diff --git a/swh/storage/archiver/worker.py b/swh/storage/archiver/worker.py index e065bfe0..a374c280 100644 --- a/swh/storage/archiver/worker.py +++ b/swh/storage/archiver/worker.py @@ -1,416 +1,415 @@ # 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 import logging import random import time from collections import defaultdict from celery import group from swh.core import hashutil, config, utils from swh.objstorage import get_objstorage from swh.objstorage.exc import Error, ObjNotFoundError from swh.scheduler.utils import get_task from .storage import ArchiverStorage from .copier import ArchiverCopier logger = logging.getLogger('archiver.worker') class BaseArchiveWorker(config.SWHConfig, metaclass=abc.ABCMeta): """Base archive worker. Inherit from this class and override: - ADDITIONAL_CONFIG: Some added configuration needed for the director to work - CONFIG_BASE_FILENAME: relative path to lookup for the configuration file - def need_archival(self, content_data): Determine if a content needs archival or not - def choose_backup_servers(self, present, missing): Choose which backup server to send copies to """ DEFAULT_CONFIG = { 'dbconn': ('str', 'dbname=softwareheritage-archiver-dev'), 'storages': ('list[dict]', [ {'host': 'uffizi', 'cls': 'pathslicing', 'args': {'root': '/tmp/softwareheritage/objects', 'slicing': '0:2/2:4/4:6'}}, {'host': 'banco', 'cls': 'remote', 'args': {'base_url': 'http://banco:5003/'}} ]) } ADDITIONAL_CONFIG = {} CONFIG_BASE_FILENAME = 'archiver/worker' objstorages = {} def __init__(self, batch): super().__init__() self.config = self.parse_config_file( additional_configs=[self.ADDITIONAL_CONFIG]) self.batch = batch self.archiver_db = ArchiverStorage(self.config['dbconn']) self.objstorages = { storage['host']: get_objstorage(storage['cls'], storage['args']) for storage in self.config.get('storages', []) } self.set_objstorages = set(self.objstorages) def run(self): """Do the task expected from the archiver worker. Process the contents in self.batch, ensure that the elements still need an archival (using archiver db), and spawn copiers to copy files in each destination according to the archiver-worker's policy. """ transfers = defaultdict(list) for obj_id in self.batch: # Get dict {'missing': [servers], 'present': [servers]} # for contents ignoring those who don't need archival. copies = self.compute_copies(self.set_objstorages, obj_id) if not copies: # could not happen if using .director module msg = 'Unknown content %s' % hashutil.hash_to_hex(obj_id) logger.warning(msg) continue if not self.need_archival(copies): continue - present = copies.get('present', []) - missing = copies.get('missing', []) + present = copies.get('present', set()) + missing = copies.get('missing', set()) if len(present) == 0: msg = 'Lost content %s' % hashutil.hash_to_hex(obj_id) logger.critical(msg) continue # Choose servers to be used as srcs and dests. for src_dest in self.choose_backup_servers(present, missing): transfers[src_dest].append(obj_id) # Then run copiers for each of the required transfers. contents_copied = [] for (src, dest), content_ids in transfers.items(): contents_copied.extend(self.run_copier(src, dest, content_ids)) # copy is done, eventually do something else with them self.copy_finished(contents_copied) def compute_copies(self, set_objstorages, content_id): """From a content_id, return present and missing copies. Args: objstorages (set): objstorage's id name content_id: the content concerned Returns: - A dictionary with keys 'present' and 'missing' that are - mapped to lists of copies ids depending on whenever the - content is present or missing on the copy. - - There is also the key 'ongoing' which is associated with a - dict that map to a copy name the mtime of the ongoing - status update. - + A dictionary with the following keys: + - 'present': set of archives where the content is present + - 'missing': set of archives where the content is missing + - 'ongoing': ongoing copies: dict mapping the archive id + with the time the copy supposedly started. """ result = self.archiver_db.content_archive_get(content_id) if not result: return None _, present, ongoing = result - set_present = set(present) - set_ongoing = set(ongoing) + set_present = set_objstorages & set(present) + set_ongoing = set_objstorages & set(ongoing) set_missing = set_objstorages - set_present - set_ongoing return { 'present': set_present, 'missing': set_missing, - 'ongoing': ongoing + 'ongoing': {archive: value + for archive, value in ongoing.items() + if archive in set_ongoing}, } def run_copier(self, source, destination, content_ids): """Run a copier in order to archive the given contents. Upload the given contents from the source to the destination. If the process fails, the whole content is considered uncopied and remains 'ongoing', waiting to be rescheduled as there is a delay. Args: source (str): source storage's identifier destination (str): destination storage's identifier content_ids ([sha1]): list of content ids to archive. """ # Check if there are any errors among the contents. content_status = self.get_contents_error(content_ids, source) # Iterates over the error detected. for content_id, real_status in content_status.items(): # Remove them from the to-archive list, # as they cannot be retrieved correctly. content_ids.remove(content_id) # Update their status to reflect their real state. self.archiver_db.content_archive_update( content_id, archive_id=source, new_status=real_status) # Now perform the copy on the remaining contents ac = ArchiverCopier( source=self.objstorages[source], destination=self.objstorages[destination], content_ids=content_ids) if ac.run(): # Once the archival complete, update the database. for content_id in content_ids: self.archiver_db.content_archive_update( content_id, archive_id=destination, new_status='present') return content_ids return [] def copy_finished(self, content_ids): """Hook to notify the content_ids archive copy is finished. (This is not an abstract method as this is optional """ pass def get_contents_error(self, content_ids, source_storage): """Indicates what is the error associated to a content when needed Check the given content on the given storage. If an error is detected, it will be reported through the returned dict. Args: content_ids ([sha1]): list of content ids to check source_storage (str): the source storage holding the contents to check. Returns: a dict that map {content_id -> error_status} for each content_id with an error. The `error_status` result may be 'missing' or 'corrupted'. """ content_status = {} storage = self.objstorages[source_storage] for content_id in content_ids: try: storage.check(content_id) except Error: content_status[content_id] = 'corrupted' logger.error('%s corrupted!' % hashutil.hash_to_hex( content_id)) except ObjNotFoundError: content_status[content_id] = 'missing' logger.error('%s missing!' % hashutil.hash_to_hex(content_id)) return content_status @abc.abstractmethod def need_archival(self, content_data): """Indicate if the content needs to be archived. Args: content_data (dict): dict that contains two lists 'present' and 'missing' with copies id corresponding to this status. Returns: True if there is not enough copies, False otherwise. """ pass @abc.abstractmethod def choose_backup_servers(self, present, missing): """Choose and yield the required amount of couple source/destination For each required copy, choose a unique destination server among the missing copies and a source server among the presents. Args: present: set of objstorage source name where the content is present missing: set of objstorage destination name where the content is missing Yields: tuple (source (str), destination (src)) for each required copy. """ pass class ArchiverWithRetentionPolicyWorker(BaseArchiveWorker): """ Do the required backups on a given batch of contents. Process the content of a content batch in order to do the needed backups on the slaves servers. """ ADDITIONAL_CONFIG = { 'retention_policy': ('int', 2), 'archival_max_age': ('int', 3600), } def __init__(self, batch): """ Constructor of the ArchiverWorker class. Args: batch: list of object's sha1 that potentially need archival. """ super().__init__(batch) config = self.config self.retention_policy = config['retention_policy'] self.archival_max_age = config['archival_max_age'] if len(self.objstorages) < self.retention_policy: raise ValueError('Retention policy is too high for the number of ' 'provided servers') def need_archival(self, content_data): """ Indicate if the content need to be archived. Args: content_data (dict): dict that contains two lists 'present' and 'missing' with copies id corresponding to this status. Returns: True if there is not enough copies, False otherwise. """ nb_presents = len(content_data.get('present', [])) for copy, mtime in content_data.get('ongoing', {}).items(): if not self._is_archival_delay_elapsed(mtime): nb_presents += 1 return nb_presents < self.retention_policy def _is_archival_delay_elapsed(self, start_time): """ Indicates if the archival delay is elapsed given the start_time Args: start_time (float): time at which the archival started. Returns: True if the archival delay is elasped, False otherwise """ elapsed = time.time() - start_time return elapsed > self.archival_max_age def choose_backup_servers(self, present, missing): """Choose and yield the required amount of couple source/destination For each required copy, choose a unique destination server among the missing copies and a source server among the presents. Each destination server is unique so after archival, the retention policy requirement will be fulfilled. However, the source server may be used multiple times. Args: present: set of objstorage source name where the content is present missing: set of objstorage destination name where the content is missing Yields: tuple (source, destination) for each required copy. """ # Transform from set to list to allow random selections missing = list(missing) present = list(present) nb_required = self.retention_policy - len(present) destinations = random.sample(missing, nb_required) sources = [random.choice(present) for dest in destinations] yield from zip(sources, destinations) class ArchiverToBackendWorker(BaseArchiveWorker): """Worker that sends copies over from a source to another backend. Process the content of a content batch from source objstorage to destination objstorage. """ CONFIG_BASE_FILENAME = 'archiver/worker-to-backend' ADDITIONAL_CONFIG = { 'next_task': ( 'dict', { 'queue': 'swh.indexer.tasks.SWHOrchestratorAllContentsTask', 'batch_size': 10, } ) } def __init__(self, destination, batch): """Constructor of the ArchiverWorkerToBackend class. Args: destination: where to copy the objects from batch: sha1s to send to destination """ super().__init__(batch) self.destination = destination next_task = self.config['next_task'] destination_queue = next_task['queue'] self.task_destination = get_task(destination_queue) self.batch_size = int(next_task['batch_size']) def need_archival(self, content_data): """Indicate if the content needs to be archived. Args: content_data (dict): dict that contains 3 lists 'present', 'ongoing' and 'missing' with copies id corresponding to this status. Returns: True if we need to archive, False otherwise """ return self.destination in content_data.get('missing', {}) def choose_backup_servers(self, present, missing): """The destination is fixed to the destination mentioned. The only variable here is the source of information that we choose randomly in 'present'. Args: present: set of objstorage source name where the content is present missing: set of objstorage destination name where the content is missing Yields: tuple (source, destination) for each required copy. """ yield (random.choice(list(present)), self.destination) def copy_finished(self, content_ids): """Once the copy is finished, we'll send those batch of contents as done in the destination queue. """ groups = [] for ids in utils.grouper(content_ids, self.batch_size): sig_ids = self.task_destination.s(list(ids)) groups.append(sig_ids) group(groups).delay() diff --git a/swh/storage/tests/test_archiver.py b/swh/storage/tests/test_archiver.py index 0b75c458..6bb358b8 100644 --- a/swh/storage/tests/test_archiver.py +++ b/swh/storage/tests/test_archiver.py @@ -1,302 +1,312 @@ # 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 tempfile import unittest import os import time import json from nose.tools import istest from nose.plugins.attrib import attr from swh.core import hashutil from swh.core.tests.db_testing import DbsTestFixture from server_testing import ServerTestFixture from swh.storage.archiver import ArchiverWithRetentionPolicyDirector from swh.storage.archiver import ArchiverWithRetentionPolicyWorker from swh.objstorage import get_objstorage from swh.objstorage.exc import ObjNotFoundError from swh.objstorage.api.server import app TEST_DIR = os.path.dirname(os.path.abspath(__file__)) TEST_DATA_DIR = os.path.join(TEST_DIR, '../../../../swh-storage-testdata') @attr('db') class TestArchiver(DbsTestFixture, ServerTestFixture, unittest.TestCase): """ Test the objstorage archiver. """ TEST_DB_NAMES = [ 'softwareheritage-archiver-test', ] TEST_DB_DUMPS = [ os.path.join(TEST_DATA_DIR, 'dumps/swh-archiver.dump'), ] TEST_DB_DUMP_TYPES = [ 'pg_dump', ] def setUp(self): # Launch the backup server dest_root = tempfile.mkdtemp(prefix='remote') self.config = { 'cls': 'pathslicing', 'args': { 'root': dest_root, 'slicing': '0:2/2:4/4:6', } } self.app = app super().setUp() # Retrieve connection (depends on the order in TEST_DB_NAMES) self.conn = self.conns[0] # archiver db's connection self.cursor = self.cursors[0] # Create source storage src_root = tempfile.mkdtemp() src_config = { 'cls': 'pathslicing', 'args': { 'root': src_root, 'slicing': '0:2/2:4/4:6' } } self.src_storage = get_objstorage(**src_config) # Create destination storage dest_config = { 'cls': 'remote', 'args': { 'url': self.url() } } self.dest_storage = get_objstorage(**dest_config) # Keep mapped the id to the storages self.storages = { 'uffizi': self.src_storage, 'banco': self.dest_storage } # Override configurations src_archiver_conf = {'host': 'uffizi'} dest_archiver_conf = {'host': 'banco'} src_archiver_conf.update(src_config) dest_archiver_conf.update(dest_config) self.archiver_storages = [src_archiver_conf, dest_archiver_conf] self._override_director_config() self._override_worker_config() # Create the base archiver self.archiver = self._create_director() def tearDown(self): self.empty_tables() super().tearDown() def empty_tables(self): # Remove all content self.cursor.execute('DELETE FROM content_archive') self.conn.commit() def _override_director_config(self, retention_policy=2): """ Override the default config of the Archiver director to allow the tests to use the *-test db instead of the default one as there is no configuration file for now. """ ArchiverWithRetentionPolicyDirector.parse_config_file = lambda obj, additional_configs: { # noqa 'dbconn': self.conn, 'batch_max_size': 5000, 'archival_max_age': 3600, 'retention_policy': retention_policy, 'asynchronous': False, } def _override_worker_config(self): """ Override the default config of the Archiver worker to allow the tests to use the *-test db instead of the default one as there is no configuration file for now. """ ArchiverWithRetentionPolicyWorker.parse_config_file = lambda obj, additional_configs: { # noqa 'retention_policy': 2, 'archival_max_age': 3600, 'dbconn': self.conn, 'storages': self.archiver_storages, 'source': 'uffizi', } def _create_director(self): return ArchiverWithRetentionPolicyDirector() def _create_worker(self, batch={}): return ArchiverWithRetentionPolicyWorker(batch) def _add_content(self, storage_name, content_data): """ Add really a content to the given objstorage This put an empty status for the added content. Args: storage_name: the concerned storage content_data: the data to insert with_row_insert: to insert a row entry in the db or not """ # Add the content to the storage obj_id = self.storages[storage_name].add(content_data) db_obj_id = r'\x' + hashutil.hash_to_hex(obj_id) self.cursor.execute(""" INSERT INTO content_archive VALUES('%s', '{}') """ % (db_obj_id)) return obj_id def _update_status(self, obj_id, storage_name, status, date=None): """ Update the db status for the given id/storage_name. This does not create the content in the storage. """ db_obj_id = r'\x' + hashutil.hash_to_hex(obj_id) self.archiver.archiver_storage.content_archive_update( db_obj_id, storage_name, status ) def _add_dated_content(self, obj_id, copies={}): """ Fully erase the previous copies field for the given content id This does not alter the contents into the objstorages. """ db_obj_id = r'\x' + hashutil.hash_to_hex(obj_id) self.cursor.execute(""" UPDATE TABLE content_archive SET copies='%s' WHERE content_id='%s' """ % (json.dumps(copies), db_obj_id)) # Integration test @istest def archive_missing_content(self): """ Run archiver on a missing content should archive it. """ obj_data = b'archive_missing_content' obj_id = self._add_content('uffizi', obj_data) self._update_status(obj_id, 'uffizi', 'present') # Content is missing on banco (entry not present in the db) try: self.dest_storage.get(obj_id) except ObjNotFoundError: pass else: self.fail('Content should not be present before archival') self.archiver.run() # now the content should be present on remote objstorage remote_data = self.dest_storage.get(obj_id) self.assertEquals(obj_data, remote_data) @istest def archive_present_content(self): """ A content that is not 'missing' shouldn't be archived. """ obj_id = self._add_content('uffizi', b'archive_present_content') self._update_status(obj_id, 'uffizi', 'present') self._update_status(obj_id, 'banco', 'present') # After the run, the content should NOT be in the archive. # As the archiver believe it was already in. self.archiver.run() with self.assertRaises(ObjNotFoundError): self.dest_storage.get(obj_id) @istest def archive_already_enough(self): """ A content missing with enough copies shouldn't be archived. """ obj_id = self._add_content('uffizi', b'archive_alread_enough') self._update_status(obj_id, 'uffizi', 'present') self._override_director_config(retention_policy=1) director = self._create_director() # Obj is present in only one archive but only one copy is required. director.run() with self.assertRaises(ObjNotFoundError): self.dest_storage.get(obj_id) # Unit tests for archive worker def archival_elapsed(self, mtime): return self._create_worker()._is_archival_delay_elapsed(mtime) @istest def vstatus_ongoing_remaining(self): self.assertFalse(self.archival_elapsed(time.time())) @istest def vstatus_ongoing_elapsed(self): past_time = ( time.time() - self._create_worker().archival_max_age ) self.assertTrue(self.archival_elapsed(past_time)) def _status(self, status, mtime=None): """ Get a dict that match the copies structure """ return {'status': status, 'mtime': mtime or time.time()} @istest def need_archival_missing(self): """ A content should need archival when it is missing. """ status_copies = {'present': ['uffizi'], 'missing': ['banco']} worker = self._create_worker() self.assertEqual(worker.need_archival(status_copies), True) @istest def need_archival_present(self): """ A content present everywhere shouldn't need archival """ status_copies = {'present': ['uffizi', 'banco']} worker = self._create_worker() self.assertEqual(worker.need_archival(status_copies), False) def _compute_copies_status(self, status): """ A content with a given status should be detected correctly """ obj_id = self._add_content( 'banco', b'compute_copies_' + bytes(status, 'utf8')) self._update_status(obj_id, 'banco', status) worker = self._create_worker() self.assertIn('banco', worker.compute_copies( set(worker.objstorages), obj_id)[status]) @istest def compute_copies_present(self): """ A present content should be detected with correct status """ self._compute_copies_status('present') @istest def compute_copies_missing(self): """ A missing content should be detected with correct status """ self._compute_copies_status('missing') + @istest + def compute_copies_extra_archive(self): + obj_id = self._add_content('banco', b'foobar') + self._update_status(obj_id, 'banco', 'present') + self._update_status(obj_id, 'random_archive', 'present') + worker = self._create_worker() + copies = worker.compute_copies(set(worker.objstorages), obj_id) + self.assertEqual(copies['present'], {'banco'}) + self.assertEqual(copies['missing'], {'uffizi'}) + def _get_backups(self, present, missing): """ Return a list of the pair src/dest from the present and missing """ worker = self._create_worker() return list(worker.choose_backup_servers(present, missing)) @istest def choose_backup_servers(self): self.assertEqual(len(self._get_backups(['uffizi', 'banco'], [])), 0) self.assertEqual(len(self._get_backups(['uffizi'], ['banco'])), 1) # Even with more possible destinations, do not take more than the # retention_policy require self.assertEqual( len(self._get_backups(['uffizi'], ['banco', 's3'])), 1 )