):
if not (isinstance(object_name, string_types) and object_name):
raise SwiftError(
"Object names must be specified as non-empty strings"
)
self.object_name = object_name
self.options = options
class SwiftCopyObject(object):
"""
Class for specifying an object copy,
allowing the destination/headers/metadata/fresh_metadata to be specified
separately for each individual object.
destination and fresh_metadata should be set in options
"""
def __init__(self, object_name, options=None):
if not (isinstance(object_name, string_types) and object_name):
raise SwiftError(
"Object names must be specified as non-empty strings"
)
self.object_name = object_name
self.options = options
if self.options is None:
self.destination = None
self.fresh_metadata = False
else:
self.destination = self.options.get('destination')
self.fresh_metadata = self.options.get('fresh_metadata', False)
if self.destination is not None:
destination_components = self.destination.split('/')
if destination_components[0] or len(destination_components) < 2:
raise SwiftError("destination must be in format /cont[/obj]")
if not destination_components[-1]:
raise SwiftError("destination must not end in a slash")
if len(destination_components) == 2:
# only container set in destination
self.destination = "{0}/{1}".format(
self.destination, object_name
)
class _SwiftReader(object):
"""
Class for downloading objects from swift and raising appropriate
errors on failures caused by either invalid md5sum or size of the
data read.
"""
def __init__(self, path, body, headers, checksum=True):
self._path = path
self._body = body
self._actual_read = 0
self._content_length = None
self._actual_md5 = None
self._expected_md5 = headers.get('etag', '')
if len(self._expected_md5) > 1 and self._expected_md5[0] == '"' \
and self._expected_md5[-1] == '"':
self._expected_md5 = self._expected_md5[1:-1]
# Some headers indicate the MD5 of the response
# definitely *won't* match the ETag
bad_md5_headers = set([
'content-range',
'x-object-manifest',
'x-static-large-object',
])
if bad_md5_headers.intersection(headers):
# This isn't a useful checksum
self._expected_md5 = ''
if self._expected_md5 and checksum:
self._actual_md5 = md5()
if 'content-length' in headers:
try:
self._content_length = int(headers.get('content-length'))
except ValueError:
raise SwiftError('content-length header must be an integer')
def __iter__(self):
for chunk in self._body:
if self._actual_md5:
self._actual_md5.update(chunk)
self._actual_read += len(chunk)
yield chunk
self._check_contents()
def _check_contents(self):
if self._actual_md5 and self._expected_md5:
etag = self._actual_md5.hexdigest()
if etag != self._expected_md5:
raise SwiftError('Error downloading {0}: md5sum != etag, '
'{1} != {2}'.format(
self._path, etag, self._expected_md5))
if (self._content_length is not None and
self._actual_read != self._content_length):
raise SwiftError('Error downloading {0}: read_length != '
'content_length, {1:d} != {2:d}'.format(
self._path, self._actual_read,
self._content_length))
def bytes_read(self):
return self._actual_read
class SwiftService(object):
"""
Service for performing swift operations
"""
def __init__(self, options=None):
if options is not None:
self._options = dict(
_default_global_options,
**dict(_default_local_options, **options)
)
else:
self._options = dict(
_default_global_options,
**_default_local_options
)
process_options(self._options)
def create_connection():
return get_conn(self._options)
self.thread_manager = MultiThreadingManager(
create_connection,
segment_threads=self._options['segment_threads'],
object_dd_threads=self._options['object_dd_threads'],
object_uu_threads=self._options['object_uu_threads'],
container_threads=self._options['container_threads']
)
self.capabilities_cache = {} # Each instance should have its own cache
def __enter__(self):
self.thread_manager.__enter__()
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.thread_manager.__exit__(exc_type, exc_val, exc_tb)
# Stat related methods
#
def stat(self, container=None, objects=None, options=None):
"""
Get account stats, container stats or information about a list of
objects in a container.
:param container: The container to query.
:param objects: A list of object paths about which to return
information (a list of strings).
:param options: A dictionary containing options to override the global
options specified during the service object creation.
These options are applied to all stat operations
performed by this call::
{
'human': False,
'header': []
}
:returns: Either a single dictionary containing stats about an account
or container, or an iterator for returning the results of the
stat operations on a list of objects.
:raises SwiftError:
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
if not container:
if objects:
raise SwiftError('Objects specified without container')
else:
res = {
'action': 'stat_account',
'success': True,
'container': container,
'object': None,
}
try:
stats_future = self.thread_manager.container_pool.submit(
stat_account, options
)
items, headers = get_future_result(stats_future)
res.update({
'items': items,
'headers': headers
})
return res
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
raise SwiftError('Account not found', exc=err)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
else:
if not objects:
res = {
'action': 'stat_container',
'container': container,
'object': None,
'success': True,
}
try:
stats_future = self.thread_manager.container_pool.submit(
stat_container, options, container
)
items, headers = get_future_result(stats_future)
res.update({
'items': items,
'headers': headers
})
return res
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
raise SwiftError('Container %r not found' % container,
container=container, exc=err)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
else:
stat_futures = []
for stat_o in objects:
stat_future = self.thread_manager.object_dd_pool.submit(
self._stat_object, container, stat_o, options
)
stat_futures.append(stat_future)
return ResultsIterator(stat_futures)
@staticmethod
def _stat_object(conn, container, obj, options):
res = {
'action': 'stat_object',
'object': obj,
'container': container,
'success': True,
}
try:
items, headers = stat_object(conn, options, container, obj)
res.update({
'items': items,
'headers': headers
})
return res
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
# Post related methods
#
def post(self, container=None, objects=None, options=None):
"""
Post operations on an account, container or list of objects
:param container: The container to make the post operation against.
:param objects: A list of object names (strings) or SwiftPostObject
instances containing an object name, and an
options dict (can be None) to override the options for
that individual post operation::
[
'object_name',
SwiftPostObject('object_name', options={...}),
...
]
The options dict is described below.
:param options: A dictionary containing options to override the global
options specified during the service object creation.
These options are applied to all post operations
performed by this call, unless overridden on a per
object basis. Possible options are given below::
{
'meta': [],
'header': [],
'read_acl': None, # For containers only
'write_acl': None, # For containers only
'sync_to': None, # For containers only
'sync_key': None # For containers only
}
:returns: Either a single result dictionary in the case of a post to a
container/account, or an iterator for returning the results
of posts to a list of objects.
:raises SwiftError:
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
res = {
'success': True,
'container': container,
'object': None,
'headers': {},
}
if not container:
res["action"] = "post_account"
if objects:
raise SwiftError('Objects specified without container')
else:
response_dict = {}
headers = split_headers(
options['meta'], 'X-Account-Meta-')
headers.update(
split_headers(options['header'], ''))
res['headers'] = headers
try:
post = self.thread_manager.container_pool.submit(
self._post_account_job, headers, response_dict
)
get_future_result(post)
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': response_dict
})
return res
raise SwiftError('Account not found', exc=err)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'response_dict': response_dict,
'traceback': traceback,
'error_timestamp': err_time
})
return res
if not objects:
res["action"] = "post_container"
response_dict = {}
headers = split_headers(
options['meta'], 'X-Container-Meta-')
headers.update(
split_headers(options['header'], ''))
if options['read_acl'] is not None:
headers['X-Container-Read'] = options['read_acl']
if options['write_acl'] is not None:
headers['X-Container-Write'] = options['write_acl']
if options['sync_to'] is not None:
headers['X-Container-Sync-To'] = options['sync_to']
if options['sync_key'] is not None:
headers['X-Container-Sync-Key'] = options['sync_key']
res['headers'] = headers
try:
post = self.thread_manager.container_pool.submit(
self._post_container_job, container,
headers, response_dict
)
get_future_result(post)
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'action': 'post_container',
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': response_dict
})
return res
raise SwiftError(
"Container '%s' not found" % container,
container=container, exc=err
)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'action': 'post_container',
'success': False,
'error': err,
'response_dict': response_dict,
'traceback': traceback,
'error_timestamp': err_time
})
return res
else:
post_futures = []
post_objects = self._make_post_objects(objects)
for post_object in post_objects:
obj = post_object.object_name
obj_options = post_object.options
response_dict = {}
headers = split_headers(
options['meta'], 'X-Object-Meta-')
# add header options to the headers object for the request.
headers.update(
split_headers(options['header'], ''))
if obj_options is not None:
if 'meta' in obj_options:
headers.update(
split_headers(
obj_options['meta'], 'X-Object-Meta-'
)
)
if 'header' in obj_options:
headers.update(
split_headers(obj_options['header'], '')
)
post = self.thread_manager.object_uu_pool.submit(
self._post_object_job, container, obj,
headers, response_dict
)
post_futures.append(post)
return ResultsIterator(post_futures)
@staticmethod
def _make_post_objects(objects):
post_objects = []
for o in objects:
if isinstance(o, string_types):
obj = SwiftPostObject(o)
post_objects.append(obj)
elif isinstance(o, SwiftPostObject):
post_objects.append(o)
else:
raise SwiftError(
"The post operation takes only strings or "
"SwiftPostObjects as input",
obj=o)
return post_objects
@staticmethod
def _post_account_job(conn, headers, result):
return conn.post_account(headers=headers, response_dict=result)
@staticmethod
def _post_container_job(conn, container, headers, result):
try:
res = conn.post_container(
container, headers=headers, response_dict=result)
except ClientException as err:
if err.http_status != 404:
raise
_response_dict = {}
res = conn.put_container(
container, headers=headers, response_dict=_response_dict)
result['post_put'] = _response_dict
return res
@staticmethod
def _post_object_job(conn, container, obj, headers, result):
res = {
'success': True,
'action': 'post_object',
'container': container,
'object': obj,
'headers': headers,
'response_dict': result
}
try:
conn.post_object(
container, obj, headers=headers, response_dict=result)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
# List related methods
#
def list(self, container=None, options=None):
"""
List operations on an account, container.
:param container: The container to make the list operation against.
:param options: A dictionary containing options to override the global
options specified during the service object creation::
{
'long': False,
'prefix': None,
'delimiter': None,
'header': []
}
:returns: A generator for returning the results of the list operation
on an account or container. Each result yielded from the
generator is either a 'list_account_part' or
'list_container_part', containing part of the listing.
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
rq = Queue(maxsize=10) # Just stop list running away consuming memory
if container is None:
listing_future = self.thread_manager.container_pool.submit(
self._list_account_job, options, rq
)
else:
listing_future = self.thread_manager.container_pool.submit(
self._list_container_job, container, options, rq
)
res = get_from_queue(rq)
while res is not None:
yield res
res = get_from_queue(rq)
# Make sure the future has completed
get_future_result(listing_future)
@staticmethod
def _list_account_job(conn, options, result_queue):
marker = ''
error = None
req_headers = split_headers(options.get('header', []))
try:
while True:
_, items = conn.get_account(
marker=marker, prefix=options['prefix'],
headers=req_headers
)
if not items:
result_queue.put(None)
return
if options['long']:
for i in items:
name = i['name']
i['meta'] = conn.head_container(name)
res = {
'action': 'list_account_part',
'container': None,
'prefix': options['prefix'],
'success': True,
'listing': items,
'marker': marker,
}
result_queue.put(res)
marker = items[-1].get('name', items[-1].get('subdir'))
except ClientException as err:
traceback, err_time = report_traceback()
logger.exception(err)
if err.http_status != 404:
error = (err, traceback, err_time)
else:
error = (
SwiftError('Account not found', exc=err),
traceback, err_time
)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
error = (err, traceback, err_time)
res = {
'action': 'list_account_part',
'container': None,
'prefix': options['prefix'],
'success': False,
'marker': marker,
'error': error[0],
'traceback': error[1],
'error_timestamp': error[2]
}
result_queue.put(res)
result_queue.put(None)
@staticmethod
def _list_container_job(conn, container, options, result_queue):
marker = options.get('marker', '')
error = None
req_headers = split_headers(options.get('header', []))
try:
while True:
_, items = conn.get_container(
container, marker=marker, prefix=options['prefix'],
delimiter=options['delimiter'], headers=req_headers
)
if not items:
result_queue.put(None)
return
res = {
'action': 'list_container_part',
'container': container,
'prefix': options['prefix'],
'success': True,
'marker': marker,
'listing': items,
}
result_queue.put(res)
marker = items[-1].get('name', items[-1].get('subdir'))
except ClientException as err:
traceback, err_time = report_traceback()
logger.exception(err)
if err.http_status != 404:
error = (err, traceback, err_time)
else:
error = (
SwiftError(
'Container %r not found' % container,
container=container, exc=err
),
traceback,
err_time
)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
error = (err, traceback, err_time)
res = {
'action': 'list_container_part',
'container': container,
'prefix': options['prefix'],
'success': False,
'marker': marker,
'error': error[0],
'traceback': error[1],
'error_timestamp': error[2]
}
result_queue.put(res)
result_queue.put(None)
# Download related methods
#
def download(self, container=None, objects=None, options=None):
"""
Download operations on an account, optional container and optional list
of objects.
:param container: The container to download from.
:param objects: A list of object names to download (a list of strings).
:param options: A dictionary containing options to override the global
options specified during the service object creation::
{
'yes_all': False,
'marker': '',
'prefix': None,
'no_download': False,
'header': [],
'skip_identical': False,
'out_directory': None,
'checksum': True,
'out_file': None,
'remove_prefix': False,
'shuffle' : False
}
:returns: A generator for returning the results of the download
operations. Each result yielded from the generator is a
'download_object' dictionary containing the results of an
individual file download.
:raises ClientException:
:raises SwiftError:
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
if not container:
# Download everything if options['yes_all'] is set
if options['yes_all']:
try:
options_copy = deepcopy(options)
options_copy["long"] = False
for part in self.list(options=options_copy):
if part["success"]:
containers = [i['name'] for i in part["listing"]]
if options['shuffle']:
shuffle(containers)
for con in containers:
for res in self._download_container(
con, options_copy):
yield res
else:
raise part["error"]
# If we see a 404 here, the listing of the account failed
except ClientException as err:
if err.http_status != 404:
raise
raise SwiftError('Account not found', exc=err)
elif objects is None:
if '/' in container:
raise SwiftError('\'/\' in container name',
container=container)
for res in self._download_container(container, options):
yield res
else:
if '/' in container:
raise SwiftError('\'/\' in container name',
container=container)
if options['out_file'] and len(objects) > 1:
options['out_file'] = None
o_downs = [
self.thread_manager.object_dd_pool.submit(
self._download_object_job, container, obj, options
) for obj in objects
]
for o_down in interruptable_as_completed(o_downs):
yield o_down.result()
def _download_object_job(self, conn, container, obj, options):
out_file = options['out_file']
results_dict = {}
req_headers = split_headers(options['header'], '')
pseudodir = False
path = join(container, obj) if options['yes_all'] else obj
path = path.lstrip(os_path_sep)
options['skip_identical'] = (options['skip_identical'] and
out_file != '-')
if options['prefix'] and options['remove_prefix']:
path = path[len(options['prefix']):].lstrip('/')
if options['out_directory']:
path = os.path.join(options['out_directory'], path)
if options['skip_identical']:
filename = out_file if out_file else path
try:
fp = open(filename, 'rb', DISK_BUFFER)
except IOError:
pass
else:
with fp:
md5sum = md5()
while True:
data = fp.read(DISK_BUFFER)
if not data:
break
md5sum.update(data)
req_headers['If-None-Match'] = md5sum.hexdigest()
try:
start_time = time()
get_args = {'resp_chunk_size': DISK_BUFFER,
'headers': req_headers,
'response_dict': results_dict}
if options['skip_identical']:
# Assume the file is a large object; if we're wrong, the query
# string is ignored and the If-None-Match header will trigger
# the behavior we want
get_args['query_string'] = 'multipart-manifest=get'
try:
headers, body = conn.get_object(container, obj, **get_args)
except ClientException as e:
if not options['skip_identical']:
raise
if e.http_status != 304: # Only handling Not Modified
raise
headers = results_dict['headers']
if 'x-object-manifest' in headers:
# DLO: most likely it has more than one page worth of
# segments and we have an empty file locally
body = []
elif config_true_value(headers.get('x-static-large-object')):
# SLO: apparently we have a copy of the manifest locally?
# provide no chunking data to force a fresh download
body = [b'[]']
else:
# Normal object: let it bubble up
raise
if options['skip_identical']:
if config_true_value(headers.get('x-static-large-object')) or \
'x-object-manifest' in headers:
# The request was chunked, so stitch it back together
chunk_data = self._get_chunk_data(conn, container, obj,
headers, b''.join(body))
else:
chunk_data = None
if chunk_data is not None:
if self._is_identical(chunk_data, filename):
raise ClientException('Large object is identical',
http_status=304)
# Large objects are different; start the real download
del get_args['query_string']
get_args['response_dict'].clear()
headers, body = conn.get_object(container, obj, **get_args)
headers_receipt = time()
obj_body = _SwiftReader(path, body, headers,
options.get('checksum', True))
no_file = options['no_download']
if out_file == "-" and not no_file:
res = {
'action': 'download_object',
'container': container,
'object': obj,
'path': path,
'pseudodir': pseudodir,
'contents': obj_body
}
return res
fp = None
try:
content_type = headers.get('content-type', '').split(';', 1)[0]
if content_type in KNOWN_DIR_MARKERS:
make_dir = not no_file and out_file != "-"
if make_dir and not isdir(path):
mkdirs(path)
else:
make_dir = not (no_file or out_file)
if make_dir:
dirpath = dirname(path)
if dirpath and not isdir(dirpath):
mkdirs(dirpath)
if not no_file:
if out_file:
fp = open(out_file, 'wb', DISK_BUFFER)
else:
if basename(path):
fp = open(path, 'wb', DISK_BUFFER)
else:
pseudodir = True
for chunk in obj_body:
if fp is not None:
fp.write(chunk)
finish_time = time()
finally:
bytes_read = obj_body.bytes_read()
if fp is not None:
fp.close()
if ('x-object-meta-mtime' in headers and not no_file and
not options['ignore_mtime']):
try:
mtime = float(headers['x-object-meta-mtime'])
except ValueError:
pass # no real harm; couldn't trust it anyway
else:
if options['out_file']:
utime(options['out_file'], (mtime, mtime))
else:
utime(path, (mtime, mtime))
res = {
'action': 'download_object',
'success': True,
'container': container,
'object': obj,
'path': path,
'pseudodir': pseudodir,
'start_time': start_time,
'finish_time': finish_time,
'headers_receipt': headers_receipt,
'auth_end_time': conn.auth_end_time,
'read_length': bytes_read,
'attempts': conn.attempts,
'response_dict': results_dict
}
return res
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res = {
'action': 'download_object',
'container': container,
'object': obj,
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': results_dict,
'path': path,
'pseudodir': pseudodir,
'attempts': conn.attempts
}
return res
def _submit_page_downloads(self, container, page_generator, options):
try:
list_page = next(page_generator)
except StopIteration:
return None
if list_page["success"]:
objects = [o["name"] for o in list_page["listing"]]
if options["shuffle"]:
shuffle(objects)
o_downs = [
self.thread_manager.object_dd_pool.submit(
self._download_object_job, container, obj, options
) for obj in objects
]
return o_downs
else:
raise list_page["error"]
def _download_container(self, container, options):
_page_generator = self.list(container=container, options=options)
try:
next_page_downs = self._submit_page_downloads(
container, _page_generator, options
)
except ClientException as err:
if err.http_status != 404:
raise
raise SwiftError(
'Container %r not found' % container,
container=container, exc=err
)
error = None
while next_page_downs:
page_downs = next_page_downs
next_page_downs = None
# Start downloading the next page of list results when
# we have completed 80% of the previous page
next_page_triggered = False
next_page_trigger_point = 0.8 * len(page_downs)
page_results_yielded = 0
for o_down in interruptable_as_completed(page_downs):
yield o_down.result()
# Do we need to start the next set of downloads yet?
if not next_page_triggered:
page_results_yielded += 1
if page_results_yielded >= next_page_trigger_point:
try:
next_page_downs = self._submit_page_downloads(
container, _page_generator, options
)
except ClientException as err:
# Allow the current page to finish downloading
logger.exception(err)
error = err
except Exception:
# Something unexpected went wrong - cancel
# remaining downloads
for _d in page_downs:
_d.cancel()
raise
finally:
# Stop counting and testing
next_page_triggered = True
if error:
raise error
# Upload related methods
#
def upload(self, container, objects, options=None):
"""
Upload a list of objects to a given container.
:param container: The container (or pseudo-folder path) to put the
uploads into.
:param objects: A list of file/directory names (strings) or
SwiftUploadObject instances containing a source for the
created object, an object name, and an options dict
(can be None) to override the options for that
individual upload operation::
[
'/path/to/file',
SwiftUploadObject('/path', object_name='obj1'),
...
]
The options dict is as described below.
The SwiftUploadObject source may be one of:
* A file-like object (with a read method)
* A string containing the path to a local
file or directory
* None, to indicate that we want an empty object
:param options: A dictionary containing options to override the global
options specified during the service object creation.
These options are applied to all upload operations
performed by this call, unless overridden on a per
object basis. Possible options are given below::
{
'meta': [],
'header': [],
'segment_size': None,
'use_slo': False,
'segment_container': None,
'leave_segments': False,
'changed': None,
'skip_identical': False,
'fail_fast': False,
'dir_marker': False # Only for None sources
}
:returns: A generator for returning the results of the uploads.
:raises SwiftError:
:raises ClientException:
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
try:
segment_size = int(0 if options['segment_size'] is None else
options['segment_size'])
except ValueError:
raise SwiftError('Segment size should be an integer value')
# Incase we have a psudeo-folder path for <container> arg, derive
# the container name from the top path and prepend the rest to
# the object name. (same as passing --object-name).
container, _sep, pseudo_folder = container.partition('/')
# Try to create the container, just in case it doesn't exist. If this
# fails, it might just be because the user doesn't have container PUT
# permissions, so we'll ignore any error. If there's really a problem,
# it'll surface on the first object PUT.
policy_header = {}
_header = split_headers(options["header"])
if POLICY in _header:
policy_header[POLICY] = \
_header[POLICY]
create_containers = [
self.thread_manager.container_pool.submit(
self._create_container_job, container, headers=policy_header)
]
# wait for first container job to complete before possibly attempting
# segment container job because segment container job may attempt
# to HEAD the first container
for r in interruptable_as_completed(create_containers):
res = r.result()
yield res
if segment_size:
seg_container = container + '_segments'
if options['segment_container']:
seg_container = options['segment_container']
if seg_container != container:
if not policy_header:
# Since no storage policy was specified on the command
# line, rather than just letting swift pick the default
# storage policy, we'll try to create the segments
# container with the same policy as the upload container
create_containers = [
self.thread_manager.container_pool.submit(
self._create_container_job, seg_container,
policy_source=container
)
]
else:
create_containers = [
self.thread_manager.container_pool.submit(
self._create_container_job, seg_container,
headers=policy_header
)
]
for r in interruptable_as_completed(create_containers):
res = r.result()
yield res
# We maintain a results queue here and a separate thread to monitor
# the futures because we want to get results back from potential
# segment uploads too
rq = Queue()
file_jobs = {}
upload_objects = self._make_upload_objects(objects, pseudo_folder)
for upload_object in upload_objects:
s = upload_object.source
o = upload_object.object_name
o_opts = upload_object.options
details = {'action': 'upload', 'container': container}
if o_opts is not None:
object_options = deepcopy(options)
object_options.update(o_opts)
else:
object_options = options
if hasattr(s, 'read'):
# We've got a file like object to upload to o
file_future = self.thread_manager.object_uu_pool.submit(
self._upload_object_job, container, s, o, object_options,
results_queue=rq
)
details['file'] = s
details['object'] = o
file_jobs[file_future] = details
elif s is not None:
# We've got a path to upload to o
details['path'] = s
details['object'] = o
if isdir(s):
dir_future = self.thread_manager.object_uu_pool.submit(
self._create_dir_marker_job, container, o,
object_options, path=s
)
file_jobs[dir_future] = details
else:
try:
stat(s)
file_future = \
self.thread_manager.object_uu_pool.submit(
self._upload_object_job, container, s, o,
object_options, results_queue=rq
)
file_jobs[file_future] = details
except OSError as err:
# Avoid tying up threads with jobs that will fail
traceback, err_time = report_traceback()
logger.exception(err)
res = {
'action': 'upload_object',
'container': container,
'object': o,
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'path': s
}
rq.put(res)
else:
# Create an empty object (as a dir marker if is_dir)
details['file'] = None
details['object'] = o
if object_options['dir_marker']:
dir_future = self.thread_manager.object_uu_pool.submit(
self._create_dir_marker_job, container, o,
object_options
)
file_jobs[dir_future] = details
else:
file_future = self.thread_manager.object_uu_pool.submit(
self._upload_object_job, container, StringIO(),
o, object_options
)
file_jobs[file_future] = details
# Start a thread to watch for upload results
Thread(
target=self._watch_futures, args=(file_jobs, rq)
).start()
# yield results as they become available, including those from
# segment uploads.
res = get_from_queue(rq)
cancelled = False
while res is not None:
yield res
if not res['success']:
if not cancelled and options['fail_fast']:
cancelled = True
for f in file_jobs:
f.cancel()
res = get_from_queue(rq)
@staticmethod
def _make_upload_objects(objects, pseudo_folder=''):
upload_objects = []
for o in objects:
if isinstance(o, string_types):
obj = SwiftUploadObject(o, urljoin(pseudo_folder,
o.lstrip('/')))
upload_objects.append(obj)
elif isinstance(o, SwiftUploadObject):
o.object_name = urljoin(pseudo_folder, o.object_name)
upload_objects.append(o)
else:
raise SwiftError(
"The upload operation takes only strings or "
"SwiftUploadObjects as input",
obj=o)
return upload_objects
@staticmethod
def _create_container_job(
conn, container, headers=None, policy_source=None):
"""
Create a container using the given connection
:param conn: The swift connection used for requests.
:param container: The container name to create.
:param headers: An optional dict of headers for the
put_container request.
:param policy_source: An optional name of a container whose policy we
should duplicate.
:return: A dict containing the results of the operation.
"""
res = {
'action': 'create_container',
'container': container,
'headers': headers
}
create_response = {}
try:
if policy_source is not None:
_meta = conn.head_container(policy_source)
if 'x-storage-policy' in _meta:
policy_header = {
POLICY: _meta.get('x-storage-policy')
}
if headers is None:
headers = policy_header
else:
headers.update(policy_header)
conn.put_container(
container, headers, response_dict=create_response
)
res.update({
'success': True,
'response_dict': create_response
})
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': create_response
})
return res
@staticmethod
def _create_dir_marker_job(conn, container, obj, options, path=None):
res = {
'action': 'create_dir_marker',
'container': container,
'object': obj,
'path': path
}
results_dict = {}
if obj.startswith('./') or obj.startswith('.\\'):
obj = obj[2:]
if obj.startswith('/'):
obj = obj[1:]
if path is not None:
put_headers = {'x-object-meta-mtime': "%f" % getmtime(path)}
else:
put_headers = {'x-object-meta-mtime': "%f" % round(time())}
res['headers'] = put_headers
if options['changed']:
try:
headers = conn.head_object(container, obj)
ct = headers.get('content-type', '').split(';', 1)[0]
cl = int(headers.get('content-length'))
et = headers.get('etag')
mt = headers.get('x-object-meta-mtime')
if (ct in KNOWN_DIR_MARKERS and
cl == 0 and
et == EMPTY_ETAG and
mt == put_headers['x-object-meta-mtime']):
res['success'] = True
return res
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
try:
conn.put_object(container, obj, '', content_length=0,
content_type=KNOWN_DIR_MARKERS[0],
headers=put_headers,
response_dict=results_dict)
res.update({
'success': True,
'response_dict': results_dict})
return res
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': results_dict})
return res
@staticmethod
def _upload_segment_job(conn, path, container, segment_name, segment_start,
segment_size, segment_index, obj_name, options,
results_queue=None):
results_dict = {}
if options['segment_container']:
segment_container = options['segment_container']
else:
segment_container = container + '_segments'
res = {
'action': 'upload_segment',
'for_container': container,
'for_object': obj_name,
'segment_index': segment_index,
'segment_size': segment_size,
'segment_location': '/%s/%s' % (segment_container,
segment_name),
'log_line': '%s segment %s' % (obj_name, segment_index),
}
fp = None
try:
fp = open(path, 'rb', DISK_BUFFER)
fp.seek(segment_start)
contents = LengthWrapper(fp, segment_size, md5=options['checksum'])
etag = conn.put_object(
segment_container,
segment_name,
contents,
content_length=segment_size,
content_type='application/swiftclient-segment',
response_dict=results_dict)
if options['checksum'] and etag and etag != contents.get_md5sum():
raise SwiftError('Segment {0}: upload verification failed: '
'md5 mismatch, local {1} != remote {2} '
'(remote segment has not been removed)'
.format(segment_index,
contents.get_md5sum(),
etag))
res.update({
'success': True,
'response_dict': results_dict,
'segment_etag': etag,
'attempts': conn.attempts
})
if results_queue is not None:
results_queue.put(res)
return res
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time,
'response_dict': results_dict,
'attempts': conn.attempts
})
if results_queue is not None:
results_queue.put(res)
return res
finally:
if fp is not None:
fp.close()
@staticmethod
def _put_object(conn, container, name, content, headers=None, md5=None):
"""
Upload object into a given container and verify the resulting ETag, if
the md5 optional parameter is passed.
:param conn: The Swift connection to use for uploads.
:param container: The container to put the object into.
:param name: The name of the object.
:param content: Object content.
:param headers: Headers (optional) to associate with the object.
:param md5: MD5 sum of the content. If passed in, will be used to
verify the returned ETag.
:returns: A dictionary as the response from calling put_object.
The keys are:
- status
- reason
- headers
On error, the dictionary contains the following keys:
- success (with value False)
- error - the encountered exception (object)
- error_timestamp
- response_dict - results from the put_object call, as
documented above
- attempts - number of attempts made
"""
if headers is None:
headers = {}
else:
headers = dict(headers)
if md5 is not None:
headers['etag'] = md5
results = {}
try:
etag = conn.put_object(
container, name, content, content_length=len(content),
headers=headers, response_dict=results)
if md5 is not None and etag != md5:
raise SwiftError('Upload verification failed for {0}: md5 '
'mismatch {1} != {2}'.format(name, md5, etag))
results['success'] = True
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
return {
'success': False,
'error': err,
'error_timestamp': err_time,
'response_dict': results,
'attempts': conn.attempts,
'traceback': traceback
}
return results
@staticmethod
def _upload_stream_segment(conn, container, object_name,
segment_container, segment_name,
segment_size, segment_index,
headers, fd):
"""
Upload a segment from a stream, buffering it in memory first. The
resulting object is placed either as a segment in the segment
container, or if it is smaller than a single segment, as the given
object name.
:param conn: Swift Connection to use.
:param container: Container in which the object would be placed.
:param object_name: Name of the final object (used in case the stream
is smaller than the segment_size)
:param segment_container: Container to hold the object segments.
:param segment_name: The name of the segment.
:param segment_size: Minimum segment size.
:param segment_index: The segment index.
:param headers: Headers to attach to the segment/object.
:param fd: File-like handle for the content. Must implement read().
:returns: Dictionary, containing the following keys:
- complete -- whether the stream is exhausted
- segment_size - the actual size of the segment (may be
smaller than the passed in segment_size)
- segment_location - path to the segment
- segment_index - index of the segment
- segment_etag - the ETag for the segment
"""
buf = []
dgst = md5()
bytes_read = 0
while bytes_read < segment_size:
data = fd.read(segment_size - bytes_read)
if not data:
break
bytes_read += len(data)
dgst.update(data)
buf.append(data)
buf = b''.join(buf)
segment_hash = dgst.hexdigest()
if not buf and segment_index > 0:
# Happens if the segment size aligns with the object size
return {'complete': True,
'segment_size': 0,
'segment_index': None,
'segment_etag': None,
'segment_location': None,
'success': True}
if segment_index == 0 and len(buf) < segment_size:
ret = SwiftService._put_object(
conn, container, object_name, buf, headers, segment_hash)
ret['segment_location'] = '/%s/%s' % (container, object_name)
else:
ret = SwiftService._put_object(
conn, segment_container, segment_name, buf, headers,
segment_hash)
ret['segment_location'] = '/%s/%s' % (
segment_container, segment_name)
ret.update(
dict(complete=len(buf) < segment_size,
segment_size=len(buf),
segment_index=segment_index,
segment_etag=segment_hash,
for_object=object_name))
return ret
def _get_chunk_data(self, conn, container, obj, headers, manifest=None):
chunks = []
if 'x-object-manifest' in headers:
scontainer, sprefix = headers['x-object-manifest'].split('/', 1)
for part in self.list(scontainer, {'prefix': sprefix}):
if part["success"]:
chunks.extend(part["listing"])
else:
raise part["error"]
elif config_true_value(headers.get('x-static-large-object')):
if manifest is None:
headers, manifest = conn.get_object(
container, obj, query_string='multipart-manifest=get')
manifest = parse_api_response(headers, manifest)
for chunk in manifest:
if chunk.get('sub_slo'):
scont, sobj = chunk['name'].lstrip('/').split('/', 1)
chunks.extend(self._get_chunk_data(
conn, scont, sobj, {'x-static-large-object': True}))
else:
chunks.append(chunk)
else:
chunks.append({'hash': headers.get('etag').strip('"'),
'bytes': int(headers.get('content-length'))})
return chunks
def _is_identical(self, chunk_data, path):
if path is None:
return False
try:
fp = open(path, 'rb', DISK_BUFFER)
except IOError:
return False
with fp:
for chunk in chunk_data:
to_read = chunk['bytes']
md5sum = md5()
while to_read:
data = fp.read(min(DISK_BUFFER, to_read))
if not data:
return False
md5sum.update(data)
to_read -= len(data)
if md5sum.hexdigest() != chunk['hash']:
return False
# Each chunk is verified; check that we're at the end of the file
return not fp.read(1)
@staticmethod
def _upload_slo_manifest(conn, segment_results, container, obj, headers):
"""
Upload an SLO manifest, given the results of uploading each segment, to
the specified container.
:param segment_results: List of response_dict structures, as populated
by _upload_segment_job. Specifically, each
entry must container the following keys:
- segment_location
- segment_etag
- segment_size
- segment_index
:param container: The container to put the manifest into.
:param obj: The name of the manifest object to use.
:param headers: Optional set of headers to attach to the manifest.
"""
if headers is None:
headers = {}
segment_results.sort(key=lambda di: di['segment_index'])
for seg in segment_results:
seg_loc = seg['segment_location'].lstrip('/')
if isinstance(seg_loc, text_type):
seg_loc = seg_loc.encode('utf-8')
manifest_data = json.dumps([
{
'path': d['segment_location'],
'etag': d['segment_etag'],
'size_bytes': d['segment_size']
} for d in segment_results
])
response = {}
conn.put_object(
container, obj, manifest_data,
headers=headers,
query_string='multipart-manifest=put',
response_dict=response)
return response
def _upload_object_job(self, conn, container, source, obj, options,
results_queue=None):
if obj.startswith('./') or obj.startswith('.\\'):
obj = obj[2:]
if obj.startswith('/'):
obj = obj[1:]
res = {
'action': 'upload_object',
'container': container,
'object': obj
}
if hasattr(source, 'read'):
stream = source
path = None
else:
path = source
res['path'] = path
try:
if path is not None:
put_headers = {'x-object-meta-mtime': "%f" % getmtime(path)}
else:
put_headers = {'x-object-meta-mtime': "%f" % round(time())}
res['headers'] = put_headers
# We need to HEAD all objects now in case we're overwriting a
# manifest object and need to delete the old segments
# ourselves.
old_manifest = None
old_slo_manifest_paths = []
new_slo_manifest_paths = set()
segment_size = int(0 if options['segment_size'] is None
else options['segment_size'])
if (options['changed'] or options['skip_identical'] or
not options['leave_segments']):
try:
headers = conn.head_object(container, obj)
is_slo = config_true_value(
headers.get('x-static-large-object'))
if options['skip_identical'] or (
is_slo and not options['leave_segments']):
chunk_data = self._get_chunk_data(
conn, container, obj, headers)
if options['skip_identical'] and self._is_identical(
chunk_data, path):
res.update({
'success': True,
'status': 'skipped-identical'
})
return res
cl = int(headers.get('content-length'))
mt = headers.get('x-object-meta-mtime')
if (path is not None and options['changed'] and
cl == getsize(path) and
mt == put_headers['x-object-meta-mtime']):
res.update({
'success': True,
'status': 'skipped-changed'
})
return res
if not options['leave_segments']:
old_manifest = headers.get('x-object-manifest')
if is_slo:
for old_seg in chunk_data:
seg_path = old_seg['name'].lstrip('/')
if isinstance(seg_path, text_type):
seg_path = seg_path.encode('utf-8')
old_slo_manifest_paths.append(seg_path)
except ClientException as err:
if err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
# Merge the command line header options to the put_headers
put_headers.update(split_headers(
options['meta'], 'X-Object-Meta-'))
put_headers.update(split_headers(options['header'], ''))
# Don't do segment job if object is not big enough, and never do
# a segment job if we're reading from a stream - we may fail if we
# go over the single object limit, but this gives us a nice way
# to create objects from memory
if (path is not None and segment_size and
(getsize(path) > segment_size)):
res['large_object'] = True
seg_container = container + '_segments'
if options['segment_container']:
seg_container = options['segment_container']
full_size = getsize(path)
segment_futures = []
segment_pool = self.thread_manager.segment_pool
segment = 0
segment_start = 0
while segment_start < full_size:
if segment_start + segment_size > full_size:
segment_size = full_size - segment_start
if options['use_slo']:
segment_name = '%s/slo/%s/%s/%s/%08d' % (
obj, put_headers['x-object-meta-mtime'],
full_size, options['segment_size'], segment
)
else:
segment_name = '%s/%s/%s/%s/%08d' % (
obj, put_headers['x-object-meta-mtime'],
full_size, options['segment_size'], segment
)
seg = segment_pool.submit(
self._upload_segment_job, path, container,
segment_name, segment_start, segment_size, segment,
obj, options, results_queue=results_queue
)
segment_futures.append(seg)
segment += 1
segment_start += segment_size
segment_results = []
errors = False
exceptions = []
for f in interruptable_as_completed(segment_futures):
try:
r = f.result()
if not r['success']:
errors = True
segment_results.append(r)
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
errors = True
exceptions.append((err, traceback, err_time))
if errors:
err = ClientException(
'Aborting manifest creation '
'because not all segments could be uploaded. %s/%s'
% (container, obj))
res.update({
'success': False,
'error': err,
'exceptions': exceptions,
'segment_results': segment_results
})
return res
res['segment_results'] = segment_results
if options['use_slo']:
response = self._upload_slo_manifest(
conn, segment_results, container, obj, put_headers)
res['manifest_response_dict'] = response
new_slo_manifest_paths = {
seg['segment_location'] for seg in segment_results}
else:
new_object_manifest = '%s/%s/%s/%s/%s/' % (
quote(seg_container.encode('utf8')),
quote(obj.encode('utf8')),
put_headers['x-object-meta-mtime'], full_size,
options['segment_size'])
if old_manifest and old_manifest.rstrip('/') == \
new_object_manifest.rstrip('/'):
old_manifest = None
put_headers['x-object-manifest'] = new_object_manifest
mr = {}
conn.put_object(
container, obj, '', content_length=0,
headers=put_headers,
response_dict=mr
)
res['manifest_response_dict'] = mr
elif options['use_slo'] and segment_size and not path:
segment = 0
results = []
while True:
segment_name = '%s/slo/%s/%s/%08d' % (
obj, put_headers['x-object-meta-mtime'],
segment_size, segment
)
seg_container = container + '_segments'
if options['segment_container']:
seg_container = options['segment_container']
ret = self._upload_stream_segment(
conn, container, obj,
seg_container,
segment_name,
segment_size,
segment,
put_headers,
stream
)
if not ret['success']:
return ret
if (ret['complete'] and segment == 0) or\
ret['segment_size'] > 0:
results.append(ret)
if results_queue is not None:
# Don't insert the 0-sized segments or objects
# themselves
if ret['segment_location'] != '/%s/%s' % (
container, obj) and ret['segment_size'] > 0:
results_queue.put(ret)
if ret['complete']:
break
segment += 1
if results[0]['segment_location'] != '/%s/%s' % (
container, obj):
response = self._upload_slo_manifest(
conn, results, container, obj, put_headers)
res['manifest_response_dict'] = response
new_slo_manifest_paths = {
r['segment_location'] for r in results}
res['large_object'] = True
else:
res['response_dict'] = ret
res['large_object'] = False
else:
res['large_object'] = False
obr = {}
fp = None
try:
if path is not None:
content_length = getsize(path)
fp = open(path, 'rb', DISK_BUFFER)
contents = LengthWrapper(fp,
content_length,
md5=options['checksum'])
else:
content_length = None
contents = ReadableToIterable(stream,
md5=options['checksum'])
etag = conn.put_object(
container, obj, contents,
content_length=content_length, headers=put_headers,
response_dict=obr
)
res['response_dict'] = obr
if (options['checksum'] and
etag and etag != contents.get_md5sum()):
raise SwiftError(
'Object upload verification failed: '
'md5 mismatch, local {0} != remote {1} '
'(remote object has not been removed)'
.format(contents.get_md5sum(), etag))
finally:
if fp is not None:
fp.close()
if old_manifest or old_slo_manifest_paths:
drs = []
delobjsmap = {}
if old_manifest:
scontainer, sprefix = old_manifest.split('/', 1)
sprefix = sprefix.rstrip('/') + '/'
delobjsmap[scontainer] = []
for part in self.list(scontainer, {'prefix': sprefix}):
if not part["success"]:
raise part["error"]
delobjsmap[scontainer].extend(
seg['name'] for seg in part['listing'])
if old_slo_manifest_paths:
for seg_to_delete in old_slo_manifest_paths:
if seg_to_delete in new_slo_manifest_paths:
continue
scont, sobj = \
seg_to_delete.split(b'/', 1)
delobjs_cont = delobjsmap.get(scont, [])
delobjs_cont.append(sobj)
delobjsmap[scont] = delobjs_cont
del_segs = []
for dscont, dsobjs in delobjsmap.items():
for dsobj in dsobjs:
del_seg = self.thread_manager.segment_pool.submit(
self._delete_segment, dscont, dsobj,
results_queue=results_queue
)
del_segs.append(del_seg)
for del_seg in interruptable_as_completed(del_segs):
drs.append(del_seg.result())
res['segment_delete_results'] = drs
# return dict for printing
res.update({
'success': True,
'status': 'uploaded',
'attempts': conn.attempts})
return res
except OSError as err:
traceback, err_time = report_traceback()
logger.exception(err)
if err.errno == ENOENT:
error = SwiftError('Local file %r not found' % path, exc=err)
else:
error = err
res.update({
'success': False,
'error': error,
'traceback': traceback,
'error_timestamp': err_time
})
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
# Delete related methods
#
def delete(self, container=None, objects=None, options=None):
"""
Delete operations on an account, optional container and optional list
of objects.
:param container: The container to delete or delete from.
:param objects: The list of objects to delete.
:param options: A dictionary containing options to override the global
options specified during the service object creation::
{
'yes_all': False,
'leave_segments': False,
'prefix': None,
'header': [],
}
:returns: A generator for returning the results of the delete
operations. Each result yielded from the generator is either
a 'delete_container', 'delete_object', 'delete_segment', or
'bulk_delete' dictionary containing the results of an
individual delete operation.
:raises ClientException:
:raises SwiftError:
"""
if options is not None:
options = dict(self._options, **options)
else:
options = self._options
if container is not None:
if objects is not None:
if options['prefix']:
objects = [obj for obj in objects
if obj.startswith(options['prefix'])]
rq = Queue()
obj_dels = {}
bulk_page_size = self._bulk_delete_page_size(objects)
if bulk_page_size > 1:
page_at_a_time = n_at_a_time(objects, bulk_page_size)
for page_slice in page_at_a_time:
for obj_slice in n_groups(
page_slice,
self._options['object_dd_threads']):
self._bulk_delete(container, obj_slice, options,
obj_dels)
else:
self._per_item_delete(container, objects, options,
obj_dels, rq)
# Start a thread to watch for delete results
Thread(
target=self._watch_futures, args=(obj_dels, rq)
).start()
# yield results as they become available, raising the first
# encountered exception
res = get_from_queue(rq)
while res is not None:
yield res
# Cancel the remaining jobs if necessary
if options['fail_fast'] and not res['success']:
for d in obj_dels.keys():
d.cancel()
res = get_from_queue(rq)
else:
for res in self._delete_container(container, options):
yield res
else:
if objects:
raise SwiftError('Objects specified without container')
if options['prefix']:
raise SwiftError('Prefix specified without container')
if options['yes_all']:
cancelled = False
containers = []
for part in self.list():
if part["success"]:
containers.extend(c['name'] for c in part['listing'])
else:
raise part["error"]
for con in containers:
if cancelled:
break
else:
for res in self._delete_container(
con, options=options):
yield res
# Cancel the remaining container deletes, but yield
# any pending results
if (not cancelled and options['fail_fast'] and
not res['success']):
cancelled = True
def _bulk_delete_page_size(self, objects):
'''
Given the iterable 'objects', will return how many items should be
deleted at a time.
:param objects: An iterable that supports 'len()'
:returns: The bulk delete page size (i.e. the max number of
objects that can be bulk deleted at once, as reported by
the cluster). If bulk delete is disabled, return 1
'''
if len(objects) <= 2 * self._options['object_dd_threads']:
# Not many objects; may as well delete one-by-one
return 1
try:
cap_result = self.capabilities()
if not cap_result['success']:
# This shouldn't actually happen, but just in case we start
# being more nuanced about our capabilities result...
return 1
except ClientException:
# Old swift, presumably; assume no bulk middleware
return 1
swift_info = cap_result['capabilities']
if 'bulk_delete' in swift_info:
return swift_info['bulk_delete'].get(
'max_deletes_per_request', 10000)
else:
return 1
def _per_item_delete(self, container, objects, options, rdict, rq):
for obj in objects:
obj_del = self.thread_manager.object_dd_pool.submit(
self._delete_object, container, obj, options,
results_queue=rq
)
obj_details = {'container': container, 'object': obj}
rdict[obj_del] = obj_details
@staticmethod
def _delete_segment(conn, container, obj, results_queue=None):
results_dict = {}
try:
res = {'success': True}
conn.delete_object(container, obj, response_dict=results_dict)
except Exception as err:
if not isinstance(err, ClientException) or err.http_status != 404:
traceback, err_time = report_traceback()
logger.exception(err)
res = {
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
}
res.update({
'action': 'delete_segment',
'container': container,
'object': obj,
'attempts': conn.attempts,
'response_dict': results_dict
})
if results_queue is not None:
results_queue.put(res)
return res
def _delete_object(self, conn, container, obj, options,
results_queue=None):
_headers = {}
_headers = split_headers(options.get('header', []))
res = {
'action': 'delete_object',
'container': container,
'object': obj
}
try:
old_manifest = None
query_string = None
if not options['leave_segments']:
try:
headers = conn.head_object(container, obj,
headers=_headers)
old_manifest = headers.get('x-object-manifest')
if config_true_value(headers.get('x-static-large-object')):
query_string = 'multipart-manifest=delete'
except ClientException as err:
if err.http_status != 404:
raise
results_dict = {}
conn.delete_object(container, obj,
headers=_headers,
query_string=query_string,
response_dict=results_dict)
if old_manifest:
dlo_segments_deleted = True
segment_pool = self.thread_manager.segment_pool
s_container, s_prefix = old_manifest.split('/', 1)
s_prefix = s_prefix.rstrip('/') + '/'
del_segs = []
for part in self.list(
container=s_container, options={'prefix': s_prefix}):
if part["success"]:
seg_list = [o["name"] for o in part["listing"]]
else:
raise part["error"]
for seg in seg_list:
del_seg = segment_pool.submit(
self._delete_segment, s_container,
seg, results_queue=results_queue
)
del_segs.append(del_seg)
for del_seg in interruptable_as_completed(del_segs):
del_res = del_seg.result()
if not del_res["success"]:
dlo_segments_deleted = False
res['dlo_segments_deleted'] = dlo_segments_deleted
res.update({
'success': True,
'response_dict': results_dict,
'attempts': conn.attempts,
})
except Exception as err:
traceback, err_time = report_traceback()
logger.exception(err)
res.update({
'success': False,
'error': err,
'traceback': traceback,
'error_timestamp': err_time
})
return res
return res
@staticmethod
def _delete_empty_container(conn, container, options):
results_dict = {}
_headers = {}
_headers = split_headers(options.get('header', []))
try:
conn.delete_container(container, headers=_headers,
response_dict=results_dict)
res = {'success': True}
except Exception as err