diff --git a/swh/objstorage/api/client.py b/swh/objstorage/api/client.py index f6d3887..bcf5661 100644 --- a/swh/objstorage/api/client.py +++ b/swh/objstorage/api/client.py @@ -1,64 +1,67 @@ # Copyright (C) 2015-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information from swh.core.api import SWHRemoteAPI from swh.model import hashutil from ..objstorage import ObjStorage, DEFAULT_CHUNK_SIZE from ..exc import ObjStorageAPIError class RemoteObjStorage(ObjStorage, SWHRemoteAPI): """Proxy to a remote object storage. This class allows to connect to an object storage server via http protocol. Attributes: url (string): The url of the server to connect. Must end with a '/' session: The session to send requests. """ def __init__(self, url): super().__init__(api_exception=ObjStorageAPIError, url=url) def check_config(self, *, check_write): return self.post('check_config', {'check_write': check_write}) def __contains__(self, obj_id): return self.post('content/contains', {'obj_id': obj_id}) def add(self, content, obj_id=None, check_presence=True): return self.post('content/add', {'content': content, 'obj_id': obj_id, 'check_presence': check_presence}) def get(self, obj_id): return self.post('content/get', {'obj_id': obj_id}) def get_batch(self, obj_ids): return self.post('content/get/batch', {'obj_ids': obj_ids}) def check(self, obj_id): return self.post('content/check', {'obj_id': obj_id}) + def delete(self, obj_id): + return self.post('content/delete', {'obj_id': obj_id}) + # Management methods def get_random(self, batch_size): return self.post('content/get/random', {'batch_size': batch_size}) # Streaming methods def add_stream(self, content_iter, obj_id, check_presence=True): obj_id = hashutil.hash_to_hex(obj_id) return self.post_stream('content/add_stream/{}'.format(obj_id), params={'check_presence': check_presence}, data=content_iter) def get_stream(self, obj_id, chunk_size=DEFAULT_CHUNK_SIZE): obj_id = hashutil.hash_to_hex(obj_id) return super().get_stream('content/get_stream/{}'.format(obj_id), chunk_size=chunk_size) diff --git a/swh/objstorage/api/server.py b/swh/objstorage/api/server.py index 024fd69..9110687 100644 --- a/swh/objstorage/api/server.py +++ b/swh/objstorage/api/server.py @@ -1,147 +1,154 @@ # Copyright (C) 2015-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import asyncio import aiohttp.web import click from swh.core import config from swh.core.api_async import (SWHRemoteAPI, decode_request, encode_data_server as encode_data) from swh.model import hashutil from swh.objstorage import get_objstorage DEFAULT_CONFIG_PATH = 'objstorage/server' DEFAULT_CONFIG = { 'cls': ('str', 'pathslicing'), 'args': ('dict', { 'root': '/srv/softwareheritage/objects', 'slicing': '0:2/2:4/4:6', }), 'client_max_size': ('int', 1024 * 1024 * 1024), } @asyncio.coroutine def index(request): return aiohttp.web.Response(body="SWH Objstorage API server") @asyncio.coroutine def check_config(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].check_config(**req)) @asyncio.coroutine def contains(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].__contains__(**req)) @asyncio.coroutine def add_bytes(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].add(**req)) @asyncio.coroutine def get_bytes(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].get(**req)) @asyncio.coroutine def get_batch(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].get_batch(**req)) @asyncio.coroutine def check(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].check(**req)) +@asyncio.coroutine +def delete(request): + req = yield from decode_request(request) + return encode_data(request.app['objstorage'].delete(**req)) + + # Management methods @asyncio.coroutine def get_random_contents(request): req = yield from decode_request(request) return encode_data(request.app['objstorage'].get_random(**req)) # Streaming methods @asyncio.coroutine def add_stream(request): hex_id = request.match_info['hex_id'] obj_id = hashutil.hash_to_bytes(hex_id) check_pres = (request.query.get('check_presence', '').lower() == 'true') objstorage = request.app['objstorage'] if check_pres and obj_id in objstorage: return encode_data(obj_id) with objstorage.chunk_writer(obj_id) as write: # XXX (3.5): use 'async for chunk in request.content.iter_any()' while not request.content.at_eof(): chunk = yield from request.content.readany() write(chunk) return encode_data(obj_id) @asyncio.coroutine def get_stream(request): hex_id = request.match_info['hex_id'] obj_id = hashutil.hash_to_bytes(hex_id) response = aiohttp.web.StreamResponse() yield from response.prepare(request) for chunk in request.app['objstorage'].get_stream(obj_id, 2 << 20): response.write(chunk) yield from response.drain() return response def make_app(config, **kwargs): if 'client_max_size' in config: kwargs['client_max_size'] = config['client_max_size'] app = SWHRemoteAPI(**kwargs) app.router.add_route('GET', '/', index) app.router.add_route('POST', '/check_config', check_config) app.router.add_route('POST', '/content/contains', contains) app.router.add_route('POST', '/content/add', add_bytes) app.router.add_route('POST', '/content/get', get_bytes) app.router.add_route('POST', '/content/get/batch', get_batch) app.router.add_route('POST', '/content/get/random', get_random_contents) app.router.add_route('POST', '/content/check', check) + app.router.add_route('POST', '/content/delete', delete) app.router.add_route('POST', '/content/add_stream/{hex_id}', add_stream) app.router.add_route('GET', '/content/get_stream/{hex_id}', get_stream) app.update(config) app['objstorage'] = get_objstorage(app['cls'], app['args']) return app def make_app_from_configfile(config_path=DEFAULT_CONFIG_PATH, **kwargs): return make_app(config.read(config_path, DEFAULT_CONFIG), **kwargs) @click.command() @click.argument('config-path', required=1) @click.option('--host', default='0.0.0.0', help="Host to run the server") @click.option('--port', default=5003, type=click.INT, help="Binding port of the server") @click.option('--debug/--nodebug', default=True, help="Indicates if the server should run in debug mode") def launch(config_path, host, port, debug): app = make_app_from_configfile(config_path, debug=bool(debug)) aiohttp.web.run_app(app, host=host, port=int(port)) if __name__ == '__main__': launch() diff --git a/swh/objstorage/cloud/objstorage_azure.py b/swh/objstorage/cloud/objstorage_azure.py index cc11ec2..982ecd1 100644 --- a/swh/objstorage/cloud/objstorage_azure.py +++ b/swh/objstorage/cloud/objstorage_azure.py @@ -1,114 +1,126 @@ # Copyright (C) 2016-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import gzip from swh.objstorage.objstorage import ObjStorage, compute_hash from swh.objstorage.exc import ObjNotFoundError, Error from swh.model import hashutil from azure.storage.blob import BlockBlobService from azure.common import AzureMissingResourceHttpError class AzureCloudObjStorage(ObjStorage): """ObjStorage with azure abilities. """ def __init__(self, account_name, api_secret_key, container_name): self.block_blob_service = BlockBlobService( account_name=account_name, account_key=api_secret_key) self.container_name = container_name def _internal_id(self, obj_id): """Internal id is the hex version in objstorage. """ return hashutil.hash_to_hex(obj_id) def check_config(self, *, check_write): """Check the configuration for this object storage""" props = self.block_blob_service.get_container_properties( self.container_name ) # FIXME: check_write is ignored here return bool(props) def __contains__(self, obj_id): """Does the storage contains the obj_id. """ hex_obj_id = self._internal_id(obj_id) return self.block_blob_service.exists( container_name=self.container_name, blob_name=hex_obj_id) def __iter__(self): """Iterate over the objects present in the storage. """ for obj in self.block_blob_service.list_blobs(self.container_name): yield hashutil.hash_to_bytes(obj.name) def __len__(self): """Compute the number of objects in the current object storage. Returns: number of objects contained in the storage. """ return sum(1 for i in self) def add(self, content, obj_id=None, check_presence=True): """Add an obj in storage if it's not there already. """ if obj_id is None: # Checksum is missing, compute it on the fly. obj_id = compute_hash(content) if check_presence and obj_id in self: return obj_id hex_obj_id = self._internal_id(obj_id) # Send the gzipped content self.block_blob_service.create_blob_from_bytes( container_name=self.container_name, blob_name=hex_obj_id, blob=gzip.compress(content)) return obj_id def restore(self, content, obj_id=None): """Restore a content. """ return self.add(content, obj_id, check_presence=False) def get(self, obj_id): """Retrieve blob's content if found. """ hex_obj_id = self._internal_id(obj_id) try: blob = self.block_blob_service.get_blob_to_bytes( container_name=self.container_name, blob_name=hex_obj_id) except AzureMissingResourceHttpError: raise ObjNotFoundError('Content %s not found!' % hex_obj_id) return gzip.decompress(blob.content) def check(self, obj_id): """Check the content integrity. """ obj_content = self.get(obj_id) content_obj_id = compute_hash(obj_content) if content_obj_id != obj_id: raise Error(obj_id) + + def delete(self, obj_id): + """Delete an object.""" + hex_obj_id = self._internal_id(obj_id) + try: + self.block_blob_service.delete_blob( + container_name=self.container_name, + blob_name=hex_obj_id) + except AzureMissingResourceHttpError: + raise ObjNotFoundError('Content {} not found!'.format(hex_obj_id)) + + return True diff --git a/swh/objstorage/cloud/objstorage_cloud.py b/swh/objstorage/cloud/objstorage_cloud.py index 805e271..c653ebc 100644 --- a/swh/objstorage/cloud/objstorage_cloud.py +++ b/swh/objstorage/cloud/objstorage_cloud.py @@ -1,163 +1,167 @@ # Copyright (C) 2016-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import abc from swh.model import hashutil from swh.objstorage.objstorage import ObjStorage, compute_hash from swh.objstorage.exc import ObjNotFoundError, Error from libcloud.storage import providers from libcloud.storage.types import Provider, ObjectDoesNotExistError class CloudObjStorage(ObjStorage, metaclass=abc.ABCMeta): """Abstract ObjStorage that connect to a cloud using Libcloud Implementations of this class must redefine the _get_provider method to make it return a driver provider (i.e. object that supports `get_driver` method) which return a LibCloud driver (see https://libcloud.readthedocs.io/en/latest/storage/api.html). """ def __init__(self, api_key, api_secret_key, container_name): self.driver = self._get_driver(api_key, api_secret_key) self.container_name = container_name self.container = self.driver.get_container( container_name=container_name) def _get_driver(self, api_key, api_secret_key): """Initialize a driver to communicate with the cloud Args: api_key: key to connect to the API. api_secret_key: secret key for authentification. Returns: a Libcloud driver to a cloud storage. """ # Get the driver class from its description. cls = providers.get_driver(self._get_provider()) # Initialize the driver. return cls(api_key, api_secret_key) @abc.abstractmethod def _get_provider(self): """Get a libcloud driver provider This method must be overriden by subclasses to specify which of the native libcloud driver the current storage should connect to. Alternatively, provider for a custom driver may be returned, in which case the provider will have to support `get_driver` method. """ raise NotImplementedError('%s must implement `get_provider` method' % type(self)) def check_config(self, *, check_write): """Check the configuration for this object storage""" # FIXME: hopefully this blew up during instantiation return True def __contains__(self, obj_id): try: self._get_object(obj_id) except ObjNotFoundError: return False else: return True def __iter__(self): """ Iterate over the objects present in the storage Warning: Iteration over the contents of a cloud-based object storage may have bad efficiency: due to the very high amount of objects in it and the fact that it is remote, get all the contents of the current object storage may result in a lot of network requests. You almost certainly don't want to use this method in production. """ yield from map(lambda obj: obj.name, self.driver.iterate_container_objects(self.container)) def __len__(self): """Compute the number of objects in the current object storage. Warning: this currently uses `__iter__`, its warning about bad performance applies. Returns: number of objects contained in the storage. """ return sum(1 for i in self) def add(self, content, obj_id=None, check_presence=True): if obj_id is None: # Checksum is missing, compute it on the fly. obj_id = compute_hash(content) if check_presence and obj_id in self: return obj_id self._put_object(content, obj_id) return obj_id def restore(self, content, obj_id=None): return self.add(content, obj_id, check_presence=False) def get(self, obj_id): return bytes(self._get_object(obj_id).as_stream()) def check(self, obj_id): # Check that the file exists, as _get_object raises ObjNotFoundError self._get_object(obj_id) # Check the content integrity obj_content = self.get(obj_id) content_obj_id = compute_hash(obj_content) if content_obj_id != obj_id: raise Error(obj_id) + def delete(self, obj_id): + obj = self._get_object(obj_id) + return self.driver.delete_object(obj) + def _get_object(self, obj_id): """Get a Libcloud wrapper for an object pointer. This wrapper does not retrieve the content of the object directly. """ hex_obj_id = hashutil.hash_to_hex(obj_id) try: return self.driver.get_object(self.container_name, hex_obj_id) except ObjectDoesNotExistError as e: raise ObjNotFoundError(e.object_name) def _put_object(self, content, obj_id): """Create an object in the cloud storage. Created object will contain the content and be referenced by the given id. """ hex_obj_id = hashutil.hash_to_hex(obj_id) self.driver.upload_object_via_stream(iter(content), self.container, hex_obj_id) class AwsCloudObjStorage(CloudObjStorage): """ Amazon's S3 Cloud-based object storage """ def _get_provider(self): return Provider.S3 class OpenStackCloudObjStorage(CloudObjStorage): """ OpenStack Swift Cloud based object storage """ def _get_provider(self): return Provider.OPENSTACK_SWIFT diff --git a/swh/objstorage/multiplexer/filter/filter.py b/swh/objstorage/multiplexer/filter/filter.py index 8f56887..739655e 100644 --- a/swh/objstorage/multiplexer/filter/filter.py +++ b/swh/objstorage/multiplexer/filter/filter.py @@ -1,75 +1,78 @@ # Copyright (C) 2015-2016 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information from ...objstorage import ObjStorage class ObjStorageFilter(ObjStorage): """Base implementation of a filter that allow inputs on ObjStorage or not. This class copy the API of ...objstorage in order to filter the inputs of this class. If the operation is allowed, return the result of this operation applied to the destination implementation. Otherwise, just return without any operation. This class is an abstract base class for a classic read/write storage. Filters can inherit from it and only redefine some methods in order to change behavior. """ def __init__(self, storage): self.storage = storage def check_config(self, *, check_write): """Check the object storage for proper configuration. Args: check_write: check whether writes to the objstorage will succeed Returns: True if the storage is properly configured """ return self.storage.check_config(check_write=check_write) def __contains__(self, *args, **kwargs): return self.storage.__contains__(*args, **kwargs) def __iter__(self): """ Iterates over the content of each storages Warning: The `__iter__` methods frequently have bad performance. You almost certainly don't want to use this method in production as the wrapped storage may cause performance issues. """ return self.storage.__iter__() def __len__(self): """ Compute the number of objects in the current object storage. Warning: performance issue in `__iter__` also applies here. Returns: number of objects contained in the storage. """ return self.storage.__len__() def add(self, content, obj_id=None, check_presence=True, *args, **kwargs): return self.storage.add(content, obj_id, check_presence, *args, **kwargs) def restore(self, content, obj_id=None, *args, **kwargs): return self.storage.restore(content, obj_id, *args, **kwargs) def get(self, obj_id, *args, **kwargs): return self.storage.get(obj_id, *args, **kwargs) def check(self, obj_id, *args, **kwargs): return self.storage.check(obj_id, *args, **kwargs) + def delete(self, obj_id, *args, **kwargs): + return self.storage.delete(obj_id, *args, **kwargs) + def get_random(self, batch_size, *args, **kwargs): return self.storage.get_random(batch_size, *args, **kwargs) diff --git a/swh/objstorage/multiplexer/multiplexer_objstorage.py b/swh/objstorage/multiplexer/multiplexer_objstorage.py index 3041445..00a043f 100644 --- a/swh/objstorage/multiplexer/multiplexer_objstorage.py +++ b/swh/objstorage/multiplexer/multiplexer_objstorage.py @@ -1,174 +1,177 @@ # Copyright (C) 2015-2016 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import random from ..objstorage import ObjStorage from ..exc import ObjNotFoundError class MultiplexerObjStorage(ObjStorage): """Implementation of ObjStorage that distributes between multiple storages. The multiplexer object storage allows an input to be demultiplexed among multiple storages that will or will not accept it by themselves (see .filter package). As the ids can be differents, no pre-computed ids should be submitted. Also, there are no guarantees that the returned ids can be used directly into the storages that the multiplexer manage. Use case examples follow. Example 1:: storage_v1 = filter.read_only(PathSlicingObjStorage('/dir1', '0:2/2:4/4:6')) storage_v2 = PathSlicingObjStorage('/dir2', '0:1/0:5') storage = MultiplexerObjStorage([storage_v1, storage_v2]) When using 'storage', all the new contents will only be added to the v2 storage, while it will be retrievable from both. Example 2:: storage_v1 = filter.id_regex( PathSlicingObjStorage('/dir1', '0:2/2:4/4:6'), r'[^012].*' ) storage_v2 = filter.if_regex( PathSlicingObjStorage('/dir2', '0:1/0:5'), r'[012]/*' ) storage = MultiplexerObjStorage([storage_v1, storage_v2]) When using this storage, the contents with a sha1 starting with 0, 1 or 2 will be redirected (read AND write) to the storage_v2, while the others will be redirected to the storage_v1. If a content starting with 0, 1 or 2 is present in the storage_v1, it would be ignored anyway. """ def __init__(self, storages): self.storages = storages def check_config(self, *, check_write): return all( storage.check_config(check_write=check_write) for storage in self.storages ) def __contains__(self, obj_id): """Check the object storage for proper configuration. Args: check_write: check whether writes to the objstorage will succeed Returns: True if the storage is properly configured """ for storage in self.storages: if obj_id in storage: return True return False def __iter__(self): """Iterates over the content of each storages Due to the demultiplexer nature, same content can be in multiple storages and may be yielded multiple times. Warning: The ``__iter__`` methods frequently have bad performance. You almost certainly don't want to use this method in production. """ for storage in self.storages: yield from storage def __len__(self): """Compute the number of objects in the current object storage. Identical objects present in multiple storages will be counted as multiple objects. Warning: this currently uses `__iter__`, its warning about bad performance applies. Returns: number of objects contained in the storage. """ return sum(map(len, self.storages)) def add(self, content, obj_id=None, check_presence=True): """ Add a new object to the object storage. If the adding step works in all the storages that accept this content, this is a success. Otherwise, the full adding step is an error even if it succeed in some of the storages. Args: content: content of the object to be added to the storage. obj_id: checksum of [bytes] using [ID_HASH_ALGO] algorithm. When given, obj_id will be trusted to match the bytes. If missing, obj_id will be computed on the fly. check_presence: indicate if the presence of the content should be verified before adding the file. Returns: an id of the object into the storage. As the write-storages are always readable as well, any id will be valid to retrieve a content. """ return [storage.add(content, obj_id, check_presence) for storage in self.storages].pop() def restore(self, content, obj_id=None): return [storage.restore(content, obj_id) for storage in self.storages].pop() def get(self, obj_id): for storage in self.storages: try: return storage.get(obj_id) except ObjNotFoundError: continue # If no storage contains this content, raise the error raise ObjNotFoundError(obj_id) def check(self, obj_id): nb_present = 0 for storage in self.storages: try: storage.check(obj_id) except ObjNotFoundError: continue else: nb_present += 1 # If there is an Error because of a corrupted file, then let it pass. # Raise the ObjNotFoundError only if the content coulnd't be found in # all the storages. if nb_present == 0: raise ObjNotFoundError(obj_id) + def delete(self, obj_id): + return all(storage.delete(obj_id) for storage in self.storages) + def get_random(self, batch_size): storages_set = [storage for storage in self.storages if len(storage) > 0] if len(storages_set) <= 0: return [] while storages_set: storage = random.choice(storages_set) try: return storage.get_random(batch_size) except NotImplementedError: storages_set.remove(storage) # There is no storage that allow the get_random operation raise NotImplementedError( "There is no storage implementation into the multiplexer that " "support the 'get_random' operation" ) diff --git a/swh/objstorage/objstorage.py b/swh/objstorage/objstorage.py index 4663c24..a7cd29a 100644 --- a/swh/objstorage/objstorage.py +++ b/swh/objstorage/objstorage.py @@ -1,243 +1,257 @@ # Copyright (C) 2015-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import abc from swh.model import hashutil from .exc import ObjNotFoundError ID_HASH_ALGO = 'sha1' ID_HASH_LENGTH = 40 # Size in bytes of the hash hexadecimal representation. DEFAULT_CHUNK_SIZE = 2 * 1024 * 1024 # Size in bytes of the streaming chunks def compute_hash(content): return hashutil.hash_data( content, algorithms=[ID_HASH_ALGO] ).get(ID_HASH_ALGO) class ObjStorage(metaclass=abc.ABCMeta): """ High-level API to manipulate the Software Heritage object storage. Conceptually, the object storage offers the following methods: - check_config() check if the object storage is properly configured - __contains__() check if an object is present, by object id - add() add a new object, returning an object id - restore() same as add() but erase an already existed content - get() retrieve the content of an object, by object id - check() check the integrity of an object, by object id + - delete() remove an object And some management methods: - get_random() get random object id of existing contents (used for the content integrity checker). Some of the methods have available streaming equivalents: - add_stream() same as add() but with a chunked iterator - restore_stream() same as add_stream() but erase already existing content - get_stream() same as get() but returns a chunked iterator Each implementation of this interface can have a different behavior and its own way to store the contents. """ @abc.abstractmethod def check_config(self, *, check_write): """Check whether the object storage is properly configured. Args: check_write (bool): if True, check if writes to the object storage can succeed. Returns: True if the configuration check worked, an exception if it didn't. """ pass @abc.abstractmethod def __contains__(self, obj_id, *args, **kwargs): """Indicate if the given object is present in the storage. Args: obj_id (bytes): object identifier. Returns: True iff the object is present in the current object storage. """ pass @abc.abstractmethod def add(self, content, obj_id=None, check_presence=True, *args, **kwargs): """Add a new object to the object storage. Args: content (bytes): object's raw content to add in storage. obj_id (bytes): checksum of [bytes] using [ID_HASH_ALGO] algorithm. When given, obj_id will be trusted to match the bytes. If missing, obj_id will be computed on the fly. check_presence (bool): indicate if the presence of the content should be verified before adding the file. Returns: the id (bytes) of the object into the storage. """ pass def restore(self, content, obj_id=None, *args, **kwargs): """Restore a content that have been corrupted. This function is identical to add but does not check if the object id is already in the file system. The default implementation provided by the current class is suitable for most cases. Args: content (bytes): object's raw content to add in storage obj_id (bytes): checksum of `bytes` as computed by ID_HASH_ALGO. When given, obj_id will be trusted to match bytes. If missing, obj_id will be computed on the fly. """ # check_presence to false will erase the potential previous content. return self.add(content, obj_id, check_presence=False) @abc.abstractmethod def get(self, obj_id, *args, **kwargs): """Retrieve the content of a given object. Args: obj_id (bytes): object id. Returns: the content of the requested object as bytes. Raises: ObjNotFoundError: if the requested object is missing. """ pass def get_batch(self, obj_ids, *args, **kwargs): """Retrieve objects' raw content in bulk from storage. Note: This function does have a default implementation in ObjStorage that is suitable for most cases. For object storages that needs to do the minimal number of requests possible (ex: remote object storages), that method can be overriden to perform a more efficient operation. Args: obj_ids ([bytes]: list of object ids. Returns: list of resulting contents, or None if the content could not be retrieved. Do not raise any exception as a fail for one content will not cancel the whole request. """ for obj_id in obj_ids: try: yield self.get(obj_id) except ObjNotFoundError: yield None @abc.abstractmethod def check(self, obj_id, *args, **kwargs): """Perform an integrity check for a given object. Verify that the file object is in place and that the gziped content matches the object id. Args: obj_id (bytes): object identifier. Raises: ObjNotFoundError: if the requested object is missing. Error: if the request object is corrupted. """ pass + @abc.abstractmethod + def delete(self, obj_id, *args, **kwargs): + """Delete an object. + + Args: + obj_id (bytes): object identifier. + + Raises: + ObjNotFoundError: if the requested object is missing. + + """ + pass + # Management methods def get_random(self, batch_size, *args, **kwargs): """Get random ids of existing contents. This method is used in order to get random ids to perform content integrity verifications on random contents. Args: batch_size (int): Number of ids that will be given Yields: An iterable of ids (bytes) of contents that are in the current object storage. """ pass # Streaming methods def add_stream(self, content_iter, obj_id, check_presence=True): """Add a new object to the object storage using streaming. This function is identical to add() except it takes a generator that yields the chunked content instead of the whole content at once. Args: content (bytes): chunked generator that yields the object's raw content to add in storage. obj_id (bytes): object identifier check_presence (bool): indicate if the presence of the content should be verified before adding the file. Returns: the id (bytes) of the object into the storage. """ raise NotImplementedError def restore_stream(self, content_iter, obj_id=None): """Restore a content that have been corrupted using streaming. This function is identical to restore() except it takes a generator that yields the chunked content instead of the whole content at once. The default implementation provided by the current class is suitable for most cases. Args: content (bytes): chunked generator that yields the object's raw content to add in storage. obj_id (bytes): object identifier """ # check_presence to false will erase the potential previous content. return self.add_stream(content_iter, obj_id, check_presence=False) def get_stream(self, obj_id, chunk_size=DEFAULT_CHUNK_SIZE): """Retrieve the content of a given object as a chunked iterator. Args: obj_id (bytes): object id. Returns: the content of the requested object as bytes. Raises: ObjNotFoundError: if the requested object is missing. """ raise NotImplementedError diff --git a/swh/objstorage/objstorage_pathslicing.py b/swh/objstorage/objstorage_pathslicing.py index 6980374..9729f4d 100644 --- a/swh/objstorage/objstorage_pathslicing.py +++ b/swh/objstorage/objstorage_pathslicing.py @@ -1,327 +1,338 @@ # Copyright (C) 2015-2017 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import functools import os import gzip import tempfile import random from contextlib import contextmanager from swh.model import hashutil from .objstorage import (ObjStorage, compute_hash, ID_HASH_ALGO, ID_HASH_LENGTH, DEFAULT_CHUNK_SIZE) from .exc import ObjNotFoundError, Error GZIP_BUFSIZ = 1048576 DIR_MODE = 0o755 FILE_MODE = 0o644 @contextmanager def _write_obj_file(hex_obj_id, objstorage): """ Context manager for writing object files to the object storage. During writing, data are written to a temporary file, which is atomically renamed to the right file name after closing. This context manager also takes care of (gzip) compressing the data on the fly. Usage sample: with _write_obj_file(hex_obj_id, objstorage): f.write(obj_data) Yields: a file-like object open for writing bytes. """ # Get the final paths and create the directory if absent. dir = objstorage._obj_dir(hex_obj_id) if not os.path.isdir(dir): os.makedirs(dir, DIR_MODE, exist_ok=True) path = os.path.join(dir, hex_obj_id) # Create a temporary file. (tmp, tmp_path) = tempfile.mkstemp(suffix='.tmp', prefix='hex_obj_id.', dir=dir) # Open the file and yield it for writing. tmp_f = os.fdopen(tmp, 'wb') with gzip.GzipFile(filename=tmp_path, fileobj=tmp_f) as f: yield f # Then close the temporary file and move it to the right directory. tmp_f.close() os.chmod(tmp_path, FILE_MODE) os.rename(tmp_path, path) @contextmanager def _read_obj_file(hex_obj_id, objstorage): """ Context manager for reading object file in the object storage. Usage sample: with _read_obj_file(hex_obj_id, objstorage) as f: b = f.read() Yields: a file-like object open for reading bytes. """ path = objstorage._obj_path(hex_obj_id) with gzip.GzipFile(path, 'rb') as f: yield f class PathSlicingObjStorage(ObjStorage): """Implementation of the ObjStorage API based on the hash of the content. On disk, an object storage is a directory tree containing files named after their object IDs. An object ID is a checksum of its content, depending on the value of the ID_HASH_ALGO constant (see swh.model.hashutil for its meaning). To avoid directories that contain too many files, the object storage has a given slicing. Each slicing correspond to a directory that is named according to the hash of its content. So for instance a file with SHA1 34973274ccef6ab4dfaaf86599792fa9c3fe4689 will be stored in the given object storages : - 0:2/2:4/4:6 : 34/97/32/34973274ccef6ab4dfaaf86599792fa9c3fe4689 - 0:1/0:5/ : 3/34973/34973274ccef6ab4dfaaf86599792fa9c3fe4689 The files in the storage are stored in gzipped compressed format. Attributes: root (string): path to the root directory of the storage on the disk. bounds: list of tuples that indicates the beginning and the end of each subdirectory for a content. """ def __init__(self, root, slicing): """ Create an object to access a hash-slicing based object storage. Args: root (string): path to the root directory of the storage on the disk. slicing (string): string that indicates the slicing to perform on the hash of the content to know the path where it should be stored. """ self.root = root # Make a list of tuples where each tuple contains the beginning # and the end of each slicing. self.bounds = [ slice(*map(int, sbounds.split(':'))) for sbounds in slicing.split('/') if sbounds ] self.check_config(check_write=False) def check_config(self, *, check_write): """Check whether this object storage is properly configured""" root = self.root if not os.path.isdir(root): raise ValueError( 'PathSlicingObjStorage root "%s" is not a directory' % root ) max_endchar = max(map(lambda bound: bound.stop, self.bounds)) if ID_HASH_LENGTH < max_endchar: raise ValueError( 'Algorithm %s has too short hash for slicing to char %d' % (ID_HASH_ALGO, max_endchar) ) if check_write: if not os.access(self.root, os.W_OK): raise PermissionError( 'PathSlicingObjStorage root "%s" is not writable' % root ) return True def __contains__(self, obj_id): hex_obj_id = hashutil.hash_to_hex(obj_id) return os.path.exists(self._obj_path(hex_obj_id)) def __iter__(self): """Iterate over the object identifiers currently available in the storage. Warning: with the current implementation of the object storage, this method will walk the filesystem to list objects, meaning that listing all objects will be very slow for large storages. You almost certainly don't want to use this method in production. Return: Iterator over object IDs """ def obj_iterator(): # XXX hackish: it does not verify that the depth of found files # matches the slicing depth of the storage for root, _dirs, files in os.walk(self.root): for f in files: yield bytes.fromhex(f) return obj_iterator() def __len__(self): """Compute the number of objects available in the storage. Warning: this currently uses `__iter__`, its warning about bad performances applies Return: number of objects contained in the storage """ return sum(1 for i in self) def _obj_dir(self, hex_obj_id): """ Compute the storage directory of an object. See also: PathSlicingObjStorage::_obj_path Args: hex_obj_id: object id as hexlified string. Returns: Path to the directory that contains the required object. """ slices = [hex_obj_id[bound] for bound in self.bounds] return os.path.join(self.root, *slices) def _obj_path(self, hex_obj_id): """ Compute the full path to an object into the current storage. See also: PathSlicingObjStorage::_obj_dir Args: hex_obj_id: object id as hexlified string. Returns: Path to the actual object corresponding to the given id. """ return os.path.join(self._obj_dir(hex_obj_id), hex_obj_id) def add(self, content, obj_id=None, check_presence=True): if obj_id is None: obj_id = compute_hash(content) if check_presence and obj_id in self: # If the object is already present, return immediately. return obj_id hex_obj_id = hashutil.hash_to_hex(obj_id) with _write_obj_file(hex_obj_id, self) as f: f.write(content) return obj_id def get(self, obj_id): if obj_id not in self: raise ObjNotFoundError(obj_id) # Open the file and return its content as bytes hex_obj_id = hashutil.hash_to_hex(obj_id) with _read_obj_file(hex_obj_id, self) as f: return f.read() def check(self, obj_id): if obj_id not in self: raise ObjNotFoundError(obj_id) hex_obj_id = hashutil.hash_to_hex(obj_id) try: with gzip.open(self._obj_path(hex_obj_id)) as f: length = None if ID_HASH_ALGO.endswith('_git'): # if the hashing algorithm is git-like, we need to know the # content size to hash on the fly. Do a first pass here to # compute the size length = 0 while True: chunk = f.read(GZIP_BUFSIZ) length += len(chunk) if not chunk: break f.rewind() checksums = hashutil.hash_file(f, length, algorithms=[ID_HASH_ALGO]) actual_obj_id = checksums[ID_HASH_ALGO] if obj_id != actual_obj_id: raise Error( 'Corrupt object %s should have id %s' % (hashutil.hash_to_hex(obj_id), hashutil.hash_to_hex(actual_obj_id)) ) except (OSError, IOError): # IOError is for compatibility with older python versions raise Error('Corrupt object %s is not a gzip file' % obj_id) + def delete(self, obj_id): + if obj_id not in self: + raise ObjNotFoundError(obj_id) + + hex_obj_id = hashutil.hash_to_hex(obj_id) + try: + os.remove(self._obj_path(hex_obj_id)) + except FileNotFoundError: + raise ObjNotFoundError(obj_id) + return True + # Management methods def get_random(self, batch_size): def get_random_content(self, batch_size): """ Get a batch of content inside a single directory. Returns: a tuple (batch size, batch). """ dirs = [] for level in range(len(self.bounds)): path = os.path.join(self.root, *dirs) dir_list = next(os.walk(path))[1] if 'tmp' in dir_list: dir_list.remove('tmp') dirs.append(random.choice(dir_list)) path = os.path.join(self.root, *dirs) content_list = next(os.walk(path))[2] length = min(batch_size, len(content_list)) return length, map(hashutil.hash_to_bytes, random.sample(content_list, length)) while batch_size: length, it = get_random_content(self, batch_size) batch_size = batch_size - length yield from it # Streaming methods @contextmanager def chunk_writer(self, obj_id): hex_obj_id = hashutil.hash_to_hex(obj_id) with _write_obj_file(hex_obj_id, self) as f: yield f.write def add_stream(self, content_iter, obj_id, check_presence=True): if check_presence and obj_id in self: return obj_id with self.chunk_writer(obj_id) as writer: for chunk in content_iter: writer(chunk) return obj_id def get_stream(self, obj_id, chunk_size=DEFAULT_CHUNK_SIZE): if obj_id not in self: raise ObjNotFoundError(obj_id) hex_obj_id = hashutil.hash_to_hex(obj_id) with _read_obj_file(hex_obj_id, self) as f: reader = functools.partial(f.read, chunk_size) yield from iter(reader, b'')