You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

737 lines
28 KiB

# Copyright 2016 Amazon.com, Inc. or its affiliates. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"). You
# may not use this file except in compliance with the License. A copy of
# the License is located at
#
# http://aws.amazon.com/apache2.0/
#
# or in the "license" file accompanying this file. This file 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 copy
import logging
import re
import threading
from s3transfer.bandwidth import BandwidthLimiter, LeakyBucket
from s3transfer.constants import ALLOWED_DOWNLOAD_ARGS, KB, MB
from s3transfer.copies import CopySubmissionTask
from s3transfer.delete import DeleteSubmissionTask
from s3transfer.download import DownloadSubmissionTask
from s3transfer.exceptions import CancelledError, FatalError
from s3transfer.futures import (
IN_MEMORY_DOWNLOAD_TAG,
IN_MEMORY_UPLOAD_TAG,
BoundedExecutor,
TransferCoordinator,
TransferFuture,
TransferMeta,
)
from s3transfer.upload import UploadSubmissionTask
from s3transfer.utils import (
CallArgs,
OSUtils,
SlidingWindowSemaphore,
TaskSemaphore,
add_s3express_defaults,
get_callbacks,
signal_not_transferring,
signal_transferring,
)
logger = logging.getLogger(__name__)
class TransferConfig:
def __init__(
self,
multipart_threshold=8 * MB,
multipart_chunksize=8 * MB,
max_request_concurrency=10,
max_submission_concurrency=5,
max_request_queue_size=1000,
max_submission_queue_size=1000,
max_io_queue_size=1000,
io_chunksize=256 * KB,
num_download_attempts=5,
max_in_memory_upload_chunks=10,
max_in_memory_download_chunks=10,
max_bandwidth=None,
):
"""Configurations for the transfer manager
:param multipart_threshold: The threshold for which multipart
transfers occur.
:param max_request_concurrency: The maximum number of S3 API
transfer-related requests that can happen at a time.
:param max_submission_concurrency: The maximum number of threads
processing a call to a TransferManager method. Processing a
call usually entails determining which S3 API requests that need
to be enqueued, but does **not** entail making any of the
S3 API data transferring requests needed to perform the transfer.
The threads controlled by ``max_request_concurrency`` is
responsible for that.
:param multipart_chunksize: The size of each transfer if a request
becomes a multipart transfer.
:param max_request_queue_size: The maximum amount of S3 API requests
that can be queued at a time.
:param max_submission_queue_size: The maximum amount of
TransferManager method calls that can be queued at a time.
:param max_io_queue_size: The maximum amount of read parts that
can be queued to be written to disk per download. The default
size for each elementin this queue is 8 KB.
:param io_chunksize: The max size of each chunk in the io queue.
Currently, this is size used when reading from the downloaded
stream as well.
:param num_download_attempts: The number of download attempts that
will be tried upon errors with downloading an object in S3. Note
that these retries account for errors that occur when streaming
down the data from s3 (i.e. socket errors and read timeouts that
occur after receiving an OK response from s3).
Other retryable exceptions such as throttling errors and 5xx errors
are already retried by botocore (this default is 5). The
``num_download_attempts`` does not take into account the
number of exceptions retried by botocore.
:param max_in_memory_upload_chunks: The number of chunks that can
be stored in memory at a time for all ongoing upload requests.
This pertains to chunks of data that need to be stored in memory
during an upload if the data is sourced from a file-like object.
The total maximum memory footprint due to a in-memory upload
chunks is roughly equal to:
max_in_memory_upload_chunks * multipart_chunksize
+ max_submission_concurrency * multipart_chunksize
``max_submission_concurrency`` has an affect on this value because
for each thread pulling data off of a file-like object, they may
be waiting with a single read chunk to be submitted for upload
because the ``max_in_memory_upload_chunks`` value has been reached
by the threads making the upload request.
:param max_in_memory_download_chunks: The number of chunks that can
be buffered in memory and **not** in the io queue at a time for all
ongoing download requests. This pertains specifically to file-like
objects that cannot be seeked. The total maximum memory footprint
due to a in-memory download chunks is roughly equal to:
max_in_memory_download_chunks * multipart_chunksize
:param max_bandwidth: The maximum bandwidth that will be consumed
in uploading and downloading file content. The value is in terms of
bytes per second.
"""
self.multipart_threshold = multipart_threshold
self.multipart_chunksize = multipart_chunksize
self.max_request_concurrency = max_request_concurrency
self.max_submission_concurrency = max_submission_concurrency
self.max_request_queue_size = max_request_queue_size
self.max_submission_queue_size = max_submission_queue_size
self.max_io_queue_size = max_io_queue_size
self.io_chunksize = io_chunksize
self.num_download_attempts = num_download_attempts
self.max_in_memory_upload_chunks = max_in_memory_upload_chunks
self.max_in_memory_download_chunks = max_in_memory_download_chunks
self.max_bandwidth = max_bandwidth
self._validate_attrs_are_nonzero()
def _validate_attrs_are_nonzero(self):
for attr, attr_val in self.__dict__.items():
if attr_val is not None and attr_val <= 0:
raise ValueError(
'Provided parameter %s of value %s must be greater than '
'0.' % (attr, attr_val)
)
class TransferManager:
ALLOWED_DOWNLOAD_ARGS = ALLOWED_DOWNLOAD_ARGS
ALLOWED_UPLOAD_ARGS = [
'ACL',
'CacheControl',
'ChecksumAlgorithm',
'ContentDisposition',
'ContentEncoding',
'ContentLanguage',
'ContentType',
'ExpectedBucketOwner',
'Expires',
'GrantFullControl',
'GrantRead',
'GrantReadACP',
'GrantWriteACP',
'Metadata',
'ObjectLockLegalHoldStatus',
'ObjectLockMode',
'ObjectLockRetainUntilDate',
'RequestPayer',
'ServerSideEncryption',
'StorageClass',
'SSECustomerAlgorithm',
'SSECustomerKey',
'SSECustomerKeyMD5',
'SSEKMSKeyId',
'SSEKMSEncryptionContext',
'Tagging',
'WebsiteRedirectLocation',
]
ALLOWED_COPY_ARGS = ALLOWED_UPLOAD_ARGS + [
'CopySourceIfMatch',
'CopySourceIfModifiedSince',
'CopySourceIfNoneMatch',
'CopySourceIfUnmodifiedSince',
'CopySourceSSECustomerAlgorithm',
'CopySourceSSECustomerKey',
'CopySourceSSECustomerKeyMD5',
'MetadataDirective',
'TaggingDirective',
]
ALLOWED_DELETE_ARGS = [
'MFA',
'VersionId',
'RequestPayer',
'ExpectedBucketOwner',
]
VALIDATE_SUPPORTED_BUCKET_VALUES = True
_UNSUPPORTED_BUCKET_PATTERNS = {
'S3 Object Lambda': re.compile(
r'^arn:(aws).*:s3-object-lambda:[a-z\-0-9]+:[0-9]{12}:'
r'accesspoint[/:][a-zA-Z0-9\-]{1,63}'
),
}
def __init__(self, client, config=None, osutil=None, executor_cls=None):
"""A transfer manager interface for Amazon S3
:param client: Client to be used by the manager
:param config: TransferConfig to associate specific configurations
:param osutil: OSUtils object to use for os-related behavior when
using with transfer manager.
:type executor_cls: s3transfer.futures.BaseExecutor
:param executor_cls: The class of executor to use with the transfer
manager. By default, concurrent.futures.ThreadPoolExecutor is used.
"""
self._client = client
self._config = config
if config is None:
self._config = TransferConfig()
self._osutil = osutil
if osutil is None:
self._osutil = OSUtils()
self._coordinator_controller = TransferCoordinatorController()
# A counter to create unique id's for each transfer submitted.
self._id_counter = 0
# The executor responsible for making S3 API transfer requests
self._request_executor = BoundedExecutor(
max_size=self._config.max_request_queue_size,
max_num_threads=self._config.max_request_concurrency,
tag_semaphores={
IN_MEMORY_UPLOAD_TAG: TaskSemaphore(
self._config.max_in_memory_upload_chunks
),
IN_MEMORY_DOWNLOAD_TAG: SlidingWindowSemaphore(
self._config.max_in_memory_download_chunks
),
},
executor_cls=executor_cls,
)
# The executor responsible for submitting the necessary tasks to
# perform the desired transfer
self._submission_executor = BoundedExecutor(
max_size=self._config.max_submission_queue_size,
max_num_threads=self._config.max_submission_concurrency,
executor_cls=executor_cls,
)
# There is one thread available for writing to disk. It will handle
# downloads for all files.
self._io_executor = BoundedExecutor(
max_size=self._config.max_io_queue_size,
max_num_threads=1,
executor_cls=executor_cls,
)
# The component responsible for limiting bandwidth usage if it
# is configured.
self._bandwidth_limiter = None
if self._config.max_bandwidth is not None:
logger.debug(
'Setting max_bandwidth to %s', self._config.max_bandwidth
)
leaky_bucket = LeakyBucket(self._config.max_bandwidth)
self._bandwidth_limiter = BandwidthLimiter(leaky_bucket)
self._register_handlers()
@property
def client(self):
return self._client
@property
def config(self):
return self._config
def upload(self, fileobj, bucket, key, extra_args=None, subscribers=None):
"""Uploads a file to S3
:type fileobj: str or seekable file-like object
:param fileobj: The name of a file to upload or a seekable file-like
object to upload. It is recommended to use a filename because
file-like objects may result in higher memory usage.
:type bucket: str
:param bucket: The name of the bucket to upload to
:type key: str
:param key: The name of the key to upload to
:type extra_args: dict
:param extra_args: Extra arguments that may be passed to the
client operation
:type subscribers: list(s3transfer.subscribers.BaseSubscriber)
:param subscribers: The list of subscribers to be invoked in the
order provided based on the event emit during the process of
the transfer request.
:rtype: s3transfer.futures.TransferFuture
:returns: Transfer future representing the upload
"""
if extra_args is None:
extra_args = {}
if subscribers is None:
subscribers = []
self._validate_all_known_args(extra_args, self.ALLOWED_UPLOAD_ARGS)
self._validate_if_bucket_supported(bucket)
self._add_operation_defaults(bucket, extra_args)
call_args = CallArgs(
fileobj=fileobj,
bucket=bucket,
key=key,
extra_args=extra_args,
subscribers=subscribers,
)
extra_main_kwargs = {}
if self._bandwidth_limiter:
extra_main_kwargs['bandwidth_limiter'] = self._bandwidth_limiter
return self._submit_transfer(
call_args, UploadSubmissionTask, extra_main_kwargs
)
def download(
self, bucket, key, fileobj, extra_args=None, subscribers=None
):
"""Downloads a file from S3
:type bucket: str
:param bucket: The name of the bucket to download from
:type key: str
:param key: The name of the key to download from
:type fileobj: str or seekable file-like object
:param fileobj: The name of a file to download or a seekable file-like
object to download. It is recommended to use a filename because
file-like objects may result in higher memory usage.
:type extra_args: dict
:param extra_args: Extra arguments that may be passed to the
client operation
:type subscribers: list(s3transfer.subscribers.BaseSubscriber)
:param subscribers: The list of subscribers to be invoked in the
order provided based on the event emit during the process of
the transfer request.
:rtype: s3transfer.futures.TransferFuture
:returns: Transfer future representing the download
"""
if extra_args is None:
extra_args = {}
if subscribers is None:
subscribers = []
self._validate_all_known_args(extra_args, self.ALLOWED_DOWNLOAD_ARGS)
self._validate_if_bucket_supported(bucket)
call_args = CallArgs(
bucket=bucket,
key=key,
fileobj=fileobj,
extra_args=extra_args,
subscribers=subscribers,
)
extra_main_kwargs = {'io_executor': self._io_executor}
if self._bandwidth_limiter:
extra_main_kwargs['bandwidth_limiter'] = self._bandwidth_limiter
return self._submit_transfer(
call_args, DownloadSubmissionTask, extra_main_kwargs
)
def copy(
self,
copy_source,
bucket,
key,
extra_args=None,
subscribers=None,
source_client=None,
):
"""Copies a file in S3
:type copy_source: dict
:param copy_source: The name of the source bucket, key name of the
source object, and optional version ID of the source object. The
dictionary format is:
``{'Bucket': 'bucket', 'Key': 'key', 'VersionId': 'id'}``. Note
that the ``VersionId`` key is optional and may be omitted.
:type bucket: str
:param bucket: The name of the bucket to copy to
:type key: str
:param key: The name of the key to copy to
:type extra_args: dict
:param extra_args: Extra arguments that may be passed to the
client operation
:type subscribers: a list of subscribers
:param subscribers: The list of subscribers to be invoked in the
order provided based on the event emit during the process of
the transfer request.
:type source_client: botocore or boto3 Client
:param source_client: The client to be used for operation that
may happen at the source object. For example, this client is
used for the head_object that determines the size of the copy.
If no client is provided, the transfer manager's client is used
as the client for the source object.
:rtype: s3transfer.futures.TransferFuture
:returns: Transfer future representing the copy
"""
if extra_args is None:
extra_args = {}
if subscribers is None:
subscribers = []
if source_client is None:
source_client = self._client
self._validate_all_known_args(extra_args, self.ALLOWED_COPY_ARGS)
if isinstance(copy_source, dict):
self._validate_if_bucket_supported(copy_source.get('Bucket'))
self._validate_if_bucket_supported(bucket)
call_args = CallArgs(
copy_source=copy_source,
bucket=bucket,
key=key,
extra_args=extra_args,
subscribers=subscribers,
source_client=source_client,
)
return self._submit_transfer(call_args, CopySubmissionTask)
def delete(self, bucket, key, extra_args=None, subscribers=None):
"""Delete an S3 object.
:type bucket: str
:param bucket: The name of the bucket.
:type key: str
:param key: The name of the S3 object to delete.
:type extra_args: dict
:param extra_args: Extra arguments that may be passed to the
DeleteObject call.
:type subscribers: list
:param subscribers: A list of subscribers to be invoked during the
process of the transfer request. Note that the ``on_progress``
callback is not invoked during object deletion.
:rtype: s3transfer.futures.TransferFuture
:return: Transfer future representing the deletion.
"""
if extra_args is None:
extra_args = {}
if subscribers is None:
subscribers = []
self._validate_all_known_args(extra_args, self.ALLOWED_DELETE_ARGS)
self._validate_if_bucket_supported(bucket)
call_args = CallArgs(
bucket=bucket,
key=key,
extra_args=extra_args,
subscribers=subscribers,
)
return self._submit_transfer(call_args, DeleteSubmissionTask)
def _validate_if_bucket_supported(self, bucket):
# s3 high level operations don't support some resources
# (eg. S3 Object Lambda) only direct API calls are available
# for such resources
if self.VALIDATE_SUPPORTED_BUCKET_VALUES:
for resource, pattern in self._UNSUPPORTED_BUCKET_PATTERNS.items():
match = pattern.match(bucket)
if match:
raise ValueError(
'TransferManager methods do not support %s '
'resource. Use direct client calls instead.' % resource
)
def _validate_all_known_args(self, actual, allowed):
for kwarg in actual:
if kwarg not in allowed:
raise ValueError(
"Invalid extra_args key '%s', "
"must be one of: %s" % (kwarg, ', '.join(allowed))
)
def _add_operation_defaults(self, bucket, extra_args):
add_s3express_defaults(bucket, extra_args)
def _submit_transfer(
self, call_args, submission_task_cls, extra_main_kwargs=None
):
if not extra_main_kwargs:
extra_main_kwargs = {}
# Create a TransferFuture to return back to the user
transfer_future, components = self._get_future_with_components(
call_args
)
# Add any provided done callbacks to the created transfer future
# to be invoked on the transfer future being complete.
for callback in get_callbacks(transfer_future, 'done'):
components['coordinator'].add_done_callback(callback)
# Get the main kwargs needed to instantiate the submission task
main_kwargs = self._get_submission_task_main_kwargs(
transfer_future, extra_main_kwargs
)
# Submit a SubmissionTask that will submit all of the necessary
# tasks needed to complete the S3 transfer.
self._submission_executor.submit(
submission_task_cls(
transfer_coordinator=components['coordinator'],
main_kwargs=main_kwargs,
)
)
# Increment the unique id counter for future transfer requests
self._id_counter += 1
return transfer_future
def _get_future_with_components(self, call_args):
transfer_id = self._id_counter
# Creates a new transfer future along with its components
transfer_coordinator = TransferCoordinator(transfer_id=transfer_id)
# Track the transfer coordinator for transfers to manage.
self._coordinator_controller.add_transfer_coordinator(
transfer_coordinator
)
# Also make sure that the transfer coordinator is removed once
# the transfer completes so it does not stick around in memory.
transfer_coordinator.add_done_callback(
self._coordinator_controller.remove_transfer_coordinator,
transfer_coordinator,
)
components = {
'meta': TransferMeta(call_args, transfer_id=transfer_id),
'coordinator': transfer_coordinator,
}
transfer_future = TransferFuture(**components)
return transfer_future, components
def _get_submission_task_main_kwargs(
self, transfer_future, extra_main_kwargs
):
main_kwargs = {
'client': self._client,
'config': self._config,
'osutil': self._osutil,
'request_executor': self._request_executor,
'transfer_future': transfer_future,
}
main_kwargs.update(extra_main_kwargs)
return main_kwargs
def _register_handlers(self):
# Register handlers to enable/disable callbacks on uploads.
event_name = 'request-created.s3'
self._client.meta.events.register_first(
event_name,
signal_not_transferring,
unique_id='s3upload-not-transferring',
)
self._client.meta.events.register_last(
event_name, signal_transferring, unique_id='s3upload-transferring'
)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_value, *args):
cancel = False
cancel_msg = ''
cancel_exc_type = FatalError
# If a exception was raised in the context handler, signal to cancel
# all of the inprogress futures in the shutdown.
if exc_type:
cancel = True
cancel_msg = str(exc_value)
if not cancel_msg:
cancel_msg = repr(exc_value)
# If it was a KeyboardInterrupt, the cancellation was initiated
# by the user.
if isinstance(exc_value, KeyboardInterrupt):
cancel_exc_type = CancelledError
self._shutdown(cancel, cancel_msg, cancel_exc_type)
def shutdown(self, cancel=False, cancel_msg=''):
"""Shutdown the TransferManager
It will wait till all transfers complete before it completely shuts
down.
:type cancel: boolean
:param cancel: If True, calls TransferFuture.cancel() for
all in-progress in transfers. This is useful if you want the
shutdown to happen quicker.
:type cancel_msg: str
:param cancel_msg: The message to specify if canceling all in-progress
transfers.
"""
self._shutdown(cancel, cancel, cancel_msg)
def _shutdown(self, cancel, cancel_msg, exc_type=CancelledError):
if cancel:
# Cancel all in-flight transfers if requested, before waiting
# for them to complete.
self._coordinator_controller.cancel(cancel_msg, exc_type)
try:
# Wait until there are no more in-progress transfers. This is
# wrapped in a try statement because this can be interrupted
# with a KeyboardInterrupt that needs to be caught.
self._coordinator_controller.wait()
except KeyboardInterrupt:
# If not errors were raised in the try block, the cancel should
# have no coordinators it needs to run cancel on. If there was
# an error raised in the try statement we want to cancel all of
# the inflight transfers before shutting down to speed that
# process up.
self._coordinator_controller.cancel('KeyboardInterrupt()')
raise
finally:
# Shutdown all of the executors.
self._submission_executor.shutdown()
self._request_executor.shutdown()
self._io_executor.shutdown()
class TransferCoordinatorController:
def __init__(self):
"""Abstraction to control all transfer coordinators
This abstraction allows the manager to wait for inprogress transfers
to complete and cancel all inprogress transfers.
"""
self._lock = threading.Lock()
self._tracked_transfer_coordinators = set()
@property
def tracked_transfer_coordinators(self):
"""The set of transfer coordinators being tracked"""
with self._lock:
# We return a copy because the set is mutable and if you were to
# iterate over the set, it may be changing in length due to
# additions and removals of transfer coordinators.
return copy.copy(self._tracked_transfer_coordinators)
def add_transfer_coordinator(self, transfer_coordinator):
"""Adds a transfer coordinator of a transfer to be canceled if needed
:type transfer_coordinator: s3transfer.futures.TransferCoordinator
:param transfer_coordinator: The transfer coordinator for the
particular transfer
"""
with self._lock:
self._tracked_transfer_coordinators.add(transfer_coordinator)
def remove_transfer_coordinator(self, transfer_coordinator):
"""Remove a transfer coordinator from cancellation consideration
Typically, this method is invoked by the transfer coordinator itself
to remove its self when it completes its transfer.
:type transfer_coordinator: s3transfer.futures.TransferCoordinator
:param transfer_coordinator: The transfer coordinator for the
particular transfer
"""
with self._lock:
self._tracked_transfer_coordinators.remove(transfer_coordinator)
def cancel(self, msg='', exc_type=CancelledError):
"""Cancels all inprogress transfers
This cancels the inprogress transfers by calling cancel() on all
tracked transfer coordinators.
:param msg: The message to pass on to each transfer coordinator that
gets cancelled.
:param exc_type: The type of exception to set for the cancellation
"""
for transfer_coordinator in self.tracked_transfer_coordinators:
transfer_coordinator.cancel(msg, exc_type)
def wait(self):
"""Wait until there are no more inprogress transfers
This will not stop when failures are encountered and not propagate any
of these errors from failed transfers, but it can be interrupted with
a KeyboardInterrupt.
"""
try:
transfer_coordinator = None
for transfer_coordinator in self.tracked_transfer_coordinators:
transfer_coordinator.result()
except KeyboardInterrupt:
logger.debug('Received KeyboardInterrupt in wait()')
# If Keyboard interrupt is raised while waiting for
# the result, then exit out of the wait and raise the
# exception
if transfer_coordinator:
logger.debug(
'On KeyboardInterrupt was waiting for %s',
transfer_coordinator,
)
raise
except Exception:
# A general exception could have been thrown because
# of result(). We just want to ignore this and continue
# because we at least know that the transfer coordinator
# has completed.
pass

Powered by BW's shoe-string budget.