class SlavedDeviceInboxStore(DeviceInboxWorkerStore, BaseSlavedStore): def __init__(self, database: Database, db_conn, hs): super(SlavedDeviceInboxStore, self).__init__(database, db_conn, hs) self._device_inbox_id_gen = SlavedIdTracker( db_conn, "device_max_stream_id", "stream_id" ) self._device_inbox_stream_cache = StreamChangeCache( "DeviceInboxStreamChangeCache", self._device_inbox_id_gen.get_current_token(), ) self._device_federation_outbox_stream_cache = StreamChangeCache( "DeviceFederationOutboxStreamChangeCache", self._device_inbox_id_gen.get_current_token(), ) self._last_device_delete_cache = ExpiringCache( cache_name="last_device_delete_cache", clock=self._clock, max_len=10000, expiry_ms=30 * 60 * 1000, ) def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == ToDeviceStream.NAME: self._device_inbox_id_gen.advance(token) for row in rows: if row.entity.startswith("@"): self._device_inbox_stream_cache.entity_has_changed( row.entity, token ) else: self._device_federation_outbox_stream_cache.entity_has_changed( row.entity, token ) return super().process_replication_rows(stream_name, instance_name, token, rows)
class SlavedGroupServerStore(GroupServerWorkerStore, BaseSlavedStore): def __init__(self, database: DatabasePool, db_conn, hs): super(SlavedGroupServerStore, self).__init__(database, db_conn, hs) self.hs = hs self._group_updates_id_gen = SlavedIdTracker(db_conn, "local_group_updates", "stream_id") self._group_updates_stream_cache = StreamChangeCache( "_group_updates_stream_cache", self._group_updates_id_gen.get_current_token(), ) def get_group_stream_token(self): return self._group_updates_id_gen.get_current_token() def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == GroupServerStream.NAME: self._group_updates_id_gen.advance(instance_name, token) for row in rows: self._group_updates_stream_cache.entity_has_changed( row.user_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows)
class SlavedAccountDataStore(TagsWorkerStore, AccountDataWorkerStore, BaseSlavedStore): def __init__(self, database: Database, db_conn, hs): self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data_max_stream_id", "stream_id") super(SlavedAccountDataStore, self).__init__(database, db_conn, hs) def get_max_account_data_stream_id(self): return self._account_data_id_gen.get_current_token() def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == "tag_account_data": self._account_data_id_gen.advance(token) for row in rows: self.get_tags_for_user.invalidate((row.user_id, )) self._account_data_stream_cache.entity_has_changed( row.user_id, token) elif stream_name == "account_data": self._account_data_id_gen.advance(token) for row in rows: if not row.room_id: self.get_global_account_data_by_type_for_user.invalidate( (row.data_type, row.user_id)) self.get_account_data_for_user.invalidate((row.user_id, )) self.get_account_data_for_room.invalidate( (row.user_id, row.room_id)) self.get_account_data_for_room_and_type.invalidate( (row.user_id, row.room_id, row.data_type)) self._account_data_stream_cache.entity_has_changed( row.user_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows)
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", extra_tables=[ ("user_signature_stream", "stream_id"), ("device_lists_outbound_pokes", "stream_id"), ], ) device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max) self._user_signature_stream_cache = StreamChangeCache( "UserSignatureStreamChangeCache", device_list_max) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max)
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): self._instance_name = hs.get_instance_name() self._receipts_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): self._can_write_to_receipts = ( self._instance_name in hs.config.worker.writers.receipts ) self._receipts_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="receipts", instance_name=self._instance_name, tables=[("receipts_linearized", "instance_name", "stream_id")], sequence_name="receipts_sequence", writers=hs.config.worker.writers.receipts, ) else: self._can_write_to_receipts = True # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if hs.get_instance_name() in hs.config.worker.writers.receipts: self._receipts_id_gen = StreamIdGenerator( db_conn, "receipts_linearized", "stream_id" ) else: self._receipts_id_gen = SlavedIdTracker( db_conn, "receipts_linearized", "stream_id" ) super().__init__(database, db_conn, hs) max_receipts_stream_id = self.get_max_receipt_stream_id() receipts_stream_prefill, min_receipts_stream_id = self.db_pool.get_cache_dict( db_conn, "receipts_linearized", entity_column="room_id", stream_column="stream_id", max_value=max_receipts_stream_id, limit=10000, ) self._receipts_stream_cache = StreamChangeCache( "ReceiptsRoomChangeCache", min_receipts_stream_id, prefilled_cache=receipts_stream_prefill, )
class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedStore): def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", extra_tables=[ ("user_signature_stream", "stream_id"), ("device_lists_outbound_pokes", "stream_id"), ("device_lists_changes_in_room", "stream_id"), ], ) super().__init__(database, db_conn, hs) def get_device_stream_token(self) -> int: return self._device_list_id_gen.get_current_token() def process_replication_rows(self, stream_name: str, instance_name: str, token: int, rows: Iterable[Any]) -> None: if stream_name == DeviceListsStream.NAME: self._device_list_id_gen.advance(instance_name, token) self._invalidate_caches_for_devices(token, rows) elif stream_name == UserSignatureStream.NAME: self._device_list_id_gen.advance(instance_name, token) for row in rows: self._user_signature_stream_cache.entity_has_changed( row.user_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows) def _invalidate_caches_for_devices( self, token: int, rows: Iterable[DeviceListsStream.DeviceListsStreamRow]) -> None: for row in rows: # The entities are either user IDs (starting with '@') whose devices # have changed, or remote servers that we need to tell about # changes. if row.entity.startswith("@"): self._device_list_stream_cache.entity_has_changed( row.entity, token) self.get_cached_devices_for_user.invalidate((row.entity, )) self._get_cached_user_device.invalidate((row.entity, )) self.get_device_list_last_stream_id_for_remote.invalidate( (row.entity, )) else: self._device_list_federation_stream_cache.entity_has_changed( row.entity, token)
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) # `_can_write_to_account_data` indicates whether the current worker is allowed # to write account data. A value of `True` implies that `_account_data_id_gen` # is an `AbstractStreamIdGenerator` and not just a tracker. self._account_data_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): self._can_write_to_account_data = ( self._instance_name in hs.config.worker.writers.account_data) self._account_data_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="account_data", instance_name=self._instance_name, tables=[ ("room_account_data", "instance_name", "stream_id"), ("room_tags_revisions", "instance_name", "stream_id"), ("account_data", "instance_name", "stream_id"), ], sequence_name="account_data_sequence", writers=hs.config.worker.writers.account_data, ) else: # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if self._instance_name in hs.config.worker.writers.account_data: self._can_write_to_account_data = True self._account_data_id_gen = StreamIdGenerator( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) else: self._account_data_id_gen = SlavedIdTracker( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( "AccountDataAndTagsChangeCache", account_max)
def __init__(self, database: DatabasePool, db_conn, hs): self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data", "stream_id", extra_tables=[ ("room_account_data", "stream_id"), ("room_tags_revisions", "stream_id"), ], ) super(SlavedAccountDataStore, self).__init__(database, db_conn, hs)
class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedStore): def __init__(self, database: Database, db_conn, hs): super(SlavedDeviceStore, self).__init__(database, db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", extra_tables=[ ("user_signature_stream", "stream_id"), ("device_lists_outbound_pokes", "stream_id"), ], ) device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max) self._user_signature_stream_cache = StreamChangeCache( "UserSignatureStreamChangeCache", device_list_max) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max) def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == DeviceListsStream.NAME: self._device_list_id_gen.advance(token) self._invalidate_caches_for_devices(token, rows) elif stream_name == UserSignatureStream.NAME: self._device_list_id_gen.advance(token) for row in rows: self._user_signature_stream_cache.entity_has_changed( row.user_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows) def _invalidate_caches_for_devices(self, token, rows): for row in rows: # The entities are either user IDs (starting with '@') whose devices # have changed, or remote servers that we need to tell about # changes. if row.entity.startswith("@"): self._device_list_stream_cache.entity_has_changed( row.entity, token) self.get_cached_devices_for_user.invalidate((row.entity, )) self._get_cached_user_device.invalidate_many((row.entity, )) self.get_device_list_last_stream_id_for_remote.invalidate( (row.entity, )) else: self._device_list_federation_stream_cache.entity_has_changed( row.entity, token)
def __init__(self, database: Database, db_conn, hs): super(SlavedGroupServerStore, self).__init__(database, db_conn, hs) self.hs = hs self._group_updates_id_gen = SlavedIdTracker( db_conn, "local_group_updates", "stream_id" ) self._group_updates_stream_cache = StreamChangeCache( "_group_updates_stream_cache", self._group_updates_id_gen.get_current_token(), )
def __init__(self, database: DatabasePool, db_conn, hs): self._instance_name = hs.get_instance_name() if isinstance(database.engine, PostgresEngine): self._can_write_to_account_data = ( self._instance_name in hs.config.worker.writers.account_data ) self._account_data_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="account_data", instance_name=self._instance_name, tables=[ ("room_account_data", "instance_name", "stream_id"), ("room_tags_revisions", "instance_name", "stream_id"), ("account_data", "instance_name", "stream_id"), ], sequence_name="account_data_sequence", writers=hs.config.worker.writers.account_data, ) else: self._can_write_to_account_data = True # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if hs.get_instance_name() in hs.config.worker.writers.account_data: self._account_data_id_gen = StreamIdGenerator( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) else: self._account_data_id_gen = SlavedIdTracker( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( "AccountDataAndTagsChangeCache", account_max ) super().__init__(database, db_conn, hs)
def __init__(self, db_conn, hs): super(SlavedDeviceStore, self).__init__(db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker(db_conn, "device_lists_stream", "stream_id") device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max)
class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedDeviceStore, self).__init__(db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker(db_conn, "device_lists_stream", "stream_id") device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max) self._user_signature_stream_cache = StreamChangeCache( "UserSignatureStreamChangeCache", device_list_max) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max) def stream_positions(self): result = super(SlavedDeviceStore, self).stream_positions() # The user signature stream uses the same stream ID generator as the # device list stream, so set them both to the device list ID # generator's current token. current_token = self._device_list_id_gen.get_current_token() result[DeviceListsStream.NAME] = current_token result[UserSignatureStream.NAME] = current_token return result def process_replication_rows(self, stream_name, token, rows): if stream_name == DeviceListsStream.NAME: self._device_list_id_gen.advance(token) for row in rows: self._invalidate_caches_for_devices(token, row.user_id, row.destination) elif stream_name == UserSignatureStream.NAME: for row in rows: self._user_signature_stream_cache.entity_has_changed( row.user_id, token) return super(SlavedDeviceStore, self).process_replication_rows(stream_name, token, rows) def _invalidate_caches_for_devices(self, token, user_id, destination): self._device_list_stream_cache.entity_has_changed(user_id, token) if destination: self._device_list_federation_stream_cache.entity_has_changed( destination, token) self._get_cached_devices_for_user.invalidate((user_id, )) self._get_cached_user_device.invalidate_many((user_id, )) self.get_device_list_last_stream_id_for_remote.invalidate((user_id, ))
class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedDeviceStore, self).__init__(db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", ) device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max, ) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max, ) def stream_positions(self): result = super(SlavedDeviceStore, self).stream_positions() result["device_lists"] = self._device_list_id_gen.get_current_token() return result def process_replication_rows(self, stream_name, token, rows): if stream_name == "device_lists": self._device_list_id_gen.advance(token) for row in rows: self._invalidate_caches_for_devices( token, row.user_id, row.destination, ) return super(SlavedDeviceStore, self).process_replication_rows(stream_name, token, rows) def _invalidate_caches_for_devices(self, token, user_id, destination): self._device_list_stream_cache.entity_has_changed(user_id, token) if destination: self._device_list_federation_stream_cache.entity_has_changed( destination, token) self._get_cached_devices_for_user.invalidate((user_id, )) self._get_cached_user_device.invalidate_many((user_id, )) self.get_device_list_last_stream_id_for_remote.invalidate((user_id, ))
def __init__(self, database: DatabasePool, db_conn, hs): super().__init__(database, db_conn, hs) if hs.config.worker.worker_app is None: self._push_rules_stream_id_gen = StreamIdGenerator( db_conn, "push_rules_stream", "stream_id") # type: Union[StreamIdGenerator, SlavedIdTracker] else: self._push_rules_stream_id_gen = SlavedIdTracker( db_conn, "push_rules_stream", "stream_id") push_rules_prefill, push_rules_id = self.db_pool.get_cache_dict( db_conn, "push_rules_stream", entity_column="user_id", stream_column="stream_id", max_value=self.get_max_push_rules_stream_id(), ) self.push_rules_stream_cache = StreamChangeCache( "PushRulesStreamChangeCache", push_rules_id, prefilled_cache=push_rules_prefill, ) self._users_new_default_push_rules = hs.config.users_new_default_push_rules
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) if hs.config.worker.worker_app is None: self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator( db_conn, "push_rules_stream", "stream_id") else: self._push_rules_stream_id_gen = SlavedIdTracker( db_conn, "push_rules_stream", "stream_id") push_rules_prefill, push_rules_id = self.db_pool.get_cache_dict( db_conn, "push_rules_stream", entity_column="user_id", stream_column="stream_id", max_value=self.get_max_push_rules_stream_id(), ) self.push_rules_stream_cache = StreamChangeCache( "PushRulesStreamChangeCache", push_rules_id, prefilled_cache=push_rules_prefill, )
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) self.hs = hs self._group_updates_id_gen = SlavedIdTracker(db_conn, "local_group_updates", "stream_id") self._group_updates_stream_cache = StreamChangeCache( "_group_updates_stream_cache", self._group_updates_id_gen.get_current_token(), )
class SlavedDeviceStore(EndToEndKeyWorkerStore, DeviceWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedDeviceStore, self).__init__(db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", ) device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max, ) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max, ) def stream_positions(self): result = super(SlavedDeviceStore, self).stream_positions() result["device_lists"] = self._device_list_id_gen.get_current_token() return result def process_replication_rows(self, stream_name, token, rows): if stream_name == "device_lists": self._device_list_id_gen.advance(token) for row in rows: self._invalidate_caches_for_devices( token, row.user_id, row.destination, ) return super(SlavedDeviceStore, self).process_replication_rows( stream_name, token, rows ) def _invalidate_caches_for_devices(self, token, user_id, destination): self._device_list_stream_cache.entity_has_changed( user_id, token ) if destination: self._device_list_federation_stream_cache.entity_has_changed( destination, token ) self._get_cached_devices_for_user.invalidate((user_id,)) self._get_cached_user_device.invalidate_many((user_id,)) self.get_device_list_last_stream_id_for_remote.invalidate((user_id,))
def __init__(self, database: DatabasePool, db_conn, hs): super(SlavedDeviceInboxStore, self).__init__(database, db_conn, hs) self._device_inbox_id_gen = SlavedIdTracker(db_conn, "device_inbox", "stream_id") self._device_inbox_stream_cache = StreamChangeCache( "DeviceInboxStreamChangeCache", self._device_inbox_id_gen.get_current_token(), ) self._device_federation_outbox_stream_cache = StreamChangeCache( "DeviceFederationOutboxStreamChangeCache", self._device_inbox_id_gen.get_current_token(), ) self._last_device_delete_cache = ExpiringCache( cache_name="last_device_delete_cache", clock=self._clock, max_len=10000, expiry_ms=30 * 60 * 1000, )
def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", extra_tables=[ ("user_signature_stream", "stream_id"), ("device_lists_outbound_pokes", "stream_id"), ("device_lists_changes_in_room", "stream_id"), ], ) super().__init__(database, db_conn, hs)
def __init__(self, db_conn, hs): super(SlavedDeviceStore, self).__init__(db_conn, hs) self.hs = hs self._device_list_id_gen = SlavedIdTracker( db_conn, "device_lists_stream", "stream_id", ) device_list_max = self._device_list_id_gen.get_current_token() self._device_list_stream_cache = StreamChangeCache( "DeviceListStreamChangeCache", device_list_max, ) self._device_list_federation_stream_cache = StreamChangeCache( "DeviceListFederationStreamChangeCache", device_list_max, )
class SlavedAccountDataStore(TagsWorkerStore, AccountDataWorkerStore, BaseSlavedStore): def __init__(self, db_conn, hs): self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data_max_stream_id", "stream_id", ) super(SlavedAccountDataStore, self).__init__(db_conn, hs) def get_max_account_data_stream_id(self): return self._account_data_id_gen.get_current_token() def stream_positions(self): result = super(SlavedAccountDataStore, self).stream_positions() position = self._account_data_id_gen.get_current_token() result["user_account_data"] = position result["room_account_data"] = position result["tag_account_data"] = position return result def process_replication_rows(self, stream_name, token, rows): if stream_name == "tag_account_data": self._account_data_id_gen.advance(token) for row in rows: self.get_tags_for_user.invalidate((row.user_id,)) self._account_data_stream_cache.entity_has_changed( row.user_id, token ) elif stream_name == "account_data": self._account_data_id_gen.advance(token) for row in rows: if not row.room_id: self.get_global_account_data_by_type_for_user.invalidate( (row.data_type, row.user_id,) ) self.get_account_data_for_user.invalidate((row.user_id,)) self.get_account_data_for_room.invalidate((row.user_id, row.room_id,)) self.get_account_data_for_room_and_type.invalidate( (row.user_id, row.room_id, row.data_type,), ) self._account_data_stream_cache.entity_has_changed( row.user_id, token ) return super(SlavedAccountDataStore, self).process_replication_rows( stream_name, token, rows )
def __init__(self, database: DatabasePool, db_conn, hs): super(EventsWorkerStore, self).__init__(database, db_conn, hs) if hs.config.worker.writers.events == hs.get_instance_name(): # We are the process in charge of generating stream ids for events, # so instantiate ID generators based on the database self._stream_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", ) self._backfill_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", step=-1, extra_tables=[("ex_outlier_stream", "event_stream_ordering")], ) else: # Another process is in charge of persisting events and generating # stream IDs: rely on the replication streams to let us know which # IDs we can process. self._stream_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering") self._backfill_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering", step=-1) self._get_event_cache = Cache( "*getEvent*", keylen=3, max_entries=hs.config.caches.event_cache_size, apply_cache_factor_from_config=False, ) self._event_fetch_lock = threading.Condition() self._event_fetch_list = [] self._event_fetch_ongoing = 0
def __init__(self, database: Database, db_conn, hs): self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data_max_stream_id", "stream_id") super(SlavedAccountDataStore, self).__init__(database, db_conn, hs)
class EventsWorkerStore(SQLBaseStore): def __init__(self, database: DatabasePool, db_conn, hs): super(EventsWorkerStore, self).__init__(database, db_conn, hs) if hs.config.worker.writers.events == hs.get_instance_name(): # We are the process in charge of generating stream ids for events, # so instantiate ID generators based on the database self._stream_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", ) self._backfill_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", step=-1, extra_tables=[("ex_outlier_stream", "event_stream_ordering")], ) else: # Another process is in charge of persisting events and generating # stream IDs: rely on the replication streams to let us know which # IDs we can process. self._stream_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering") self._backfill_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering", step=-1) self._get_event_cache = Cache( "*getEvent*", keylen=3, max_entries=hs.config.caches.event_cache_size, apply_cache_factor_from_config=False, ) self._event_fetch_lock = threading.Condition() self._event_fetch_list = [] self._event_fetch_ongoing = 0 def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == EventsStream.NAME: self._stream_id_gen.advance(instance_name, token) elif stream_name == BackfillStream.NAME: self._backfill_id_gen.advance(instance_name, -token) super().process_replication_rows(stream_name, instance_name, token, rows) async def get_received_ts(self, event_id: str) -> Optional[int]: """Get received_ts (when it was persisted) for the event. Raises an exception for unknown events. Args: event_id: The event ID to query. Returns: Timestamp in milliseconds, or None for events that were persisted before received_ts was implemented. """ return await self.db_pool.simple_select_one_onecol( table="events", keyvalues={"event_id": event_id}, retcol="received_ts", desc="get_received_ts", ) # Inform mypy that if allow_none is False (the default) then get_event # always returns an EventBase. @overload async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: Literal[False] = False, check_room_id: Optional[str] = None, ) -> EventBase: ... @overload async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: Literal[True] = False, check_room_id: Optional[str] = None, ) -> Optional[EventBase]: ... async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: bool = False, check_room_id: Optional[str] = None, ) -> Optional[EventBase]: """Get an event from the database by event_id. Args: event_id: The event_id of the event to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (behave as per allow_none if the event is redacted) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, behave as per allow_none. allow_none: If True, return None if no event found, if False throw a NotFoundError check_room_id: if not None, check the room of the found event. If there is a mismatch, behave as per allow_none. Returns: The event, or None if the event was not found. """ if not isinstance(event_id, str): raise TypeError("Invalid event event_id %r" % (event_id, )) events = await self.get_events_as_list( [event_id], redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) event = events[0] if events else None if event is not None and check_room_id is not None: if event.room_id != check_room_id: event = None if event is None and not allow_none: raise NotFoundError("Could not find event %s" % (event_id, )) return event async def get_events( self, event_ids: Iterable[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ) -> Dict[str, EventBase]: """Get events from the database Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejeted events from the response. Returns: A mapping from event_id to event. """ events = await self.get_events_as_list( event_ids, redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) return {e.event_id: e for e in events} async def get_events_as_list( self, event_ids: Collection[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ) -> List[EventBase]: """Get events from the database and return in a list in the same order as given by `event_ids` arg. Unknown events will be omitted from the response. Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejected events from the response. Returns: List of events fetched from the database. The events are in the same order as `event_ids` arg. Note that the returned list may be smaller than the list of event IDs if not all events could be fetched. """ if not event_ids: return [] # there may be duplicates so we cast the list to a set event_entry_map = await self._get_events_from_cache_or_db( set(event_ids), allow_rejected=allow_rejected) events = [] for event_id in event_ids: entry = event_entry_map.get(event_id, None) if not entry: continue if not allow_rejected: assert not entry.event.rejected_reason, ( "rejected event returned from _get_events_from_cache_or_db despite " "allow_rejected=False") # We may not have had the original event when we received a redaction, so # we have to recheck auth now. if not allow_rejected and entry.event.type == EventTypes.Redaction: if entry.event.redacts is None: # A redacted redaction doesn't have a `redacts` key, in # which case lets just withhold the event. # # Note: Most of the time if the redactions has been # redacted we still have the un-redacted event in the DB # and so we'll still see the `redacts` key. However, this # isn't always true e.g. if we have censored the event. logger.debug( "Withholding redaction event %s as we don't have redacts key", event_id, ) continue redacted_event_id = entry.event.redacts event_map = await self._get_events_from_cache_or_db( [redacted_event_id]) original_event_entry = event_map.get(redacted_event_id) if not original_event_entry: # we don't have the redacted event (or it was rejected). # # We assume that the redaction isn't authorized for now; if the # redacted event later turns up, the redaction will be re-checked, # and if it is found valid, the original will get redacted before it # is served to the client. logger.debug( "Withholding redaction event %s since we don't (yet) have the " "original %s", event_id, redacted_event_id, ) continue original_event = original_event_entry.event if original_event.type == EventTypes.Create: # we never serve redactions of Creates to clients. logger.info( "Withholding redaction %s of create event %s", event_id, redacted_event_id, ) continue if original_event.room_id != entry.event.room_id: logger.info( "Withholding redaction %s of event %s from a different room", event_id, redacted_event_id, ) continue if entry.event.internal_metadata.need_to_check_redaction(): original_domain = get_domain_from_id(original_event.sender) redaction_domain = get_domain_from_id(entry.event.sender) if original_domain != redaction_domain: # the senders don't match, so this is forbidden logger.info( "Withholding redaction %s whose sender domain %s doesn't " "match that of redacted event %s %s", event_id, redaction_domain, redacted_event_id, original_domain, ) continue # Update the cache to save doing the checks again. entry.event.internal_metadata.recheck_redaction = False event = entry.event if entry.redacted_event: if redact_behaviour == EventRedactBehaviour.BLOCK: # Skip this event continue elif redact_behaviour == EventRedactBehaviour.REDACT: event = entry.redacted_event events.append(event) if get_prev_content: if "replaces_state" in event.unsigned: prev = await self.get_event( event.unsigned["replaces_state"], get_prev_content=False, allow_none=True, ) if prev: event.unsigned = dict(event.unsigned) event.unsigned["prev_content"] = prev.content event.unsigned["prev_sender"] = prev.sender return events async def _get_events_from_cache_or_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the cache or the database. If events are pulled from the database, they will be cached for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Dict[str, _EventCacheEntry]: map from event id to result """ event_entry_map = self._get_events_from_cache( event_ids, allow_rejected=allow_rejected) missing_events_ids = [e for e in event_ids if e not in event_entry_map] if missing_events_ids: log_ctx = current_context() log_ctx.record_event_fetch(len(missing_events_ids)) # Note that _get_events_from_db is also responsible for turning db rows # into FrozenEvents (via _get_event_from_row), which involves seeing if # the events have been redacted, and if so pulling the redaction event out # of the database to check it. # missing_events = await self._get_events_from_db( missing_events_ids, allow_rejected=allow_rejected) event_entry_map.update(missing_events) return event_entry_map def _invalidate_get_event_cache(self, event_id): self._get_event_cache.invalidate((event_id, )) def _get_events_from_cache(self, events, allow_rejected, update_metrics=True): """Fetch events from the caches Args: events (Iterable[str]): list of event_ids to fetch allow_rejected (bool): Whether to return events that were rejected update_metrics (bool): Whether to update the cache hit ratio metrics Returns: dict of event_id -> _EventCacheEntry for each event_id in cache. If allow_rejected is `False` then there will still be an entry but it will be `None` """ event_map = {} for event_id in events: ret = self._get_event_cache.get((event_id, ), None, update_metrics=update_metrics) if not ret: continue if allow_rejected or not ret.event.rejected_reason: event_map[event_id] = ret else: event_map[event_id] = None return event_map def _do_fetch(self, conn): """Takes a database connection and waits for requests for events from the _event_fetch_list queue. """ i = 0 while True: with self._event_fetch_lock: event_list = self._event_fetch_list self._event_fetch_list = [] if not event_list: single_threaded = self.database_engine.single_threaded if single_threaded or i > EVENT_QUEUE_ITERATIONS: self._event_fetch_ongoing -= 1 return else: self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S) i += 1 continue i = 0 self._fetch_event_list(conn, event_list) def _fetch_event_list(self, conn, event_list): """Handle a load of requests from the _event_fetch_list queue Args: conn (twisted.enterprise.adbapi.Connection): database connection event_list (list[Tuple[list[str], Deferred]]): The fetch requests. Each entry consists of a list of event ids to be fetched, and a deferred to be completed once the events have been fetched. The deferreds are callbacked with a dictionary mapping from event id to event row. Note that it may well contain additional events that were not part of this request. """ with Measure(self._clock, "_fetch_event_list"): try: events_to_fetch = { event_id for events, _ in event_list for event_id in events } row_dict = self.db_pool.new_transaction( conn, "do_fetch", [], [], self._fetch_event_rows, events_to_fetch) # We only want to resolve deferreds from the main thread def fire(): for _, d in event_list: d.callback(row_dict) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire) except Exception as e: logger.exception("do_fetch") # We only want to resolve deferreds from the main thread def fire(evs, exc): for _, d in evs: if not d.called: with PreserveLoggingContext(): d.errback(exc) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire, event_list, e) async def _get_events_from_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the database. Returned events will be added to the cache for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Dict[str, _EventCacheEntry]: map from event id to result. May return extra events which weren't asked for. """ fetched_events = {} events_to_fetch = event_ids while events_to_fetch: row_map = await self._enqueue_events(events_to_fetch) # we need to recursively fetch any redactions of those events redaction_ids = set() for event_id in events_to_fetch: row = row_map.get(event_id) fetched_events[event_id] = row if row: redaction_ids.update(row["redactions"]) events_to_fetch = redaction_ids.difference(fetched_events.keys()) if events_to_fetch: logger.debug("Also fetching redaction events %s", events_to_fetch) # build a map from event_id to EventBase event_map = {} for event_id, row in fetched_events.items(): if not row: continue assert row["event_id"] == event_id rejected_reason = row["rejected_reason"] if not allow_rejected and rejected_reason: continue # If the event or metadata cannot be parsed, log the error and act # as if the event is unknown. try: d = db_to_json(row["json"]) except ValueError: logger.error("Unable to parse json from event: %s", event_id) continue try: internal_metadata = db_to_json(row["internal_metadata"]) except ValueError: logger.error( "Unable to parse internal_metadata from event: %s", event_id) continue format_version = row["format_version"] if format_version is None: # This means that we stored the event before we had the concept # of a event format version, so it must be a V1 event. format_version = EventFormatVersions.V1 room_version_id = row["room_version_id"] if not room_version_id: # this should only happen for out-of-band membership events which # arrived before #6983 landed. For all other events, we should have # an entry in the 'rooms' table. # # However, the 'out_of_band_membership' flag is unreliable for older # invites, so just accept it for all membership events. # if d["type"] != EventTypes.Member: raise Exception("Room %s for event %s is unknown" % (d["room_id"], event_id)) # so, assuming this is an out-of-band-invite that arrived before #6983 # landed, we know that the room version must be v5 or earlier (because # v6 hadn't been invented at that point, so invites from such rooms # would have been rejected.) # # The main reason we need to know the room version here (other than # choosing the right python Event class) is in case the event later has # to be redacted - and all the room versions up to v5 used the same # redaction algorithm. # # So, the following approximations should be adequate. if format_version == EventFormatVersions.V1: # if it's event format v1 then it must be room v1 or v2 room_version = RoomVersions.V1 elif format_version == EventFormatVersions.V2: # if it's event format v2 then it must be room v3 room_version = RoomVersions.V3 else: # if it's event format v3 then it must be room v4 or v5 room_version = RoomVersions.V5 else: room_version = KNOWN_ROOM_VERSIONS.get(room_version_id) if not room_version: logger.warning( "Event %s in room %s has unknown room version %s", event_id, d["room_id"], room_version_id, ) continue if room_version.event_format != format_version: logger.error( "Event %s in room %s with version %s has wrong format: " "expected %s, was %s", event_id, d["room_id"], room_version_id, room_version.event_format, format_version, ) continue original_ev = make_event_from_dict( event_dict=d, room_version=room_version, internal_metadata_dict=internal_metadata, rejected_reason=rejected_reason, ) event_map[event_id] = original_ev # finally, we can decide whether each one needs redacting, and build # the cache entries. result_map = {} for event_id, original_ev in event_map.items(): redactions = fetched_events[event_id]["redactions"] redacted_event = self._maybe_redact_event_row( original_ev, redactions, event_map) cache_entry = _EventCacheEntry(event=original_ev, redacted_event=redacted_event) self._get_event_cache.prefill((event_id, ), cache_entry) result_map[event_id] = cache_entry return result_map async def _enqueue_events(self, events): """Fetches events from the database using the _event_fetch_list. This allows batch and bulk fetching of events - it allows us to fetch events without having to create a new transaction for each request for events. Args: events (Iterable[str]): events to be fetched. Returns: Dict[str, Dict]: map from event id to row data from the database. May contain events that weren't requested. """ events_d = defer.Deferred() with self._event_fetch_lock: self._event_fetch_list.append((events, events_d)) self._event_fetch_lock.notify() if self._event_fetch_ongoing < EVENT_QUEUE_THREADS: self._event_fetch_ongoing += 1 should_start = True else: should_start = False if should_start: run_as_background_process("fetch_events", self.db_pool.runWithConnection, self._do_fetch) logger.debug("Loading %d events: %s", len(events), events) with PreserveLoggingContext(): row_map = await events_d logger.debug("Loaded %d events (%d rows)", len(events), len(row_map)) return row_map def _fetch_event_rows(self, txn, event_ids): """Fetch event rows from the database Events which are not found are omitted from the result. The returned per-event dicts contain the following keys: * event_id (str) * json (str): json-encoded event structure * internal_metadata (str): json-encoded internal metadata dict * format_version (int|None): The format of the event. Hopefully one of EventFormatVersions. 'None' means the event predates EventFormatVersions (so the event is format V1). * room_version_id (str|None): The version of the room which contains the event. Hopefully one of RoomVersions. Due to historical reasons, there may be a few events in the database which do not have an associated room; in this case None will be returned here. * rejected_reason (str|None): if the event was rejected, the reason why. * redactions (List[str]): a list of event-ids which (claim to) redact this event. Args: txn (twisted.enterprise.adbapi.Connection): event_ids (Iterable[str]): event IDs to fetch Returns: Dict[str, Dict]: a map from event id to event info. """ event_dict = {} for evs in batch_iter(event_ids, 200): sql = """\ SELECT e.event_id, e.internal_metadata, e.json, e.format_version, r.room_version, rej.reason FROM event_json as e LEFT JOIN rooms r USING (room_id) LEFT JOIN rejections as rej USING (event_id) WHERE """ clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", evs) txn.execute(sql + clause, args) for row in txn: event_id = row[0] event_dict[event_id] = { "event_id": event_id, "internal_metadata": row[1], "json": row[2], "format_version": row[3], "room_version_id": row[4], "rejected_reason": row[5], "redactions": [], } # check for redactions redactions_sql = "SELECT event_id, redacts FROM redactions WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "redacts", evs) txn.execute(redactions_sql + clause, args) for (redacter, redacted) in txn: d = event_dict.get(redacted) if d: d["redactions"].append(redacter) return event_dict def _maybe_redact_event_row( self, original_ev: EventBase, redactions: Iterable[str], event_map: Dict[str, EventBase], ) -> Optional[EventBase]: """Given an event object and a list of possible redacting event ids, determine whether to honour any of those redactions and if so return a redacted event. Args: original_ev: The original event. redactions: list of event ids of potential redaction events event_map: other events which have been fetched, in which we can look up the redaaction events. Map from event id to event. Returns: If the event should be redacted, a pruned event object. Otherwise, None. """ if original_ev.type == "m.room.create": # we choose to ignore redactions of m.room.create events. return None for redaction_id in redactions: redaction_event = event_map.get(redaction_id) if not redaction_event or redaction_event.rejected_reason: # we don't have the redaction event, or the redaction event was not # authorized. logger.debug( "%s was redacted by %s but redaction not found/authed", original_ev.event_id, redaction_id, ) continue if redaction_event.room_id != original_ev.room_id: logger.debug( "%s was redacted by %s but redaction was in a different room!", original_ev.event_id, redaction_id, ) continue # Starting in room version v3, some redactions need to be # rechecked if we didn't have the redacted event at the # time, so we recheck on read instead. if redaction_event.internal_metadata.need_to_check_redaction(): expected_domain = get_domain_from_id(original_ev.sender) if get_domain_from_id( redaction_event.sender) == expected_domain: # This redaction event is allowed. Mark as not needing a recheck. redaction_event.internal_metadata.recheck_redaction = False else: # Senders don't match, so the event isn't actually redacted logger.debug( "%s was redacted by %s but the senders don't match", original_ev.event_id, redaction_id, ) continue logger.debug("Redacting %s due to %s", original_ev.event_id, redaction_id) # we found a good redaction event. Redact! redacted_event = prune_event(original_ev) redacted_event.unsigned["redacted_by"] = redaction_id # It's fine to add the event directly, since get_pdu_json # will serialise this field correctly redacted_event.unsigned["redacted_because"] = redaction_event return redacted_event # no valid redaction found for this event return None async def have_events_in_timeline(self, event_ids): """Given a list of event ids, check if we have already processed and stored them as non outliers. """ rows = await self.db_pool.simple_select_many_batch( table="events", retcols=("event_id", ), column="event_id", iterable=list(event_ids), keyvalues={"outlier": False}, desc="have_events_in_timeline", ) return {r["event_id"] for r in rows} async def have_seen_events(self, event_ids): """Given a list of event ids, check if we have already processed them. Args: event_ids (iterable[str]): Returns: set[str]: The events we have already seen. """ results = set() def have_seen_events_txn(txn, chunk): sql = "SELECT event_id FROM events as e WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", chunk) txn.execute(sql + clause, args) for (event_id, ) in txn: results.add(event_id) # break the input up into chunks of 100 input_iterator = iter(event_ids) for chunk in iter(lambda: list(itertools.islice(input_iterator, 100)), []): await self.db_pool.runInteraction("have_seen_events", have_seen_events_txn, chunk) return results def _get_current_state_event_counts_txn(self, txn, room_id): """ See get_current_state_event_counts. """ sql = "SELECT COUNT(*) FROM current_state_events WHERE room_id=?" txn.execute(sql, (room_id, )) row = txn.fetchone() return row[0] if row else 0 async def get_current_state_event_counts(self, room_id: str) -> int: """ Gets the current number of state events in a room. Args: room_id: The room ID to query. Returns: The current number of state events. """ return await self.db_pool.runInteraction( "get_current_state_event_counts", self._get_current_state_event_counts_txn, room_id, ) async def get_room_complexity(self, room_id): """ Get a rough approximation of the complexity of the room. This is used by remote servers to decide whether they wish to join the room or not. Higher complexity value indicates that being in the room will consume more resources. Args: room_id (str) Returns: dict[str:int] of complexity version to complexity. """ state_events = await self.get_current_state_event_counts(room_id) # Call this one "v1", so we can introduce new ones as we want to develop # it. complexity_v1 = round(state_events / 500, 2) return {"v1": complexity_v1} def get_current_backfill_token(self): """The current minimum token that backfilled events have reached""" return -self._backfill_id_gen.get_current_token() def get_current_events_token(self): """The current maximum token that events have reached""" return self._stream_id_gen.get_current_token() async def get_all_new_forward_event_rows(self, last_id: int, current_id: int, limit: int) -> List[Tuple]: """Returns new events, for the Events replication stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: a list of events stream rows. Each tuple consists of a stream id as the first element, followed by fields suitable for casting into an EventsStreamRow. """ def get_all_new_forward_event_rows(txn): sql = ("SELECT e.stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " LEFT JOIN event_relations USING (event_id)" " WHERE ? < stream_ordering AND stream_ordering <= ?" " ORDER BY stream_ordering ASC" " LIMIT ?") txn.execute(sql, (last_id, current_id, limit)) return txn.fetchall() return await self.db_pool.runInteraction( "get_all_new_forward_event_rows", get_all_new_forward_event_rows) async def get_ex_outlier_stream_rows(self, last_id: int, current_id: int) -> List[Tuple]: """Returns de-outliered events, for the Events replication stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to Returns: a list of events stream rows. Each tuple consists of a stream id as the first element, followed by fields suitable for casting into an EventsStreamRow. """ def get_ex_outlier_stream_rows_txn(txn): sql = ( "SELECT event_stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " INNER JOIN ex_outlier_stream USING (event_id)" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " LEFT JOIN event_relations USING (event_id)" " WHERE ? < event_stream_ordering" " AND event_stream_ordering <= ?" " ORDER BY event_stream_ordering ASC") txn.execute(sql, (last_id, current_id)) return txn.fetchall() return await self.db_pool.runInteraction( "get_ex_outlier_stream_rows", get_ex_outlier_stream_rows_txn) async def get_all_new_backfill_event_rows( self, instance_name: str, last_id: int, current_id: int, limit: int) -> Tuple[List[Tuple[int, list]], int, bool]: """Get updates for backfill replication stream, including all new backfilled events and events that have gone from being outliers to not. Args: instance_name: The writer we want to fetch updates from. Unused here since there is only ever one writer. last_id: The token to fetch updates from. Exclusive. current_id: The token to fetch updates up to. Inclusive. limit: The requested limit for the number of rows to return. The function may return more or fewer rows. Returns: A tuple consisting of: the updates, a token to use to fetch subsequent updates, and whether we returned fewer rows than exists between the requested tokens due to the limit. The token returned can be used in a subsequent call to this function to get further updatees. The updates are a list of 2-tuples of stream ID and the row data """ if last_id == current_id: return [], current_id, False def get_all_new_backfill_event_rows(txn): sql = ("SELECT -e.stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " LEFT JOIN event_relations USING (event_id)" " WHERE ? > stream_ordering AND stream_ordering >= ?" " ORDER BY stream_ordering ASC" " LIMIT ?") txn.execute(sql, (-last_id, -current_id, limit)) new_event_updates = [(row[0], row[1:]) for row in txn] limited = False if len(new_event_updates) == limit: upper_bound = new_event_updates[-1][0] limited = True else: upper_bound = current_id sql = ( "SELECT -event_stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " INNER JOIN ex_outlier_stream USING (event_id)" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " LEFT JOIN event_relations USING (event_id)" " WHERE ? > event_stream_ordering" " AND event_stream_ordering >= ?" " ORDER BY event_stream_ordering DESC") txn.execute(sql, (-last_id, -upper_bound)) new_event_updates.extend((row[0], row[1:]) for row in txn) if len(new_event_updates) >= limit: upper_bound = new_event_updates[-1][0] limited = True return new_event_updates, upper_bound, limited return await self.db_pool.runInteraction( "get_all_new_backfill_event_rows", get_all_new_backfill_event_rows) async def get_all_updated_current_state_deltas( self, from_token: int, to_token: int, target_row_count: int) -> Tuple[List[Tuple], int, bool]: """Fetch updates from current_state_delta_stream Args: from_token: The previous stream token. Updates from this stream id will be excluded. to_token: The current stream token (ie the upper limit). Updates up to this stream id will be included (modulo the 'limit' param) target_row_count: The number of rows to try to return. If more rows are available, we will set 'limited' in the result. In the event of a large batch, we may return more rows than this. Returns: A triplet `(updates, new_last_token, limited)`, where: * `updates` is a list of database tuples. * `new_last_token` is the new position in stream. * `limited` is whether there are more updates to fetch. """ def get_all_updated_current_state_deltas_txn(txn): sql = """ SELECT stream_id, room_id, type, state_key, event_id FROM current_state_delta_stream WHERE ? < stream_id AND stream_id <= ? ORDER BY stream_id ASC LIMIT ? """ txn.execute(sql, (from_token, to_token, target_row_count)) return txn.fetchall() def get_deltas_for_stream_id_txn(txn, stream_id): sql = """ SELECT stream_id, room_id, type, state_key, event_id FROM current_state_delta_stream WHERE stream_id = ? """ txn.execute(sql, [stream_id]) return txn.fetchall() # we need to make sure that, for every stream id in the results, we get *all* # the rows with that stream id. rows = await self.db_pool.runInteraction( "get_all_updated_current_state_deltas", get_all_updated_current_state_deltas_txn, ) # type: List[Tuple] # if we've got fewer rows than the limit, we're good if len(rows) < target_row_count: return rows, to_token, False # we hit the limit, so reduce the upper limit so that we exclude the stream id # of the last row in the result. assert rows[-1][0] <= to_token to_token = rows[-1][0] - 1 # search backwards through the list for the point to truncate for idx in range(len(rows) - 1, 0, -1): if rows[idx - 1][0] <= to_token: return rows[:idx], to_token, True # bother. We didn't get a full set of changes for even a single # stream id. let's run the query again, without a row limit, but for # just one stream id. to_token += 1 rows = await self.db_pool.runInteraction("get_deltas_for_stream_id", get_deltas_for_stream_id_txn, to_token) return rows, to_token, True async def is_event_after(self, event_id1, event_id2): """Returns True if event_id1 is after event_id2 in the stream """ to_1, so_1 = await self.get_event_ordering(event_id1) to_2, so_2 = await self.get_event_ordering(event_id2) return (to_1, so_1) > (to_2, so_2) @cached(max_entries=5000) async def get_event_ordering(self, event_id): res = await self.db_pool.simple_select_one( table="events", retcols=["topological_ordering", "stream_ordering"], keyvalues={"event_id": event_id}, allow_none=True, ) if not res: raise SynapseError(404, "Could not find event %s" % (event_id, )) return (int(res["topological_ordering"]), int(res["stream_ordering"])) async def get_next_event_to_expire(self) -> Optional[Tuple[str, int]]: """Retrieve the entry with the lowest expiry timestamp in the event_expiry table, or None if there's no more event to expire. Returns: A tuple containing the event ID as its first element and an expiry timestamp as its second one, if there's at least one row in the event_expiry table. None otherwise. """ def get_next_event_to_expire_txn(txn): txn.execute(""" SELECT event_id, expiry_ts FROM event_expiry ORDER BY expiry_ts ASC LIMIT 1 """) return txn.fetchone() return await self.db_pool.runInteraction( desc="get_next_event_to_expire", func=get_next_event_to_expire_txn)
class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore): def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) # `_can_write_to_account_data` indicates whether the current worker is allowed # to write account data. A value of `True` implies that `_account_data_id_gen` # is an `AbstractStreamIdGenerator` and not just a tracker. self._account_data_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): self._can_write_to_account_data = ( self._instance_name in hs.config.worker.writers.account_data) self._account_data_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="account_data", instance_name=self._instance_name, tables=[ ("room_account_data", "instance_name", "stream_id"), ("room_tags_revisions", "instance_name", "stream_id"), ("account_data", "instance_name", "stream_id"), ], sequence_name="account_data_sequence", writers=hs.config.worker.writers.account_data, ) else: # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if self._instance_name in hs.config.worker.writers.account_data: self._can_write_to_account_data = True self._account_data_id_gen = StreamIdGenerator( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) else: self._account_data_id_gen = SlavedIdTracker( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( "AccountDataAndTagsChangeCache", account_max) self.db_pool.updates.register_background_update_handler( "delete_account_data_for_deactivated_users", self._delete_account_data_for_deactivated_users, ) def get_max_account_data_stream_id(self) -> int: """Get the current max stream ID for account data stream Returns: int """ return self._account_data_id_gen.get_current_token() @cached() async def get_account_data_for_user( self, user_id: str ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: """Get all the client account_data for a user. Args: user_id: The user to get the account_data for. Returns: A 2-tuple of a dict of global account_data and a dict mapping from room_id string to per room account_data dicts. """ def get_account_data_for_user_txn( txn: LoggingTransaction, ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: rows = self.db_pool.simple_select_list_txn( txn, "account_data", {"user_id": user_id}, ["account_data_type", "content"], ) global_account_data = { row["account_data_type"]: db_to_json(row["content"]) for row in rows } rows = self.db_pool.simple_select_list_txn( txn, "room_account_data", {"user_id": user_id}, ["room_id", "account_data_type", "content"], ) by_room: Dict[str, Dict[str, JsonDict]] = {} for row in rows: room_data = by_room.setdefault(row["room_id"], {}) room_data[row["account_data_type"]] = db_to_json( row["content"]) return global_account_data, by_room return await self.db_pool.runInteraction( "get_account_data_for_user", get_account_data_for_user_txn) @cached(num_args=2, max_entries=5000, tree=True) async def get_global_account_data_by_type_for_user( self, user_id: str, data_type: str) -> Optional[JsonDict]: """ Returns: The account data. """ result = await self.db_pool.simple_select_one_onecol( table="account_data", keyvalues={ "user_id": user_id, "account_data_type": data_type }, retcol="content", desc="get_global_account_data_by_type_for_user", allow_none=True, ) if result: return db_to_json(result) else: return None @cached(num_args=2, tree=True) async def get_account_data_for_room(self, user_id: str, room_id: str) -> Dict[str, JsonDict]: """Get all the client account_data for a user for a room. Args: user_id: The user to get the account_data for. room_id: The room to get the account_data for. Returns: A dict of the room account_data """ def get_account_data_for_room_txn( txn: LoggingTransaction, ) -> Dict[str, JsonDict]: rows = self.db_pool.simple_select_list_txn( txn, "room_account_data", { "user_id": user_id, "room_id": room_id }, ["account_data_type", "content"], ) return { row["account_data_type"]: db_to_json(row["content"]) for row in rows } return await self.db_pool.runInteraction( "get_account_data_for_room", get_account_data_for_room_txn) @cached(num_args=3, max_entries=5000, tree=True) async def get_account_data_for_room_and_type( self, user_id: str, room_id: str, account_data_type: str) -> Optional[JsonDict]: """Get the client account_data of given type for a user for a room. Args: user_id: The user to get the account_data for. room_id: The room to get the account_data for. account_data_type: The account data type to get. Returns: The room account_data for that type, or None if there isn't any set. """ def get_account_data_for_room_and_type_txn( txn: LoggingTransaction, ) -> Optional[JsonDict]: content_json = self.db_pool.simple_select_one_onecol_txn( txn, table="room_account_data", keyvalues={ "user_id": user_id, "room_id": room_id, "account_data_type": account_data_type, }, retcol="content", allow_none=True, ) return db_to_json(content_json) if content_json else None return await self.db_pool.runInteraction( "get_account_data_for_room_and_type", get_account_data_for_room_and_type_txn) async def get_updated_global_account_data( self, last_id: int, current_id: int, limit: int) -> List[Tuple[int, str, str]]: """Get the global account_data that has changed, for the account_data stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: A list of tuples of stream_id int, user_id string, and type string. """ if last_id == current_id: return [] def get_updated_global_account_data_txn( txn: LoggingTransaction, ) -> List[Tuple[int, str, str]]: sql = ("SELECT stream_id, user_id, account_data_type" " FROM account_data WHERE ? < stream_id AND stream_id <= ?" " ORDER BY stream_id ASC LIMIT ?") txn.execute(sql, (last_id, current_id, limit)) return cast(List[Tuple[int, str, str]], txn.fetchall()) return await self.db_pool.runInteraction( "get_updated_global_account_data", get_updated_global_account_data_txn) async def get_updated_room_account_data( self, last_id: int, current_id: int, limit: int) -> List[Tuple[int, str, str, str]]: """Get the global account_data that has changed, for the account_data stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: A list of tuples of stream_id int, user_id string, room_id string and type string. """ if last_id == current_id: return [] def get_updated_room_account_data_txn( txn: LoggingTransaction, ) -> List[Tuple[int, str, str, str]]: sql = ( "SELECT stream_id, user_id, room_id, account_data_type" " FROM room_account_data WHERE ? < stream_id AND stream_id <= ?" " ORDER BY stream_id ASC LIMIT ?") txn.execute(sql, (last_id, current_id, limit)) return cast(List[Tuple[int, str, str, str]], txn.fetchall()) return await self.db_pool.runInteraction( "get_updated_room_account_data", get_updated_room_account_data_txn) async def get_updated_account_data_for_user( self, user_id: str, stream_id: int ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: """Get all the client account_data for a that's changed for a user Args: user_id: The user to get the account_data for. stream_id: The point in the stream since which to get updates Returns: A deferred pair of a dict of global account_data and a dict mapping from room_id string to per room account_data dicts. """ def get_updated_account_data_for_user_txn( txn: LoggingTransaction, ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: sql = ("SELECT account_data_type, content FROM account_data" " WHERE user_id = ? AND stream_id > ?") txn.execute(sql, (user_id, stream_id)) global_account_data = {row[0]: db_to_json(row[1]) for row in txn} sql = ( "SELECT room_id, account_data_type, content FROM room_account_data" " WHERE user_id = ? AND stream_id > ?") txn.execute(sql, (user_id, stream_id)) account_data_by_room: Dict[str, Dict[str, JsonDict]] = {} for row in txn: room_account_data = account_data_by_room.setdefault(row[0], {}) room_account_data[row[1]] = db_to_json(row[2]) return global_account_data, account_data_by_room changed = self._account_data_stream_cache.has_entity_changed( user_id, int(stream_id)) if not changed: return {}, {} return await self.db_pool.runInteraction( "get_updated_account_data_for_user", get_updated_account_data_for_user_txn) @cached(max_entries=5000, iterable=True) async def ignored_by(self, user_id: str) -> FrozenSet[str]: """ Get users which ignore the given user. Params: user_id: The user ID which might be ignored. Return: The user IDs which ignore the given user. """ return frozenset(await self.db_pool.simple_select_onecol( table="ignored_users", keyvalues={"ignored_user_id": user_id}, retcol="ignorer_user_id", desc="ignored_by", )) @cached(max_entries=5000, iterable=True) async def ignored_users(self, user_id: str) -> FrozenSet[str]: """ Get users which the given user ignores. Params: user_id: The user ID which is making the request. Return: The user IDs which are ignored by the given user. """ return frozenset(await self.db_pool.simple_select_onecol( table="ignored_users", keyvalues={"ignorer_user_id": user_id}, retcol="ignored_user_id", desc="ignored_users", )) def process_replication_rows( self, stream_name: str, instance_name: str, token: int, rows: Iterable[Any], ) -> None: if stream_name == TagAccountDataStream.NAME: self._account_data_id_gen.advance(instance_name, token) elif stream_name == AccountDataStream.NAME: self._account_data_id_gen.advance(instance_name, token) for row in rows: if not row.room_id: self.get_global_account_data_by_type_for_user.invalidate( (row.user_id, row.data_type)) self.get_account_data_for_user.invalidate((row.user_id, )) self.get_account_data_for_room.invalidate( (row.user_id, row.room_id)) self.get_account_data_for_room_and_type.invalidate( (row.user_id, row.room_id, row.data_type)) self._account_data_stream_cache.entity_has_changed( row.user_id, token) super().process_replication_rows(stream_name, instance_name, token, rows) async def add_account_data_to_room(self, user_id: str, room_id: str, account_data_type: str, content: JsonDict) -> int: """Add some account_data to a room for a user. Args: user_id: The user to add a tag for. room_id: The room to add a tag for. account_data_type: The type of account_data to add. content: A json object to associate with the tag. Returns: The maximum stream ID. """ assert self._can_write_to_account_data assert isinstance(self._account_data_id_gen, AbstractStreamIdGenerator) content_json = json_encoder.encode(content) async with self._account_data_id_gen.get_next() as next_id: # no need to lock here as room_account_data has a unique constraint # on (user_id, room_id, account_data_type) so simple_upsert will # retry if there is a conflict. await self.db_pool.simple_upsert( desc="add_room_account_data", table="room_account_data", keyvalues={ "user_id": user_id, "room_id": room_id, "account_data_type": account_data_type, }, values={ "stream_id": next_id, "content": content_json }, lock=False, ) self._account_data_stream_cache.entity_has_changed( user_id, next_id) self.get_account_data_for_user.invalidate((user_id, )) self.get_account_data_for_room.invalidate((user_id, room_id)) self.get_account_data_for_room_and_type.prefill( (user_id, room_id, account_data_type), content) return self._account_data_id_gen.get_current_token() async def add_account_data_for_user(self, user_id: str, account_data_type: str, content: JsonDict) -> int: """Add some global account_data for a user. Args: user_id: The user to add a tag for. account_data_type: The type of account_data to add. content: A json object to associate with the tag. Returns: The maximum stream ID. """ assert self._can_write_to_account_data assert isinstance(self._account_data_id_gen, AbstractStreamIdGenerator) async with self._account_data_id_gen.get_next() as next_id: await self.db_pool.runInteraction( "add_user_account_data", self._add_account_data_for_user, next_id, user_id, account_data_type, content, ) self._account_data_stream_cache.entity_has_changed( user_id, next_id) self.get_account_data_for_user.invalidate((user_id, )) self.get_global_account_data_by_type_for_user.invalidate( (user_id, account_data_type)) return self._account_data_id_gen.get_current_token() def _add_account_data_for_user( self, txn: LoggingTransaction, next_id: int, user_id: str, account_data_type: str, content: JsonDict, ) -> None: content_json = json_encoder.encode(content) # no need to lock here as account_data has a unique constraint on # (user_id, account_data_type) so simple_upsert will retry if # there is a conflict. self.db_pool.simple_upsert_txn( txn, table="account_data", keyvalues={ "user_id": user_id, "account_data_type": account_data_type }, values={ "stream_id": next_id, "content": content_json }, lock=False, ) # Ignored users get denormalized into a separate table as an optimisation. if account_data_type != AccountDataTypes.IGNORED_USER_LIST: return # Insert / delete to sync the list of ignored users. previously_ignored_users = set( self.db_pool.simple_select_onecol_txn( txn, table="ignored_users", keyvalues={"ignorer_user_id": user_id}, retcol="ignored_user_id", )) # If the data is invalid, no one is ignored. ignored_users_content = content.get("ignored_users", {}) if isinstance(ignored_users_content, dict): currently_ignored_users = set(ignored_users_content) else: currently_ignored_users = set() # If the data has not changed, nothing to do. if previously_ignored_users == currently_ignored_users: return # Delete entries which are no longer ignored. self.db_pool.simple_delete_many_txn( txn, table="ignored_users", column="ignored_user_id", values=previously_ignored_users - currently_ignored_users, keyvalues={"ignorer_user_id": user_id}, ) # Add entries which are newly ignored. self.db_pool.simple_insert_many_txn( txn, table="ignored_users", keys=("ignorer_user_id", "ignored_user_id"), values=[(user_id, u) for u in currently_ignored_users - previously_ignored_users ], ) # Invalidate the cache for any ignored users which were added or removed. for ignored_user_id in previously_ignored_users ^ currently_ignored_users: self._invalidate_cache_and_stream(txn, self.ignored_by, (ignored_user_id, )) self._invalidate_cache_and_stream(txn, self.ignored_users, (user_id, )) async def purge_account_data_for_user(self, user_id: str) -> None: """ Removes ALL the account data for a user. Intended to be used upon user deactivation. Also purges the user from the ignored_users cache table and the push_rules cache tables. """ await self.db_pool.runInteraction( "purge_account_data_for_user_txn", self._purge_account_data_for_user_txn, user_id, ) def _purge_account_data_for_user_txn(self, txn: LoggingTransaction, user_id: str) -> None: """ See `purge_account_data_for_user`. """ # Purge from the primary account_data tables. self.db_pool.simple_delete_txn(txn, table="account_data", keyvalues={"user_id": user_id}) self.db_pool.simple_delete_txn(txn, table="room_account_data", keyvalues={"user_id": user_id}) # Purge from ignored_users where this user is the ignorer. # N.B. We don't purge where this user is the ignoree, because that # interferes with other users' account data. # It's also not this user's data to delete! self.db_pool.simple_delete_txn(txn, table="ignored_users", keyvalues={"ignorer_user_id": user_id}) # Remove the push rules self.db_pool.simple_delete_txn(txn, table="push_rules", keyvalues={"user_name": user_id}) self.db_pool.simple_delete_txn(txn, table="push_rules_enable", keyvalues={"user_name": user_id}) self.db_pool.simple_delete_txn(txn, table="push_rules_stream", keyvalues={"user_id": user_id}) # Invalidate caches as appropriate self._invalidate_cache_and_stream( txn, self.get_account_data_for_room_and_type, (user_id, )) self._invalidate_cache_and_stream(txn, self.get_account_data_for_user, (user_id, )) self._invalidate_cache_and_stream( txn, self.get_global_account_data_by_type_for_user, (user_id, )) self._invalidate_cache_and_stream(txn, self.get_account_data_for_room, (user_id, )) self._invalidate_cache_and_stream(txn, self.get_push_rules_for_user, (user_id, )) self._invalidate_cache_and_stream(txn, self.get_push_rules_enabled_for_user, (user_id, )) # This user might be contained in the ignored_by cache for other users, # so we have to invalidate it all. self._invalidate_all_cache_and_stream(txn, self.ignored_by) async def _delete_account_data_for_deactivated_users( self, progress: dict, batch_size: int) -> int: """ Retroactively purges account data for users that have already been deactivated. Gets run as a background update caused by a schema delta. """ last_user: str = progress.get("last_user", "") def _delete_account_data_for_deactivated_users_txn( txn: LoggingTransaction, ) -> int: sql = """ SELECT name FROM users WHERE deactivated = ? and name > ? ORDER BY name ASC LIMIT ? """ txn.execute(sql, (1, last_user, batch_size)) users = [row[0] for row in txn] for user in users: self._purge_account_data_for_user_txn(txn, user_id=user) if users: self.db_pool.updates._background_update_progress_txn( txn, "delete_account_data_for_deactivated_users", {"last_user": users[-1]}, ) return len(users) number_deleted = await self.db_pool.runInteraction( "_delete_account_data_for_deactivated_users", _delete_account_data_for_deactivated_users_txn, ) if number_deleted < batch_size: await self.db_pool.updates._end_background_update( "delete_account_data_for_deactivated_users") return number_deleted
class ReceiptsWorkerStore(SQLBaseStore): def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): self._instance_name = hs.get_instance_name() self._receipts_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): self._can_write_to_receipts = ( self._instance_name in hs.config.worker.writers.receipts ) self._receipts_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="receipts", instance_name=self._instance_name, tables=[("receipts_linearized", "instance_name", "stream_id")], sequence_name="receipts_sequence", writers=hs.config.worker.writers.receipts, ) else: self._can_write_to_receipts = True # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if hs.get_instance_name() in hs.config.worker.writers.receipts: self._receipts_id_gen = StreamIdGenerator( db_conn, "receipts_linearized", "stream_id" ) else: self._receipts_id_gen = SlavedIdTracker( db_conn, "receipts_linearized", "stream_id" ) super().__init__(database, db_conn, hs) max_receipts_stream_id = self.get_max_receipt_stream_id() receipts_stream_prefill, min_receipts_stream_id = self.db_pool.get_cache_dict( db_conn, "receipts_linearized", entity_column="room_id", stream_column="stream_id", max_value=max_receipts_stream_id, limit=10000, ) self._receipts_stream_cache = StreamChangeCache( "ReceiptsRoomChangeCache", min_receipts_stream_id, prefilled_cache=receipts_stream_prefill, ) def get_max_receipt_stream_id(self) -> int: """Get the current max stream ID for receipts stream""" return self._receipts_id_gen.get_current_token() async def get_last_receipt_event_id_for_user( self, user_id: str, room_id: str, receipt_types: Collection[str] ) -> Optional[str]: """ Fetch the event ID for the latest receipt in a room with one of the given receipt types. Args: user_id: The user to fetch receipts for. room_id: The room ID to fetch the receipt for. receipt_type: The receipt types to fetch. Returns: The latest receipt, if one exists. """ result = await self.db_pool.runInteraction( "get_last_receipt_event_id_for_user", self.get_last_receipt_for_user_txn, user_id, room_id, receipt_types, ) if not result: return None event_id, _ = result return event_id def get_last_receipt_for_user_txn( self, txn: LoggingTransaction, user_id: str, room_id: str, receipt_types: Collection[str], ) -> Optional[Tuple[str, int]]: """ Fetch the event ID and stream_ordering for the latest receipt in a room with one of the given receipt types. Args: user_id: The user to fetch receipts for. room_id: The room ID to fetch the receipt for. receipt_type: The receipt types to fetch. Returns: The latest receipt, if one exists. """ clause, args = make_in_list_sql_clause( self.database_engine, "receipt_type", receipt_types ) sql = f""" SELECT event_id, stream_ordering FROM receipts_linearized INNER JOIN events USING (room_id, event_id) WHERE {clause} AND user_id = ? AND room_id = ? ORDER BY stream_ordering DESC LIMIT 1 """ args.extend((user_id, room_id)) txn.execute(sql, args) return cast(Optional[Tuple[str, int]], txn.fetchone()) async def get_receipts_for_user( self, user_id: str, receipt_types: Iterable[str] ) -> Dict[str, str]: """ Fetch the event IDs for the latest receipts sent by the given user. Args: user_id: The user to fetch receipts for. receipt_types: The receipt types to check. Returns: A map of room ID to the event ID of the latest receipt for that room. If the user has not sent a receipt to a room then it will not appear in the returned dictionary. """ results = await self.get_receipts_for_user_with_orderings( user_id, receipt_types ) # Reduce the result to room ID -> event ID. return { room_id: room_result["event_id"] for room_id, room_result in results.items() } async def get_receipts_for_user_with_orderings( self, user_id: str, receipt_types: Iterable[str] ) -> JsonDict: """ Fetch receipts for all rooms that the given user is joined to. Args: user_id: The user to fetch receipts for. receipt_types: The receipt types to fetch. Earlier receipt types are given priority if multiple receipts point to the same event. Returns: A map of room ID to the latest receipt (for the given types). """ results: JsonDict = {} for receipt_type in receipt_types: partial_result = await self._get_receipts_for_user_with_orderings( user_id, receipt_type ) for room_id, room_result in partial_result.items(): # If the room has not yet been seen, or the receipt is newer, # use it. if ( room_id not in results or results[room_id]["stream_ordering"] < room_result["stream_ordering"] ): results[room_id] = room_result return results @cached() async def _get_receipts_for_user_with_orderings( self, user_id: str, receipt_type: str ) -> JsonDict: """ Fetch receipts for all rooms that the given user is joined to. Args: user_id: The user to fetch receipts for. receipt_type: The receipt type to fetch. Returns: A map of room ID to the latest receipt information. """ def f(txn: LoggingTransaction) -> List[Tuple[str, str, int, int]]: sql = ( "SELECT rl.room_id, rl.event_id," " e.topological_ordering, e.stream_ordering" " FROM receipts_linearized AS rl" " INNER JOIN events AS e USING (room_id, event_id)" " WHERE rl.room_id = e.room_id" " AND rl.event_id = e.event_id" " AND user_id = ?" " AND receipt_type = ?" ) txn.execute(sql, (user_id, receipt_type)) return cast(List[Tuple[str, str, int, int]], txn.fetchall()) rows = await self.db_pool.runInteraction( "get_receipts_for_user_with_orderings", f ) return { row[0]: { "event_id": row[1], "topological_ordering": row[2], "stream_ordering": row[3], } for row in rows } async def get_linearized_receipts_for_rooms( self, room_ids: Iterable[str], to_key: int, from_key: Optional[int] = None ) -> List[dict]: """Get receipts for multiple rooms for sending to clients. Args: room_id: The room IDs to fetch receipts of. to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A list of receipts. """ room_ids = set(room_ids) if from_key is not None: # Only ask the database about rooms where there have been new # receipts added since `from_key` room_ids = self._receipts_stream_cache.get_entities_changed( room_ids, from_key ) results = await self._get_linearized_receipts_for_rooms( room_ids, to_key, from_key=from_key ) return [ev for res in results.values() for ev in res] async def get_linearized_receipts_for_room( self, room_id: str, to_key: int, from_key: Optional[int] = None ) -> List[dict]: """Get receipts for a single room for sending to clients. Args: room_ids: The room id. to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A list of receipts. """ if from_key is not None: # Check the cache first to see if any new receipts have been added # since`from_key`. If not we can no-op. if not self._receipts_stream_cache.has_entity_changed(room_id, from_key): return [] return await self._get_linearized_receipts_for_room(room_id, to_key, from_key) @cached(tree=True) async def _get_linearized_receipts_for_room( self, room_id: str, to_key: int, from_key: Optional[int] = None ) -> List[JsonDict]: """See get_linearized_receipts_for_room""" def f(txn: LoggingTransaction) -> List[Dict[str, Any]]: if from_key: sql = ( "SELECT * FROM receipts_linearized WHERE" " room_id = ? AND stream_id > ? AND stream_id <= ?" ) txn.execute(sql, (room_id, from_key, to_key)) else: sql = ( "SELECT * FROM receipts_linearized WHERE" " room_id = ? AND stream_id <= ?" ) txn.execute(sql, (room_id, to_key)) rows = self.db_pool.cursor_to_dict(txn) return rows rows = await self.db_pool.runInteraction("get_linearized_receipts_for_room", f) if not rows: return [] content: JsonDict = {} for row in rows: content.setdefault(row["event_id"], {}).setdefault(row["receipt_type"], {})[ row["user_id"] ] = db_to_json(row["data"]) return [{"type": EduTypes.RECEIPT, "room_id": room_id, "content": content}] @cachedList( cached_method_name="_get_linearized_receipts_for_room", list_name="room_ids", num_args=3, ) async def _get_linearized_receipts_for_rooms( self, room_ids: Collection[str], to_key: int, from_key: Optional[int] = None ) -> Dict[str, List[JsonDict]]: if not room_ids: return {} def f(txn: LoggingTransaction) -> List[Dict[str, Any]]: if from_key: sql = """ SELECT * FROM receipts_linearized WHERE stream_id > ? AND stream_id <= ? AND """ clause, args = make_in_list_sql_clause( self.database_engine, "room_id", room_ids ) txn.execute(sql + clause, [from_key, to_key] + list(args)) else: sql = """ SELECT * FROM receipts_linearized WHERE stream_id <= ? AND """ clause, args = make_in_list_sql_clause( self.database_engine, "room_id", room_ids ) txn.execute(sql + clause, [to_key] + list(args)) return self.db_pool.cursor_to_dict(txn) txn_results = await self.db_pool.runInteraction( "_get_linearized_receipts_for_rooms", f ) results: JsonDict = {} for row in txn_results: # We want a single event per room, since we want to batch the # receipts by room, event and type. room_event = results.setdefault( row["room_id"], {"type": EduTypes.RECEIPT, "room_id": row["room_id"], "content": {}}, ) # The content is of the form: # {"$foo:bar": { "read": { "@user:host": <receipt> }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) receipt_type[row["user_id"]] = db_to_json(row["data"]) results = { room_id: [results[room_id]] if room_id in results else [] for room_id in room_ids } return results @cached( num_args=2, ) async def get_linearized_receipts_for_all_rooms( self, to_key: int, from_key: Optional[int] = None ) -> Dict[str, JsonDict]: """Get receipts for all rooms between two stream_ids, up to a limit of the latest 100 read receipts. Args: to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A dictionary of roomids to a list of receipts. """ def f(txn: LoggingTransaction) -> List[Dict[str, Any]]: if from_key: sql = """ SELECT * FROM receipts_linearized WHERE stream_id > ? AND stream_id <= ? ORDER BY stream_id DESC LIMIT 100 """ txn.execute(sql, [from_key, to_key]) else: sql = """ SELECT * FROM receipts_linearized WHERE stream_id <= ? ORDER BY stream_id DESC LIMIT 100 """ txn.execute(sql, [to_key]) return self.db_pool.cursor_to_dict(txn) txn_results = await self.db_pool.runInteraction( "get_linearized_receipts_for_all_rooms", f ) results: JsonDict = {} for row in txn_results: # We want a single event per room, since we want to batch the # receipts by room, event and type. room_event = results.setdefault( row["room_id"], {"type": EduTypes.RECEIPT, "room_id": row["room_id"], "content": {}}, ) # The content is of the form: # {"$foo:bar": { "read": { "@user:host": <receipt> }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) receipt_type[row["user_id"]] = db_to_json(row["data"]) return results async def get_users_sent_receipts_between( self, last_id: int, current_id: int ) -> List[str]: """Get all users who sent receipts between `last_id` exclusive and `current_id` inclusive. Returns: The list of users. """ if last_id == current_id: return [] def _get_users_sent_receipts_between_txn(txn: LoggingTransaction) -> List[str]: sql = """ SELECT DISTINCT user_id FROM receipts_linearized WHERE ? < stream_id AND stream_id <= ? """ txn.execute(sql, (last_id, current_id)) return [r[0] for r in txn] return await self.db_pool.runInteraction( "get_users_sent_receipts_between", _get_users_sent_receipts_between_txn ) async def get_all_updated_receipts( self, instance_name: str, last_id: int, current_id: int, limit: int ) -> Tuple[List[Tuple[int, list]], int, bool]: """Get updates for receipts replication stream. Args: instance_name: The writer we want to fetch updates from. Unused here since there is only ever one writer. last_id: The token to fetch updates from. Exclusive. current_id: The token to fetch updates up to. Inclusive. limit: The requested limit for the number of rows to return. The function may return more or fewer rows. Returns: A tuple consisting of: the updates, a token to use to fetch subsequent updates, and whether we returned fewer rows than exists between the requested tokens due to the limit. The token returned can be used in a subsequent call to this function to get further updatees. The updates are a list of 2-tuples of stream ID and the row data """ if last_id == current_id: return [], current_id, False def get_all_updated_receipts_txn( txn: LoggingTransaction, ) -> Tuple[List[Tuple[int, list]], int, bool]: sql = """ SELECT stream_id, room_id, receipt_type, user_id, event_id, data FROM receipts_linearized WHERE ? < stream_id AND stream_id <= ? ORDER BY stream_id ASC LIMIT ? """ txn.execute(sql, (last_id, current_id, limit)) updates = cast( List[Tuple[int, list]], [(r[0], r[1:5] + (db_to_json(r[5]),)) for r in txn], ) limited = False upper_bound = current_id if len(updates) == limit: limited = True upper_bound = updates[-1][0] return updates, upper_bound, limited return await self.db_pool.runInteraction( "get_all_updated_receipts", get_all_updated_receipts_txn ) def invalidate_caches_for_receipt( self, room_id: str, receipt_type: str, user_id: str ) -> None: self._get_receipts_for_user_with_orderings.invalidate((user_id, receipt_type)) self._get_linearized_receipts_for_room.invalidate((room_id,)) # We use this method to invalidate so that we don't end up with circular # dependencies between the receipts and push action stores. self._attempt_to_invalidate_cache( "get_unread_event_push_actions_by_room_for_user", (room_id,) ) def process_replication_rows( self, stream_name: str, instance_name: str, token: int, rows: Iterable[Any], ) -> None: if stream_name == ReceiptsStream.NAME: self._receipts_id_gen.advance(instance_name, token) for row in rows: self.invalidate_caches_for_receipt( row.room_id, row.receipt_type, row.user_id ) self._receipts_stream_cache.entity_has_changed(row.room_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows) def _insert_linearized_receipt_txn( self, txn: LoggingTransaction, room_id: str, receipt_type: str, user_id: str, event_id: str, data: JsonDict, stream_id: int, ) -> Optional[int]: """Inserts a receipt into the database if it's newer than the current one. Returns: None if the receipt is older than the current receipt otherwise, the rx timestamp of the event that the receipt corresponds to (or 0 if the event is unknown) """ assert self._can_write_to_receipts res = self.db_pool.simple_select_one_txn( txn, table="events", retcols=["stream_ordering", "received_ts"], keyvalues={"event_id": event_id}, allow_none=True, ) stream_ordering = int(res["stream_ordering"]) if res else None rx_ts = res["received_ts"] if res else 0 # We don't want to clobber receipts for more recent events, so we # have to compare orderings of existing receipts if stream_ordering is not None: sql = ( "SELECT stream_ordering, event_id FROM events" " INNER JOIN receipts_linearized AS r USING (event_id, room_id)" " WHERE r.room_id = ? AND r.receipt_type = ? AND r.user_id = ?" ) txn.execute(sql, (room_id, receipt_type, user_id)) for so, eid in txn: if int(so) >= stream_ordering: logger.debug( "Ignoring new receipt for %s in favour of existing " "one for later event %s", event_id, eid, ) return None txn.call_after( self.invalidate_caches_for_receipt, room_id, receipt_type, user_id ) txn.call_after( self._receipts_stream_cache.entity_has_changed, room_id, stream_id ) self.db_pool.simple_upsert_txn( txn, table="receipts_linearized", keyvalues={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, }, values={ "stream_id": stream_id, "event_id": event_id, "data": json_encoder.encode(data), }, # receipts_linearized has a unique constraint on # (user_id, room_id, receipt_type), so no need to lock lock=False, ) return rx_ts def _graph_to_linear( self, txn: LoggingTransaction, room_id: str, event_ids: List[str] ) -> str: """ Generate a linearized event from a list of events (i.e. a list of forward extremities in the room). This should allow for calculation of the correct read receipt even if servers have different event ordering. Args: txn: The transaction room_id: The room ID the events are in. event_ids: The list of event IDs to linearize. Returns: The linearized event ID. """ # TODO: Make this better. clause, args = make_in_list_sql_clause( self.database_engine, "event_id", event_ids ) sql = """ SELECT event_id WHERE room_id = ? AND stream_ordering IN ( SELECT max(stream_ordering) WHERE %s ) """ % ( clause, ) txn.execute(sql, [room_id] + list(args)) rows = txn.fetchall() if rows: return rows[0][0] else: raise RuntimeError("Unrecognized event_ids: %r" % (event_ids,)) async def insert_receipt( self, room_id: str, receipt_type: str, user_id: str, event_ids: List[str], data: dict, ) -> Optional[Tuple[int, int]]: """Insert a receipt, either from local client or remote server. Automatically does conversion between linearized and graph representations. Returns: The new receipts stream ID and token, if the receipt is newer than what was previously persisted. None, otherwise. """ assert self._can_write_to_receipts if not event_ids: return None if len(event_ids) == 1: linearized_event_id = event_ids[0] else: # we need to points in graph -> linearized form. linearized_event_id = await self.db_pool.runInteraction( "insert_receipt_conv", self._graph_to_linear, room_id, event_ids ) async with self._receipts_id_gen.get_next() as stream_id: # type: ignore[attr-defined] event_ts = await self.db_pool.runInteraction( "insert_linearized_receipt", self._insert_linearized_receipt_txn, room_id, receipt_type, user_id, linearized_event_id, data, stream_id=stream_id, # Read committed is actually beneficial here because we check for a receipt with # greater stream order, and checking the very latest data at select time is better # than the data at transaction start time. isolation_level=IsolationLevel.READ_COMMITTED, ) # If the receipt was older than the currently persisted one, nothing to do. if event_ts is None: return None now = self._clock.time_msec() logger.debug( "RR for event %s in %s (%i ms old)", linearized_event_id, room_id, now - event_ts, ) await self.db_pool.runInteraction( "insert_graph_receipt", self._insert_graph_receipt_txn, room_id, receipt_type, user_id, event_ids, data, ) max_persisted_id = self._receipts_id_gen.get_current_token() return stream_id, max_persisted_id def _insert_graph_receipt_txn( self, txn: LoggingTransaction, room_id: str, receipt_type: str, user_id: str, event_ids: List[str], data: JsonDict, ) -> None: assert self._can_write_to_receipts txn.call_after( self._get_receipts_for_user_with_orderings.invalidate, (user_id, receipt_type), ) # FIXME: This shouldn't invalidate the whole cache txn.call_after(self._get_linearized_receipts_for_room.invalidate, (room_id,)) self.db_pool.simple_delete_txn( txn, table="receipts_graph", keyvalues={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, }, ) self.db_pool.simple_insert_txn( txn, table="receipts_graph", values={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, "event_ids": json_encoder.encode(event_ids), "data": json_encoder.encode(data), }, )
def __init__(self, db_conn, hs): self._account_data_id_gen = SlavedIdTracker( db_conn, "account_data_max_stream_id", "stream_id", ) super(SlavedAccountDataStore, self).__init__(db_conn, hs)
class AccountDataWorkerStore(SQLBaseStore): """This is an abstract base class where subclasses must implement `get_max_account_data_stream_id` which can be called in the initializer. """ def __init__(self, database: DatabasePool, db_conn, hs): self._instance_name = hs.get_instance_name() if isinstance(database.engine, PostgresEngine): self._can_write_to_account_data = ( self._instance_name in hs.config.worker.writers.account_data) self._account_data_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="account_data", instance_name=self._instance_name, tables=[ ("room_account_data", "instance_name", "stream_id"), ("room_tags_revisions", "instance_name", "stream_id"), ("account_data", "instance_name", "stream_id"), ], sequence_name="account_data_sequence", writers=hs.config.worker.writers.account_data, ) else: self._can_write_to_account_data = True # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if hs.get_instance_name() in hs.config.worker.writers.account_data: self._account_data_id_gen = StreamIdGenerator( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) else: self._account_data_id_gen = SlavedIdTracker( db_conn, "room_account_data", "stream_id", extra_tables=[("room_tags_revisions", "stream_id")], ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( "AccountDataAndTagsChangeCache", account_max) super().__init__(database, db_conn, hs) def get_max_account_data_stream_id(self) -> int: """Get the current max stream ID for account data stream Returns: int """ return self._account_data_id_gen.get_current_token() @cached() async def get_account_data_for_user( self, user_id: str ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: """Get all the client account_data for a user. Args: user_id: The user to get the account_data for. Returns: A 2-tuple of a dict of global account_data and a dict mapping from room_id string to per room account_data dicts. """ def get_account_data_for_user_txn(txn): rows = self.db_pool.simple_select_list_txn( txn, "account_data", {"user_id": user_id}, ["account_data_type", "content"], ) global_account_data = { row["account_data_type"]: db_to_json(row["content"]) for row in rows } rows = self.db_pool.simple_select_list_txn( txn, "room_account_data", {"user_id": user_id}, ["room_id", "account_data_type", "content"], ) by_room = {} for row in rows: room_data = by_room.setdefault(row["room_id"], {}) room_data[row["account_data_type"]] = db_to_json( row["content"]) return global_account_data, by_room return await self.db_pool.runInteraction( "get_account_data_for_user", get_account_data_for_user_txn) @cached(num_args=2, max_entries=5000) async def get_global_account_data_by_type_for_user( self, data_type: str, user_id: str) -> Optional[JsonDict]: """ Returns: The account data. """ result = await self.db_pool.simple_select_one_onecol( table="account_data", keyvalues={ "user_id": user_id, "account_data_type": data_type }, retcol="content", desc="get_global_account_data_by_type_for_user", allow_none=True, ) if result: return db_to_json(result) else: return None @cached(num_args=2) async def get_account_data_for_room(self, user_id: str, room_id: str) -> Dict[str, JsonDict]: """Get all the client account_data for a user for a room. Args: user_id: The user to get the account_data for. room_id: The room to get the account_data for. Returns: A dict of the room account_data """ def get_account_data_for_room_txn(txn): rows = self.db_pool.simple_select_list_txn( txn, "room_account_data", { "user_id": user_id, "room_id": room_id }, ["account_data_type", "content"], ) return { row["account_data_type"]: db_to_json(row["content"]) for row in rows } return await self.db_pool.runInteraction( "get_account_data_for_room", get_account_data_for_room_txn) @cached(num_args=3, max_entries=5000) async def get_account_data_for_room_and_type( self, user_id: str, room_id: str, account_data_type: str) -> Optional[JsonDict]: """Get the client account_data of given type for a user for a room. Args: user_id: The user to get the account_data for. room_id: The room to get the account_data for. account_data_type: The account data type to get. Returns: The room account_data for that type, or None if there isn't any set. """ def get_account_data_for_room_and_type_txn(txn): content_json = self.db_pool.simple_select_one_onecol_txn( txn, table="room_account_data", keyvalues={ "user_id": user_id, "room_id": room_id, "account_data_type": account_data_type, }, retcol="content", allow_none=True, ) return db_to_json(content_json) if content_json else None return await self.db_pool.runInteraction( "get_account_data_for_room_and_type", get_account_data_for_room_and_type_txn) async def get_updated_global_account_data( self, last_id: int, current_id: int, limit: int) -> List[Tuple[int, str, str]]: """Get the global account_data that has changed, for the account_data stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: A list of tuples of stream_id int, user_id string, and type string. """ if last_id == current_id: return [] def get_updated_global_account_data_txn(txn): sql = ("SELECT stream_id, user_id, account_data_type" " FROM account_data WHERE ? < stream_id AND stream_id <= ?" " ORDER BY stream_id ASC LIMIT ?") txn.execute(sql, (last_id, current_id, limit)) return txn.fetchall() return await self.db_pool.runInteraction( "get_updated_global_account_data", get_updated_global_account_data_txn) async def get_updated_room_account_data( self, last_id: int, current_id: int, limit: int) -> List[Tuple[int, str, str, str]]: """Get the global account_data that has changed, for the account_data stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: A list of tuples of stream_id int, user_id string, room_id string and type string. """ if last_id == current_id: return [] def get_updated_room_account_data_txn(txn): sql = ( "SELECT stream_id, user_id, room_id, account_data_type" " FROM room_account_data WHERE ? < stream_id AND stream_id <= ?" " ORDER BY stream_id ASC LIMIT ?") txn.execute(sql, (last_id, current_id, limit)) return txn.fetchall() return await self.db_pool.runInteraction( "get_updated_room_account_data", get_updated_room_account_data_txn) async def get_updated_account_data_for_user( self, user_id: str, stream_id: int ) -> Tuple[Dict[str, JsonDict], Dict[str, Dict[str, JsonDict]]]: """Get all the client account_data for a that's changed for a user Args: user_id: The user to get the account_data for. stream_id: The point in the stream since which to get updates Returns: A deferred pair of a dict of global account_data and a dict mapping from room_id string to per room account_data dicts. """ def get_updated_account_data_for_user_txn(txn): sql = ("SELECT account_data_type, content FROM account_data" " WHERE user_id = ? AND stream_id > ?") txn.execute(sql, (user_id, stream_id)) global_account_data = {row[0]: db_to_json(row[1]) for row in txn} sql = ( "SELECT room_id, account_data_type, content FROM room_account_data" " WHERE user_id = ? AND stream_id > ?") txn.execute(sql, (user_id, stream_id)) account_data_by_room = {} for row in txn: room_account_data = account_data_by_room.setdefault(row[0], {}) room_account_data[row[1]] = db_to_json(row[2]) return global_account_data, account_data_by_room changed = self._account_data_stream_cache.has_entity_changed( user_id, int(stream_id)) if not changed: return ({}, {}) return await self.db_pool.runInteraction( "get_updated_account_data_for_user", get_updated_account_data_for_user_txn) @cached(max_entries=5000, iterable=True) async def ignored_by(self, user_id: str) -> Set[str]: """ Get users which ignore the given user. Params: user_id: The user ID which might be ignored. Return: The user IDs which ignore the given user. """ return set(await self.db_pool.simple_select_onecol( table="ignored_users", keyvalues={"ignored_user_id": user_id}, retcol="ignorer_user_id", desc="ignored_by", )) def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == TagAccountDataStream.NAME: self._account_data_id_gen.advance(instance_name, token) for row in rows: self.get_tags_for_user.invalidate((row.user_id, )) self._account_data_stream_cache.entity_has_changed( row.user_id, token) elif stream_name == AccountDataStream.NAME: self._account_data_id_gen.advance(instance_name, token) for row in rows: if not row.room_id: self.get_global_account_data_by_type_for_user.invalidate( (row.data_type, row.user_id)) self.get_account_data_for_user.invalidate((row.user_id, )) self.get_account_data_for_room.invalidate( (row.user_id, row.room_id)) self.get_account_data_for_room_and_type.invalidate( (row.user_id, row.room_id, row.data_type)) self._account_data_stream_cache.entity_has_changed( row.user_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows) async def add_account_data_to_room(self, user_id: str, room_id: str, account_data_type: str, content: JsonDict) -> int: """Add some account_data to a room for a user. Args: user_id: The user to add a tag for. room_id: The room to add a tag for. account_data_type: The type of account_data to add. content: A json object to associate with the tag. Returns: The maximum stream ID. """ assert self._can_write_to_account_data content_json = json_encoder.encode(content) async with self._account_data_id_gen.get_next() as next_id: # no need to lock here as room_account_data has a unique constraint # on (user_id, room_id, account_data_type) so simple_upsert will # retry if there is a conflict. await self.db_pool.simple_upsert( desc="add_room_account_data", table="room_account_data", keyvalues={ "user_id": user_id, "room_id": room_id, "account_data_type": account_data_type, }, values={ "stream_id": next_id, "content": content_json }, lock=False, ) self._account_data_stream_cache.entity_has_changed( user_id, next_id) self.get_account_data_for_user.invalidate((user_id, )) self.get_account_data_for_room.invalidate((user_id, room_id)) self.get_account_data_for_room_and_type.prefill( (user_id, room_id, account_data_type), content) return self._account_data_id_gen.get_current_token() async def add_account_data_for_user(self, user_id: str, account_data_type: str, content: JsonDict) -> int: """Add some account_data to a room for a user. Args: user_id: The user to add a tag for. account_data_type: The type of account_data to add. content: A json object to associate with the tag. Returns: The maximum stream ID. """ assert self._can_write_to_account_data async with self._account_data_id_gen.get_next() as next_id: await self.db_pool.runInteraction( "add_user_account_data", self._add_account_data_for_user, next_id, user_id, account_data_type, content, ) self._account_data_stream_cache.entity_has_changed( user_id, next_id) self.get_account_data_for_user.invalidate((user_id, )) self.get_global_account_data_by_type_for_user.invalidate( (account_data_type, user_id)) return self._account_data_id_gen.get_current_token() def _add_account_data_for_user( self, txn, next_id: int, user_id: str, account_data_type: str, content: JsonDict, ) -> None: content_json = json_encoder.encode(content) # no need to lock here as account_data has a unique constraint on # (user_id, account_data_type) so simple_upsert will retry if # there is a conflict. self.db_pool.simple_upsert_txn( txn, table="account_data", keyvalues={ "user_id": user_id, "account_data_type": account_data_type }, values={ "stream_id": next_id, "content": content_json }, lock=False, ) # Ignored users get denormalized into a separate table as an optimisation. if account_data_type != AccountDataTypes.IGNORED_USER_LIST: return # Insert / delete to sync the list of ignored users. previously_ignored_users = set( self.db_pool.simple_select_onecol_txn( txn, table="ignored_users", keyvalues={"ignorer_user_id": user_id}, retcol="ignored_user_id", )) # If the data is invalid, no one is ignored. ignored_users_content = content.get("ignored_users", {}) if isinstance(ignored_users_content, dict): currently_ignored_users = set(ignored_users_content) else: currently_ignored_users = set() # Delete entries which are no longer ignored. self.db_pool.simple_delete_many_txn( txn, table="ignored_users", column="ignored_user_id", iterable=previously_ignored_users - currently_ignored_users, keyvalues={"ignorer_user_id": user_id}, ) # Add entries which are newly ignored. self.db_pool.simple_insert_many_txn( txn, table="ignored_users", values=[{ "ignorer_user_id": user_id, "ignored_user_id": u } for u in currently_ignored_users - previously_ignored_users], ) # Invalidate the cache for any ignored users which were added or removed. for ignored_user_id in previously_ignored_users ^ currently_ignored_users: self._invalidate_cache_and_stream(txn, self.ignored_by, (ignored_user_id, ))
class ReceiptsWorkerStore(SQLBaseStore): def __init__(self, database: DatabasePool, db_conn, hs): self._instance_name = hs.get_instance_name() if isinstance(database.engine, PostgresEngine): self._can_write_to_receipts = ( self._instance_name in hs.config.worker.writers.receipts) self._receipts_id_gen = MultiWriterIdGenerator( db_conn=db_conn, db=database, stream_name="receipts", instance_name=self._instance_name, tables=[("receipts_linearized", "instance_name", "stream_id")], sequence_name="receipts_sequence", writers=hs.config.worker.writers.receipts, ) else: self._can_write_to_receipts = True # We shouldn't be running in worker mode with SQLite, but its useful # to support it for unit tests. # # If this process is the writer than we need to use # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). if hs.get_instance_name() in hs.config.worker.writers.receipts: self._receipts_id_gen = StreamIdGenerator( db_conn, "receipts_linearized", "stream_id") else: self._receipts_id_gen = SlavedIdTracker( db_conn, "receipts_linearized", "stream_id") super().__init__(database, db_conn, hs) self._receipts_stream_cache = StreamChangeCache( "ReceiptsRoomChangeCache", self.get_max_receipt_stream_id()) def get_max_receipt_stream_id(self): """Get the current max stream ID for receipts stream Returns: int """ return self._receipts_id_gen.get_current_token() @cached() async def get_users_with_read_receipts_in_room(self, room_id): receipts = await self.get_receipts_for_room(room_id, "m.read") return {r["user_id"] for r in receipts} @cached(num_args=2) async def get_receipts_for_room(self, room_id: str, receipt_type: str) -> List[Dict[str, Any]]: return await self.db_pool.simple_select_list( table="receipts_linearized", keyvalues={ "room_id": room_id, "receipt_type": receipt_type }, retcols=("user_id", "event_id"), desc="get_receipts_for_room", ) @cached(num_args=3) async def get_last_receipt_event_id_for_user( self, user_id: str, room_id: str, receipt_type: str) -> Optional[str]: return await self.db_pool.simple_select_one_onecol( table="receipts_linearized", keyvalues={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, }, retcol="event_id", desc="get_own_receipt_for_user", allow_none=True, ) @cached(num_args=2) async def get_receipts_for_user(self, user_id, receipt_type): rows = await self.db_pool.simple_select_list( table="receipts_linearized", keyvalues={ "user_id": user_id, "receipt_type": receipt_type }, retcols=("room_id", "event_id"), desc="get_receipts_for_user", ) return {row["room_id"]: row["event_id"] for row in rows} async def get_receipts_for_user_with_orderings(self, user_id, receipt_type): def f(txn): sql = ("SELECT rl.room_id, rl.event_id," " e.topological_ordering, e.stream_ordering" " FROM receipts_linearized AS rl" " INNER JOIN events AS e USING (room_id, event_id)" " WHERE rl.room_id = e.room_id" " AND rl.event_id = e.event_id" " AND user_id = ?") txn.execute(sql, (user_id, )) return txn.fetchall() rows = await self.db_pool.runInteraction( "get_receipts_for_user_with_orderings", f) return { row[0]: { "event_id": row[1], "topological_ordering": row[2], "stream_ordering": row[3], } for row in rows } async def get_linearized_receipts_for_rooms( self, room_ids: List[str], to_key: int, from_key: Optional[int] = None) -> List[dict]: """Get receipts for multiple rooms for sending to clients. Args: room_id: List of room_ids. to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A list of receipts. """ room_ids = set(room_ids) if from_key is not None: # Only ask the database about rooms where there have been new # receipts added since `from_key` room_ids = self._receipts_stream_cache.get_entities_changed( room_ids, from_key) results = await self._get_linearized_receipts_for_rooms( room_ids, to_key, from_key=from_key) return [ev for res in results.values() for ev in res] async def get_linearized_receipts_for_room( self, room_id: str, to_key: int, from_key: Optional[int] = None) -> List[dict]: """Get receipts for a single room for sending to clients. Args: room_ids: The room id. to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A list of receipts. """ if from_key is not None: # Check the cache first to see if any new receipts have been added # since`from_key`. If not we can no-op. if not self._receipts_stream_cache.has_entity_changed( room_id, from_key): return [] return await self._get_linearized_receipts_for_room( room_id, to_key, from_key) @cached(num_args=3, tree=True) async def _get_linearized_receipts_for_room( self, room_id: str, to_key: int, from_key: Optional[int] = None) -> List[dict]: """See get_linearized_receipts_for_room""" def f(txn): if from_key: sql = ("SELECT * FROM receipts_linearized WHERE" " room_id = ? AND stream_id > ? AND stream_id <= ?") txn.execute(sql, (room_id, from_key, to_key)) else: sql = ("SELECT * FROM receipts_linearized WHERE" " room_id = ? AND stream_id <= ?") txn.execute(sql, (room_id, to_key)) rows = self.db_pool.cursor_to_dict(txn) return rows rows = await self.db_pool.runInteraction( "get_linearized_receipts_for_room", f) if not rows: return [] content = {} for row in rows: content.setdefault(row["event_id"], {}).setdefault( row["receipt_type"], {})[row["user_id"]] = db_to_json(row["data"]) return [{"type": "m.receipt", "room_id": room_id, "content": content}] @cachedList( cached_method_name="_get_linearized_receipts_for_room", list_name="room_ids", num_args=3, ) async def _get_linearized_receipts_for_rooms(self, room_ids, to_key, from_key=None): if not room_ids: return {} def f(txn): if from_key: sql = """ SELECT * FROM receipts_linearized WHERE stream_id > ? AND stream_id <= ? AND """ clause, args = make_in_list_sql_clause(self.database_engine, "room_id", room_ids) txn.execute(sql + clause, [from_key, to_key] + list(args)) else: sql = """ SELECT * FROM receipts_linearized WHERE stream_id <= ? AND """ clause, args = make_in_list_sql_clause(self.database_engine, "room_id", room_ids) txn.execute(sql + clause, [to_key] + list(args)) return self.db_pool.cursor_to_dict(txn) txn_results = await self.db_pool.runInteraction( "_get_linearized_receipts_for_rooms", f) results = {} for row in txn_results: # We want a single event per room, since we want to batch the # receipts by room, event and type. room_event = results.setdefault( row["room_id"], { "type": "m.receipt", "room_id": row["room_id"], "content": {} }, ) # The content is of the form: # {"$foo:bar": { "read": { "@user:host": <receipt> }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) receipt_type[row["user_id"]] = db_to_json(row["data"]) results = { room_id: [results[room_id]] if room_id in results else [] for room_id in room_ids } return results @cached( num_args=2, ) async def get_linearized_receipts_for_all_rooms( self, to_key: int, from_key: Optional[int] = None) -> Dict[str, JsonDict]: """Get receipts for all rooms between two stream_ids, up to a limit of the latest 100 read receipts. Args: to_key: Max stream id to fetch receipts up to. from_key: Min stream id to fetch receipts from. None fetches from the start. Returns: A dictionary of roomids to a list of receipts. """ def f(txn): if from_key: sql = """ SELECT * FROM receipts_linearized WHERE stream_id > ? AND stream_id <= ? ORDER BY stream_id DESC LIMIT 100 """ txn.execute(sql, [from_key, to_key]) else: sql = """ SELECT * FROM receipts_linearized WHERE stream_id <= ? ORDER BY stream_id DESC LIMIT 100 """ txn.execute(sql, [to_key]) return self.db_pool.cursor_to_dict(txn) txn_results = await self.db_pool.runInteraction( "get_linearized_receipts_for_all_rooms", f) results = {} for row in txn_results: # We want a single event per room, since we want to batch the # receipts by room, event and type. room_event = results.setdefault( row["room_id"], { "type": "m.receipt", "room_id": row["room_id"], "content": {} }, ) # The content is of the form: # {"$foo:bar": { "read": { "@user:host": <receipt> }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) receipt_type[row["user_id"]] = db_to_json(row["data"]) return results async def get_users_sent_receipts_between(self, last_id: int, current_id: int) -> List[str]: """Get all users who sent receipts between `last_id` exclusive and `current_id` inclusive. Returns: The list of users. """ if last_id == current_id: return defer.succeed([]) def _get_users_sent_receipts_between_txn(txn): sql = """ SELECT DISTINCT user_id FROM receipts_linearized WHERE ? < stream_id AND stream_id <= ? """ txn.execute(sql, (last_id, current_id)) return [r[0] for r in txn] return await self.db_pool.runInteraction( "get_users_sent_receipts_between", _get_users_sent_receipts_between_txn) async def get_all_updated_receipts( self, instance_name: str, last_id: int, current_id: int, limit: int) -> Tuple[List[Tuple[int, list]], int, bool]: """Get updates for receipts replication stream. Args: instance_name: The writer we want to fetch updates from. Unused here since there is only ever one writer. last_id: The token to fetch updates from. Exclusive. current_id: The token to fetch updates up to. Inclusive. limit: The requested limit for the number of rows to return. The function may return more or fewer rows. Returns: A tuple consisting of: the updates, a token to use to fetch subsequent updates, and whether we returned fewer rows than exists between the requested tokens due to the limit. The token returned can be used in a subsequent call to this function to get further updatees. The updates are a list of 2-tuples of stream ID and the row data """ if last_id == current_id: return [], current_id, False def get_all_updated_receipts_txn(txn): sql = """ SELECT stream_id, room_id, receipt_type, user_id, event_id, data FROM receipts_linearized WHERE ? < stream_id AND stream_id <= ? ORDER BY stream_id ASC LIMIT ? """ txn.execute(sql, (last_id, current_id, limit)) updates = [(r[0], r[1:5] + (db_to_json(r[5]), )) for r in txn] limited = False upper_bound = current_id if len(updates) == limit: limited = True upper_bound = updates[-1][0] return updates, upper_bound, limited return await self.db_pool.runInteraction("get_all_updated_receipts", get_all_updated_receipts_txn) def _invalidate_get_users_with_receipts_in_room(self, room_id: str, receipt_type: str, user_id: str): if receipt_type != "m.read": return res = self.get_users_with_read_receipts_in_room.cache.get_immediate( room_id, None, update_metrics=False) if res and user_id in res: # We'd only be adding to the set, so no point invalidating if the # user is already there return self.get_users_with_read_receipts_in_room.invalidate((room_id, )) def invalidate_caches_for_receipt(self, room_id, receipt_type, user_id): self.get_receipts_for_user.invalidate((user_id, receipt_type)) self._get_linearized_receipts_for_room.invalidate_many((room_id, )) self.get_last_receipt_event_id_for_user.invalidate( (user_id, room_id, receipt_type)) self._invalidate_get_users_with_receipts_in_room( room_id, receipt_type, user_id) self.get_receipts_for_room.invalidate((room_id, receipt_type)) def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == ReceiptsStream.NAME: self._receipts_id_gen.advance(instance_name, token) for row in rows: self.invalidate_caches_for_receipt(row.room_id, row.receipt_type, row.user_id) self._receipts_stream_cache.entity_has_changed( row.room_id, token) return super().process_replication_rows(stream_name, instance_name, token, rows) def insert_linearized_receipt_txn(self, txn, room_id, receipt_type, user_id, event_id, data, stream_id): """Inserts a read-receipt into the database if it's newer than the current RR Returns: int|None None if the RR is older than the current RR otherwise, the rx timestamp of the event that the RR corresponds to (or 0 if the event is unknown) """ assert self._can_write_to_receipts res = self.db_pool.simple_select_one_txn( txn, table="events", retcols=["stream_ordering", "received_ts"], keyvalues={"event_id": event_id}, allow_none=True, ) stream_ordering = int(res["stream_ordering"]) if res else None rx_ts = res["received_ts"] if res else 0 # We don't want to clobber receipts for more recent events, so we # have to compare orderings of existing receipts if stream_ordering is not None: sql = ( "SELECT stream_ordering, event_id FROM events" " INNER JOIN receipts_linearized as r USING (event_id, room_id)" " WHERE r.room_id = ? AND r.receipt_type = ? AND r.user_id = ?" ) txn.execute(sql, (room_id, receipt_type, user_id)) for so, eid in txn: if int(so) >= stream_ordering: logger.debug( "Ignoring new receipt for %s in favour of existing " "one for later event %s", event_id, eid, ) return None txn.call_after(self.invalidate_caches_for_receipt, room_id, receipt_type, user_id) txn.call_after(self._receipts_stream_cache.entity_has_changed, room_id, stream_id) self.db_pool.simple_upsert_txn( txn, table="receipts_linearized", keyvalues={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, }, values={ "stream_id": stream_id, "event_id": event_id, "data": json_encoder.encode(data), }, # receipts_linearized has a unique constraint on # (user_id, room_id, receipt_type), so no need to lock lock=False, ) if receipt_type == "m.read" and stream_ordering is not None: self._remove_old_push_actions_before_txn( txn, room_id=room_id, user_id=user_id, stream_ordering=stream_ordering) return rx_ts async def insert_receipt( self, room_id: str, receipt_type: str, user_id: str, event_ids: List[str], data: dict, ) -> Optional[Tuple[int, int]]: """Insert a receipt, either from local client or remote server. Automatically does conversion between linearized and graph representations. """ assert self._can_write_to_receipts if not event_ids: return None if len(event_ids) == 1: linearized_event_id = event_ids[0] else: # we need to points in graph -> linearized form. # TODO: Make this better. def graph_to_linear(txn): clause, args = make_in_list_sql_clause(self.database_engine, "event_id", event_ids) sql = """ SELECT event_id WHERE room_id = ? AND stream_ordering IN ( SELECT max(stream_ordering) WHERE %s ) """ % (clause, ) txn.execute(sql, [room_id] + list(args)) rows = txn.fetchall() if rows: return rows[0][0] else: raise RuntimeError("Unrecognized event_ids: %r" % (event_ids, )) linearized_event_id = await self.db_pool.runInteraction( "insert_receipt_conv", graph_to_linear) async with self._receipts_id_gen.get_next() as stream_id: event_ts = await self.db_pool.runInteraction( "insert_linearized_receipt", self.insert_linearized_receipt_txn, room_id, receipt_type, user_id, linearized_event_id, data, stream_id=stream_id, ) if event_ts is None: return None now = self._clock.time_msec() logger.debug( "RR for event %s in %s (%i ms old)", linearized_event_id, room_id, now - event_ts, ) await self.insert_graph_receipt(room_id, receipt_type, user_id, event_ids, data) max_persisted_id = self._receipts_id_gen.get_current_token() return stream_id, max_persisted_id async def insert_graph_receipt(self, room_id, receipt_type, user_id, event_ids, data): assert self._can_write_to_receipts return await self.db_pool.runInteraction( "insert_graph_receipt", self.insert_graph_receipt_txn, room_id, receipt_type, user_id, event_ids, data, ) def insert_graph_receipt_txn(self, txn, room_id, receipt_type, user_id, event_ids, data): assert self._can_write_to_receipts txn.call_after(self.get_receipts_for_room.invalidate, (room_id, receipt_type)) txn.call_after( self._invalidate_get_users_with_receipts_in_room, room_id, receipt_type, user_id, ) txn.call_after(self.get_receipts_for_user.invalidate, (user_id, receipt_type)) # FIXME: This shouldn't invalidate the whole cache txn.call_after(self._get_linearized_receipts_for_room.invalidate_many, (room_id, )) self.db_pool.simple_delete_txn( txn, table="receipts_graph", keyvalues={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, }, ) self.db_pool.simple_insert_txn( txn, table="receipts_graph", values={ "room_id": room_id, "receipt_type": receipt_type, "user_id": user_id, "event_ids": json_encoder.encode(event_ids), "data": json_encoder.encode(data), }, )