From e6e876b9b158f47811b6dfedd8783f658ce960a4 Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Wed, 12 Oct 2022 12:18:34 -0400 Subject: Return the thread ID properly down sync. (#14159) A receipt's thread ID, if one exists, should be added to the body of a receipt. --- synapse/storage/databases/main/receipts.py | 2 ++ 1 file changed, 2 insertions(+) (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index 246f78ac1f..b04026c21b 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -416,6 +416,8 @@ class ReceiptsWorkerStore(SQLBaseStore): # {"$foo:bar": { "read": { "@user:host": }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) + if row["thread_id"]: + receipt_type[row["user_id"]]["thread_id"] = row["thread_id"] receipt_type[row["user_id"]] = db_to_json(row["data"]) -- cgit 1.5.1 From 7d59a515bb97dc4f8253aa9a5a560221a0ef4702 Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Thu, 13 Oct 2022 12:15:41 -0400 Subject: Properly return the thread ID down sync. (#14159) Fix a broken conflict in e6e876b9b158f47811b6dfedd8783f658ce960a4, by not stomping over a field right after creating it. --- synapse/storage/databases/main/receipts.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index b04026c21b..dc6989527e 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -416,10 +416,10 @@ class ReceiptsWorkerStore(SQLBaseStore): # {"$foo:bar": { "read": { "@user:host": }, .. }, .. } event_entry = room_event["content"].setdefault(row["event_id"], {}) receipt_type = event_entry.setdefault(row["receipt_type"], {}) - if row["thread_id"]: - receipt_type[row["user_id"]]["thread_id"] = row["thread_id"] receipt_type[row["user_id"]] = db_to_json(row["data"]) + if row["thread_id"]: + receipt_type[row["user_id"]]["thread_id"] = row["thread_id"] results = { room_id: [results[room_id]] if room_id in results else [] -- cgit 1.5.1 From 36097e88c4da51fce6556a58c49bd675f4cf20ab Mon Sep 17 00:00:00 2001 From: Nick Mills-Barrett Date: Mon, 14 Nov 2022 17:31:36 +0000 Subject: Remove slaved id tracker (#14376) This matches the multi instance writer ID generator class which can both handle advancing the current token over replication and by calling the database. --- changelog.d/14376.misc | 1 + synapse/replication/slave/__init__.py | 13 ------ synapse/replication/slave/storage/__init__.py | 13 ------ .../slave/storage/_slaved_id_tracker.py | 50 ---------------------- synapse/storage/databases/main/account_data.py | 30 +++++-------- synapse/storage/databases/main/devices.py | 36 ++++++---------- synapse/storage/databases/main/events_worker.py | 35 ++++++--------- synapse/storage/databases/main/push_rule.py | 17 ++++---- synapse/storage/databases/main/pusher.py | 24 ++++------- synapse/storage/databases/main/receipts.py | 18 ++++---- synapse/storage/util/id_generators.py | 13 ++++-- 11 files changed, 74 insertions(+), 176 deletions(-) create mode 100644 changelog.d/14376.misc delete mode 100644 synapse/replication/slave/__init__.py delete mode 100644 synapse/replication/slave/storage/__init__.py delete mode 100644 synapse/replication/slave/storage/_slaved_id_tracker.py (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/changelog.d/14376.misc b/changelog.d/14376.misc new file mode 100644 index 0000000000..2ca326fea6 --- /dev/null +++ b/changelog.d/14376.misc @@ -0,0 +1 @@ +Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar). diff --git a/synapse/replication/slave/__init__.py b/synapse/replication/slave/__init__.py deleted file mode 100644 index f43a360a80..0000000000 --- a/synapse/replication/slave/__init__.py +++ /dev/null @@ -1,13 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. diff --git a/synapse/replication/slave/storage/__init__.py b/synapse/replication/slave/storage/__init__.py deleted file mode 100644 index f43a360a80..0000000000 --- a/synapse/replication/slave/storage/__init__.py +++ /dev/null @@ -1,13 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. diff --git a/synapse/replication/slave/storage/_slaved_id_tracker.py b/synapse/replication/slave/storage/_slaved_id_tracker.py deleted file mode 100644 index 8f3f953ed4..0000000000 --- a/synapse/replication/slave/storage/_slaved_id_tracker.py +++ /dev/null @@ -1,50 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -from typing import List, Optional, Tuple - -from synapse.storage.database import LoggingDatabaseConnection -from synapse.storage.util.id_generators import AbstractStreamIdTracker, _load_current_id - - -class SlavedIdTracker(AbstractStreamIdTracker): - """Tracks the "current" stream ID of a stream with a single writer. - - See `AbstractStreamIdTracker` for more details. - - Note that this class does not work correctly when there are multiple - writers. - """ - - def __init__( - self, - db_conn: LoggingDatabaseConnection, - table: str, - column: str, - extra_tables: Optional[List[Tuple[str, str]]] = None, - step: int = 1, - ): - self.step = step - self._current = _load_current_id(db_conn, table, column, step) - if extra_tables: - for table, column in extra_tables: - self.advance(None, _load_current_id(db_conn, table, column)) - - def advance(self, instance_name: Optional[str], new_id: int) -> None: - self._current = (max if self.step > 0 else min)(self._current, new_id) - - def get_current_token(self) -> int: - return self._current - - def get_current_token_for_writer(self, instance_name: str) -> int: - return self.get_current_token() diff --git a/synapse/storage/databases/main/account_data.py b/synapse/storage/databases/main/account_data.py index c38b8a9e5a..282687ebce 100644 --- a/synapse/storage/databases/main/account_data.py +++ b/synapse/storage/databases/main/account_data.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.api.constants import AccountDataTypes -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import AccountDataStream, TagAccountDataStream from synapse.storage._base import db_to_json from synapse.storage.database import ( @@ -68,12 +67,11 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # 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 + self._can_write_to_account_data = ( + self._instance_name in hs.config.worker.writers.account_data + ) 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, @@ -95,21 +93,13 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # `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")], - ) + self._account_data_id_gen = StreamIdGenerator( + db_conn, + "room_account_data", + "stream_id", + extra_tables=[("room_tags_revisions", "stream_id")], + is_writer=self._instance_name in hs.config.worker.writers.account_data, + ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py index aa58c2adc3..3e5c16b15b 100644 --- a/synapse/storage/databases/main/devices.py +++ b/synapse/storage/databases/main/devices.py @@ -38,7 +38,6 @@ from synapse.logging.opentracing import ( whitelisted_homeserver, ) from synapse.metrics.background_process_metrics import wrap_as_background_process -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams._base import DeviceListsStream, UserSignatureStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -86,28 +85,19 @@ class DeviceWorkerStore(RoomMemberWorkerStore, EndToEndKeyWorkerStore): ): super().__init__(database, db_conn, hs) - if hs.config.worker.worker_app is None: - self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - 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"), - ], - ) - else: - 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"), - ], - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + 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"), + ], + is_writer=hs.config.worker.worker_app is None, + ) # Type-ignore: _device_list_id_gen is mixed in from either DataStore (as a # StreamIdGenerator) or SlavedDataStore (as a SlavedIdTracker). diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index a79091952a..7a003ab88f 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -59,7 +59,6 @@ from synapse.metrics.background_process_metrics import ( run_as_background_process, wrap_as_background_process, ) -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import BackfillStream from synapse.replication.tcp.streams.events import EventsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause @@ -213,26 +212,20 @@ class EventsWorkerStore(SQLBaseStore): # `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.events: - 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: - self._stream_id_gen = SlavedIdTracker( - db_conn, "events", "stream_ordering" - ) - self._backfill_id_gen = SlavedIdTracker( - db_conn, "events", "stream_ordering", step=-1 - ) + self._stream_id_gen = StreamIdGenerator( + db_conn, + "events", + "stream_ordering", + is_writer=hs.get_instance_name() in hs.config.worker.writers.events, + ) + self._backfill_id_gen = StreamIdGenerator( + db_conn, + "events", + "stream_ordering", + step=-1, + extra_tables=[("ex_outlier_stream", "event_stream_ordering")], + is_writer=hs.get_instance_name() in hs.config.worker.writers.events, + ) events_max = self._stream_id_gen.get_current_token() curr_state_delta_prefill, min_curr_state_delta_id = self.db_pool.get_cache_dict( diff --git a/synapse/storage/databases/main/push_rule.py b/synapse/storage/databases/main/push_rule.py index 8ae10f6127..12ad44dbb3 100644 --- a/synapse/storage/databases/main/push_rule.py +++ b/synapse/storage/databases/main/push_rule.py @@ -30,7 +30,6 @@ from typing import ( from synapse.api.errors import StoreError from synapse.config.homeserver import ExperimentalConfig -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushRulesStream from synapse.storage._base import SQLBaseStore from synapse.storage.database import ( @@ -111,14 +110,14 @@ class PushRulesWorkerStore( ): 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" - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + db_conn, + "push_rules_stream", + "stream_id", + is_writer=hs.config.worker.worker_app is None, + ) push_rules_prefill, push_rules_id = self.db_pool.get_cache_dict( db_conn, diff --git a/synapse/storage/databases/main/pusher.py b/synapse/storage/databases/main/pusher.py index 4a01562d45..fee37b9ce4 100644 --- a/synapse/storage/databases/main/pusher.py +++ b/synapse/storage/databases/main/pusher.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.push import PusherConfig, ThrottleParams -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushersStream from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import ( @@ -59,20 +58,15 @@ class PusherWorkerStore(SQLBaseStore): ): super().__init__(database, db_conn, hs) - if hs.config.worker.worker_app is None: - self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - db_conn, - "pushers", - "id", - extra_tables=[("deleted_pushers", "stream_id")], - ) - else: - self._pushers_id_gen = SlavedIdTracker( - db_conn, - "pushers", - "id", - extra_tables=[("deleted_pushers", "stream_id")], - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + db_conn, + "pushers", + "id", + extra_tables=[("deleted_pushers", "stream_id")], + is_writer=hs.config.worker.worker_app is None, + ) self.db_pool.updates.register_background_update_handler( "remove_deactivated_pushers", diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index dc6989527e..64519587f8 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.api.constants import EduTypes -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import ReceiptsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -61,6 +60,9 @@ class ReceiptsWorkerStore(SQLBaseStore): hs: "HomeServer", ): self._instance_name = hs.get_instance_name() + + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. self._receipts_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): @@ -87,14 +89,12 @@ class ReceiptsWorkerStore(SQLBaseStore): # `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" - ) + self._receipts_id_gen = StreamIdGenerator( + db_conn, + "receipts_linearized", + "stream_id", + is_writer=hs.get_instance_name() in hs.config.worker.writers.receipts, + ) super().__init__(database, db_conn, hs) diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py index 2dfe4c0b66..1af0af1266 100644 --- a/synapse/storage/util/id_generators.py +++ b/synapse/storage/util/id_generators.py @@ -186,11 +186,13 @@ class StreamIdGenerator(AbstractStreamIdGenerator): column: str, extra_tables: Iterable[Tuple[str, str]] = (), step: int = 1, + is_writer: bool = True, ) -> None: assert step != 0 self._lock = threading.Lock() self._step: int = step self._current: int = _load_current_id(db_conn, table, column, step) + self._is_writer = is_writer for table, column in extra_tables: self._current = (max if step > 0 else min)( self._current, _load_current_id(db_conn, table, column, step) @@ -204,9 +206,11 @@ class StreamIdGenerator(AbstractStreamIdGenerator): self._unfinished_ids: OrderedDict[int, int] = OrderedDict() def advance(self, instance_name: str, new_id: int) -> None: - # `StreamIdGenerator` should only be used when there is a single writer, - # so replication should never happen. - raise Exception("Replication is not supported by StreamIdGenerator") + # Advance should never be called on a writer instance, only over replication + if self._is_writer: + raise Exception("Replication is not supported by writer StreamIdGenerator") + + self._current = (max if self._step > 0 else min)(self._current, new_id) def get_next(self) -> AsyncContextManager[int]: with self._lock: @@ -249,6 +253,9 @@ class StreamIdGenerator(AbstractStreamIdGenerator): return _AsyncCtxManagerWrapper(manager()) def get_current_token(self) -> int: + if self._is_writer: + return self._current + with self._lock: if self._unfinished_ids: return next(iter(self._unfinished_ids)) - self._step -- cgit 1.5.1 From d63814fd736fed5d3d45ff3af5e6d3bfae50c439 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Wed, 16 Nov 2022 13:50:07 +0000 Subject: Revert "Remove slaved id tracker (#14376)" (#14463) This reverts commit 36097e88c4da51fce6556a58c49bd675f4cf20ab. --- changelog.d/14376.misc | 1 - synapse/replication/slave/__init__.py | 13 ++++++ synapse/replication/slave/storage/__init__.py | 13 ++++++ .../slave/storage/_slaved_id_tracker.py | 50 ++++++++++++++++++++++ synapse/storage/databases/main/account_data.py | 30 ++++++++----- synapse/storage/databases/main/devices.py | 36 ++++++++++------ synapse/storage/databases/main/events_worker.py | 35 +++++++++------ synapse/storage/databases/main/push_rule.py | 17 ++++---- synapse/storage/databases/main/pusher.py | 24 +++++++---- synapse/storage/databases/main/receipts.py | 18 ++++---- synapse/storage/util/id_generators.py | 13 ++---- 11 files changed, 176 insertions(+), 74 deletions(-) delete mode 100644 changelog.d/14376.misc create mode 100644 synapse/replication/slave/__init__.py create mode 100644 synapse/replication/slave/storage/__init__.py create mode 100644 synapse/replication/slave/storage/_slaved_id_tracker.py (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/changelog.d/14376.misc b/changelog.d/14376.misc deleted file mode 100644 index 2ca326fea6..0000000000 --- a/changelog.d/14376.misc +++ /dev/null @@ -1 +0,0 @@ -Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar). diff --git a/synapse/replication/slave/__init__.py b/synapse/replication/slave/__init__.py new file mode 100644 index 0000000000..f43a360a80 --- /dev/null +++ b/synapse/replication/slave/__init__.py @@ -0,0 +1,13 @@ +# Copyright 2016 OpenMarket Ltd +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/synapse/replication/slave/storage/__init__.py b/synapse/replication/slave/storage/__init__.py new file mode 100644 index 0000000000..f43a360a80 --- /dev/null +++ b/synapse/replication/slave/storage/__init__.py @@ -0,0 +1,13 @@ +# Copyright 2016 OpenMarket Ltd +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/synapse/replication/slave/storage/_slaved_id_tracker.py b/synapse/replication/slave/storage/_slaved_id_tracker.py new file mode 100644 index 0000000000..8f3f953ed4 --- /dev/null +++ b/synapse/replication/slave/storage/_slaved_id_tracker.py @@ -0,0 +1,50 @@ +# Copyright 2016 OpenMarket Ltd +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from typing import List, Optional, Tuple + +from synapse.storage.database import LoggingDatabaseConnection +from synapse.storage.util.id_generators import AbstractStreamIdTracker, _load_current_id + + +class SlavedIdTracker(AbstractStreamIdTracker): + """Tracks the "current" stream ID of a stream with a single writer. + + See `AbstractStreamIdTracker` for more details. + + Note that this class does not work correctly when there are multiple + writers. + """ + + def __init__( + self, + db_conn: LoggingDatabaseConnection, + table: str, + column: str, + extra_tables: Optional[List[Tuple[str, str]]] = None, + step: int = 1, + ): + self.step = step + self._current = _load_current_id(db_conn, table, column, step) + if extra_tables: + for table, column in extra_tables: + self.advance(None, _load_current_id(db_conn, table, column)) + + def advance(self, instance_name: Optional[str], new_id: int) -> None: + self._current = (max if self.step > 0 else min)(self._current, new_id) + + def get_current_token(self) -> int: + return self._current + + def get_current_token_for_writer(self, instance_name: str) -> int: + return self.get_current_token() diff --git a/synapse/storage/databases/main/account_data.py b/synapse/storage/databases/main/account_data.py index 282687ebce..c38b8a9e5a 100644 --- a/synapse/storage/databases/main/account_data.py +++ b/synapse/storage/databases/main/account_data.py @@ -27,6 +27,7 @@ from typing import ( ) from synapse.api.constants import AccountDataTypes +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import AccountDataStream, TagAccountDataStream from synapse.storage._base import db_to_json from synapse.storage.database import ( @@ -67,11 +68,12 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # 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 - self._can_write_to_account_data = ( - self._instance_name in hs.config.worker.writers.account_data - ) 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, @@ -93,13 +95,21 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). - self._account_data_id_gen = StreamIdGenerator( - db_conn, - "room_account_data", - "stream_id", - extra_tables=[("room_tags_revisions", "stream_id")], - is_writer=self._instance_name in hs.config.worker.writers.account_data, - ) + 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( diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py index 3e5c16b15b..aa58c2adc3 100644 --- a/synapse/storage/databases/main/devices.py +++ b/synapse/storage/databases/main/devices.py @@ -38,6 +38,7 @@ from synapse.logging.opentracing import ( whitelisted_homeserver, ) from synapse.metrics.background_process_metrics import wrap_as_background_process +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams._base import DeviceListsStream, UserSignatureStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -85,19 +86,28 @@ class DeviceWorkerStore(RoomMemberWorkerStore, EndToEndKeyWorkerStore): ): super().__init__(database, db_conn, hs) - # In the worker store this is an ID tracker which we overwrite in the non-worker - # class below that is used on the main process. - self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - 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"), - ], - is_writer=hs.config.worker.worker_app is None, - ) + if hs.config.worker.worker_app is None: + self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + 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"), + ], + ) + else: + 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"), + ], + ) # Type-ignore: _device_list_id_gen is mixed in from either DataStore (as a # StreamIdGenerator) or SlavedDataStore (as a SlavedIdTracker). diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index 296e50d661..467d20253d 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -59,6 +59,7 @@ from synapse.metrics.background_process_metrics import ( run_as_background_process, wrap_as_background_process, ) +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import BackfillStream from synapse.replication.tcp.streams.events import EventsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause @@ -212,20 +213,26 @@ class EventsWorkerStore(SQLBaseStore): # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). - self._stream_id_gen = StreamIdGenerator( - db_conn, - "events", - "stream_ordering", - is_writer=hs.get_instance_name() in hs.config.worker.writers.events, - ) - self._backfill_id_gen = StreamIdGenerator( - db_conn, - "events", - "stream_ordering", - step=-1, - extra_tables=[("ex_outlier_stream", "event_stream_ordering")], - is_writer=hs.get_instance_name() in hs.config.worker.writers.events, - ) + if hs.get_instance_name() in hs.config.worker.writers.events: + 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: + self._stream_id_gen = SlavedIdTracker( + db_conn, "events", "stream_ordering" + ) + self._backfill_id_gen = SlavedIdTracker( + db_conn, "events", "stream_ordering", step=-1 + ) events_max = self._stream_id_gen.get_current_token() curr_state_delta_prefill, min_curr_state_delta_id = self.db_pool.get_cache_dict( diff --git a/synapse/storage/databases/main/push_rule.py b/synapse/storage/databases/main/push_rule.py index 12ad44dbb3..8ae10f6127 100644 --- a/synapse/storage/databases/main/push_rule.py +++ b/synapse/storage/databases/main/push_rule.py @@ -30,6 +30,7 @@ from typing import ( from synapse.api.errors import StoreError from synapse.config.homeserver import ExperimentalConfig +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushRulesStream from synapse.storage._base import SQLBaseStore from synapse.storage.database import ( @@ -110,14 +111,14 @@ class PushRulesWorkerStore( ): super().__init__(database, db_conn, hs) - # In the worker store this is an ID tracker which we overwrite in the non-worker - # class below that is used on the main process. - self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - db_conn, - "push_rules_stream", - "stream_id", - is_writer=hs.config.worker.worker_app is None, - ) + 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, diff --git a/synapse/storage/databases/main/pusher.py b/synapse/storage/databases/main/pusher.py index fee37b9ce4..4a01562d45 100644 --- a/synapse/storage/databases/main/pusher.py +++ b/synapse/storage/databases/main/pusher.py @@ -27,6 +27,7 @@ from typing import ( ) from synapse.push import PusherConfig, ThrottleParams +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushersStream from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import ( @@ -58,15 +59,20 @@ class PusherWorkerStore(SQLBaseStore): ): super().__init__(database, db_conn, hs) - # In the worker store this is an ID tracker which we overwrite in the non-worker - # class below that is used on the main process. - self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - db_conn, - "pushers", - "id", - extra_tables=[("deleted_pushers", "stream_id")], - is_writer=hs.config.worker.worker_app is None, - ) + if hs.config.worker.worker_app is None: + self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + db_conn, + "pushers", + "id", + extra_tables=[("deleted_pushers", "stream_id")], + ) + else: + self._pushers_id_gen = SlavedIdTracker( + db_conn, + "pushers", + "id", + extra_tables=[("deleted_pushers", "stream_id")], + ) self.db_pool.updates.register_background_update_handler( "remove_deactivated_pushers", diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index 64519587f8..dc6989527e 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -27,6 +27,7 @@ from typing import ( ) from synapse.api.constants import EduTypes +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import ReceiptsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -60,9 +61,6 @@ class ReceiptsWorkerStore(SQLBaseStore): hs: "HomeServer", ): self._instance_name = hs.get_instance_name() - - # In the worker store this is an ID tracker which we overwrite in the non-worker - # class below that is used on the main process. self._receipts_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): @@ -89,12 +87,14 @@ class ReceiptsWorkerStore(SQLBaseStore): # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets # updated over replication. (Multiple writers are not supported for # SQLite). - self._receipts_id_gen = StreamIdGenerator( - db_conn, - "receipts_linearized", - "stream_id", - is_writer=hs.get_instance_name() in hs.config.worker.writers.receipts, - ) + 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) diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py index 1af0af1266..2dfe4c0b66 100644 --- a/synapse/storage/util/id_generators.py +++ b/synapse/storage/util/id_generators.py @@ -186,13 +186,11 @@ class StreamIdGenerator(AbstractStreamIdGenerator): column: str, extra_tables: Iterable[Tuple[str, str]] = (), step: int = 1, - is_writer: bool = True, ) -> None: assert step != 0 self._lock = threading.Lock() self._step: int = step self._current: int = _load_current_id(db_conn, table, column, step) - self._is_writer = is_writer for table, column in extra_tables: self._current = (max if step > 0 else min)( self._current, _load_current_id(db_conn, table, column, step) @@ -206,11 +204,9 @@ class StreamIdGenerator(AbstractStreamIdGenerator): self._unfinished_ids: OrderedDict[int, int] = OrderedDict() def advance(self, instance_name: str, new_id: int) -> None: - # Advance should never be called on a writer instance, only over replication - if self._is_writer: - raise Exception("Replication is not supported by writer StreamIdGenerator") - - self._current = (max if self._step > 0 else min)(self._current, new_id) + # `StreamIdGenerator` should only be used when there is a single writer, + # so replication should never happen. + raise Exception("Replication is not supported by StreamIdGenerator") def get_next(self) -> AsyncContextManager[int]: with self._lock: @@ -253,9 +249,6 @@ class StreamIdGenerator(AbstractStreamIdGenerator): return _AsyncCtxManagerWrapper(manager()) def get_current_token(self) -> int: - if self._is_writer: - return self._current - with self._lock: if self._unfinished_ids: return next(iter(self._unfinished_ids)) - self._step -- cgit 1.5.1 From 882277008c7b43ab26e3445ab94a38aa25ad0965 Mon Sep 17 00:00:00 2001 From: Sean Quah <8349537+squahtx@users.noreply.github.com> Date: Wed, 16 Nov 2022 15:01:22 +0000 Subject: Fix background updates failing to add unique indexes on receipts (#14453) As part of the database migration to support threaded receipts, there is a possible window in between `73/08thread_receipts_non_null.sql.postgres` removing the original unique constraints on `receipts_linearized` and `receipts_graph` and the `reeipts_linearized_unique_index` and `receipts_graph_unique_index` background updates from `72/08thread_receipts.sql` completing where the unique constraints on `receipts_linearized` and `receipts_graph` are missing. Any emulated upserts on these tables must therefore be performed with a lock held, otherwise duplicate rows can end up in the tables when there are concurrent emulated upserts. Fix the missing lock. Note that emulated upserts no longer happen by default on sqlite, since the minimum supported version of sqlite supports native upserts by default now. Finally, clean up any duplicate receipts that may have crept in before trying to create the `receipts_graph_unique_index` and `receipts_linearized_unique_index` unique indexes. Signed-off-by: Sean Quah --- changelog.d/14453.bugfix | 1 + synapse/storage/databases/main/receipts.py | 171 ++++++++++++++++++--- tests/storage/databases/main/test_receipts.py | 209 ++++++++++++++++++++++++++ 3 files changed, 357 insertions(+), 24 deletions(-) create mode 100644 changelog.d/14453.bugfix create mode 100644 tests/storage/databases/main/test_receipts.py (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/changelog.d/14453.bugfix b/changelog.d/14453.bugfix new file mode 100644 index 0000000000..4969e5450c --- /dev/null +++ b/changelog.d/14453.bugfix @@ -0,0 +1 @@ +Fix a bug introduced in Synapse 1.70.0 where the background updates to add non-thread unique indexes on receipts could fail when upgrading from 1.67.0 or earlier. diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index dc6989527e..fbf27497ec 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -113,24 +113,6 @@ class ReceiptsWorkerStore(SQLBaseStore): prefilled_cache=receipts_stream_prefill, ) - self.db_pool.updates.register_background_index_update( - "receipts_linearized_unique_index", - index_name="receipts_linearized_unique_index", - table="receipts_linearized", - columns=["room_id", "receipt_type", "user_id"], - where_clause="thread_id IS NULL", - unique=True, - ) - - self.db_pool.updates.register_background_index_update( - "receipts_graph_unique_index", - index_name="receipts_graph_unique_index", - table="receipts_graph", - columns=["room_id", "receipt_type", "user_id"], - where_clause="thread_id IS NULL", - unique=True, - ) - 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() @@ -702,9 +684,6 @@ class ReceiptsWorkerStore(SQLBaseStore): "data": json_encoder.encode(data), }, where_clause=where_clause, - # receipts_linearized has a unique constraint on - # (user_id, room_id, receipt_type), so no need to lock - lock=False, ) return rx_ts @@ -862,14 +841,13 @@ class ReceiptsWorkerStore(SQLBaseStore): "data": json_encoder.encode(data), }, where_clause=where_clause, - # receipts_graph has a unique constraint on - # (user_id, room_id, receipt_type), so no need to lock - lock=False, ) class ReceiptsBackgroundUpdateStore(SQLBaseStore): POPULATE_RECEIPT_EVENT_STREAM_ORDERING = "populate_event_stream_ordering" + RECEIPTS_LINEARIZED_UNIQUE_INDEX_UPDATE_NAME = "receipts_linearized_unique_index" + RECEIPTS_GRAPH_UNIQUE_INDEX_UPDATE_NAME = "receipts_graph_unique_index" def __init__( self, @@ -883,6 +861,14 @@ class ReceiptsBackgroundUpdateStore(SQLBaseStore): self.POPULATE_RECEIPT_EVENT_STREAM_ORDERING, self._populate_receipt_event_stream_ordering, ) + self.db_pool.updates.register_background_update_handler( + self.RECEIPTS_LINEARIZED_UNIQUE_INDEX_UPDATE_NAME, + self._background_receipts_linearized_unique_index, + ) + self.db_pool.updates.register_background_update_handler( + self.RECEIPTS_GRAPH_UNIQUE_INDEX_UPDATE_NAME, + self._background_receipts_graph_unique_index, + ) async def _populate_receipt_event_stream_ordering( self, progress: JsonDict, batch_size: int @@ -938,6 +924,143 @@ class ReceiptsBackgroundUpdateStore(SQLBaseStore): return batch_size + async def _create_receipts_index(self, index_name: str, table: str) -> None: + """Adds a unique index on `(room_id, receipt_type, user_id)` to the given + receipts table, for non-thread receipts.""" + + def _create_index(conn: LoggingDatabaseConnection) -> None: + conn.rollback() + + # we have to set autocommit, because postgres refuses to + # CREATE INDEX CONCURRENTLY without it. + if isinstance(self.database_engine, PostgresEngine): + conn.set_session(autocommit=True) + + try: + c = conn.cursor() + + # Now that the duplicates are gone, we can create the index. + concurrently = ( + "CONCURRENTLY" + if isinstance(self.database_engine, PostgresEngine) + else "" + ) + sql = f""" + CREATE UNIQUE INDEX {concurrently} {index_name} + ON {table}(room_id, receipt_type, user_id) + WHERE thread_id IS NULL + """ + c.execute(sql) + finally: + if isinstance(self.database_engine, PostgresEngine): + conn.set_session(autocommit=False) + + await self.db_pool.runWithConnection(_create_index) + + async def _background_receipts_linearized_unique_index( + self, progress: dict, batch_size: int + ) -> int: + """Removes duplicate receipts and adds a unique index on + `(room_id, receipt_type, user_id)` to `receipts_linearized`, for non-thread + receipts.""" + + def _remote_duplicate_receipts_txn(txn: LoggingTransaction) -> None: + # Identify any duplicate receipts arising from + # https://github.com/matrix-org/synapse/issues/14406. + # We expect the following query to use the per-thread receipt index and take + # less than a minute. + sql = """ + SELECT MAX(stream_id), room_id, receipt_type, user_id + FROM receipts_linearized + WHERE thread_id IS NULL + GROUP BY room_id, receipt_type, user_id + HAVING COUNT(*) > 1 + """ + txn.execute(sql) + duplicate_keys = cast(List[Tuple[int, str, str, str]], list(txn)) + + # Then remove duplicate receipts, keeping the one with the highest + # `stream_id`. There should only be a single receipt with any given + # `stream_id`. + for max_stream_id, room_id, receipt_type, user_id in duplicate_keys: + sql = """ + DELETE FROM receipts_linearized + WHERE + room_id = ? AND + receipt_type = ? AND + user_id = ? AND + thread_id IS NULL AND + stream_id < ? + """ + txn.execute(sql, (room_id, receipt_type, user_id, max_stream_id)) + + await self.db_pool.runInteraction( + self.RECEIPTS_LINEARIZED_UNIQUE_INDEX_UPDATE_NAME, + _remote_duplicate_receipts_txn, + ) + + await self._create_receipts_index( + "receipts_linearized_unique_index", + "receipts_linearized", + ) + + await self.db_pool.updates._end_background_update( + self.RECEIPTS_LINEARIZED_UNIQUE_INDEX_UPDATE_NAME + ) + + return 1 + + async def _background_receipts_graph_unique_index( + self, progress: dict, batch_size: int + ) -> int: + """Removes duplicate receipts and adds a unique index on + `(room_id, receipt_type, user_id)` to `receipts_graph`, for non-thread + receipts.""" + + def _remote_duplicate_receipts_txn(txn: LoggingTransaction) -> None: + # Identify any duplicate receipts arising from + # https://github.com/matrix-org/synapse/issues/14406. + # We expect the following query to use the per-thread receipt index and take + # less than a minute. + sql = """ + SELECT room_id, receipt_type, user_id FROM receipts_graph + WHERE thread_id IS NULL + GROUP BY room_id, receipt_type, user_id + HAVING COUNT(*) > 1 + """ + txn.execute(sql) + duplicate_keys = cast(List[Tuple[str, str, str]], list(txn)) + + # Then remove all duplicate receipts. + # We could be clever and try to keep the latest receipt out of every set of + # duplicates, but it's far simpler to remove them all. + for room_id, receipt_type, user_id in duplicate_keys: + sql = """ + DELETE FROM receipts_graph + WHERE + room_id = ? AND + receipt_type = ? AND + user_id = ? AND + thread_id IS NULL + """ + txn.execute(sql, (room_id, receipt_type, user_id)) + + await self.db_pool.runInteraction( + self.RECEIPTS_GRAPH_UNIQUE_INDEX_UPDATE_NAME, + _remote_duplicate_receipts_txn, + ) + + await self._create_receipts_index( + "receipts_graph_unique_index", + "receipts_graph", + ) + + await self.db_pool.updates._end_background_update( + self.RECEIPTS_GRAPH_UNIQUE_INDEX_UPDATE_NAME + ) + + return 1 + class ReceiptsStore(ReceiptsWorkerStore, ReceiptsBackgroundUpdateStore): pass diff --git a/tests/storage/databases/main/test_receipts.py b/tests/storage/databases/main/test_receipts.py new file mode 100644 index 0000000000..c4f12d81d7 --- /dev/null +++ b/tests/storage/databases/main/test_receipts.py @@ -0,0 +1,209 @@ +# Copyright 2022 The Matrix.org Foundation C.I.C. +# +# Licensed under the Apache License, Version 2.0 (the 'License'); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an 'AS IS' BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from typing import Any, Dict, Optional, Sequence, Tuple + +from twisted.test.proto_helpers import MemoryReactor + +from synapse.rest import admin +from synapse.rest.client import login, room +from synapse.server import HomeServer +from synapse.storage.database import LoggingTransaction +from synapse.util import Clock + +from tests.unittest import HomeserverTestCase + + +class ReceiptsBackgroundUpdateStoreTestCase(HomeserverTestCase): + + servlets = [ + admin.register_servlets, + room.register_servlets, + login.register_servlets, + ] + + def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer): + self.store = hs.get_datastores().main + self.user_id = self.register_user("foo", "pass") + self.token = self.login("foo", "pass") + self.room_id = self.helper.create_room_as(self.user_id, tok=self.token) + self.other_room_id = self.helper.create_room_as(self.user_id, tok=self.token) + + def _test_background_receipts_unique_index( + self, + update_name: str, + index_name: str, + table: str, + receipts: Dict[Tuple[str, str, str], Sequence[Dict[str, Any]]], + expected_unique_receipts: Dict[Tuple[str, str, str], Optional[Dict[str, Any]]], + ): + """Test that the background update to uniqueify non-thread receipts in + the given receipts table works properly. + + Args: + update_name: The name of the background update to test. + index_name: The name of the index that the background update creates. + table: The table of receipts that the background update fixes. + receipts: The test data containing duplicate receipts. + A list of receipt rows to insert, grouped by + `(room_id, receipt_type, user_id)`. + expected_unique_receipts: A dictionary of `(room_id, receipt_type, user_id)` + keys and expected receipt key-values after duplicate receipts have been + removed. + """ + # First, undo the background update. + def drop_receipts_unique_index(txn: LoggingTransaction) -> None: + txn.execute(f"DROP INDEX IF EXISTS {index_name}") + + self.get_success( + self.store.db_pool.runInteraction( + "drop_receipts_unique_index", + drop_receipts_unique_index, + ) + ) + + # Populate the receipts table, including duplicates. + for (room_id, receipt_type, user_id), rows in receipts.items(): + for row in rows: + self.get_success( + self.store.db_pool.simple_insert( + table, + { + "room_id": room_id, + "receipt_type": receipt_type, + "user_id": user_id, + "thread_id": None, + "data": "{}", + **row, + }, + ) + ) + + # Insert and run the background update. + self.get_success( + self.store.db_pool.simple_insert( + "background_updates", + { + "update_name": update_name, + "progress_json": "{}", + }, + ) + ) + + self.store.db_pool.updates._all_done = False + + self.wait_for_background_updates() + + # Check that the remaining receipts match expectations. + for ( + room_id, + receipt_type, + user_id, + ), expected_row in expected_unique_receipts.items(): + # Include the receipt key in the returned columns, for more informative + # assertion messages. + columns = ["room_id", "receipt_type", "user_id"] + if expected_row is not None: + columns += expected_row.keys() + + rows = self.get_success( + self.store.db_pool.simple_select_list( + table=table, + keyvalues={ + "room_id": room_id, + "receipt_type": receipt_type, + "user_id": user_id, + # `simple_select_onecol` does not support NULL filters, + # so skip the filter on `thread_id`. + }, + retcols=columns, + desc="get_receipt", + ) + ) + + if expected_row is not None: + self.assertEqual( + len(rows), + 1, + f"Background update did not leave behind latest receipt in {table}", + ) + self.assertEqual( + rows[0], + { + "room_id": room_id, + "receipt_type": receipt_type, + "user_id": user_id, + **expected_row, + }, + ) + else: + self.assertEqual( + len(rows), + 0, + f"Background update did not remove all duplicate receipts from {table}", + ) + + def test_background_receipts_linearized_unique_index(self): + """Test that the background update to uniqueify non-thread receipts in + `receipts_linearized` works properly. + """ + self._test_background_receipts_unique_index( + "receipts_linearized_unique_index", + "receipts_linearized_unique_index", + "receipts_linearized", + receipts={ + (self.room_id, "m.read", self.user_id): [ + {"stream_id": 5, "event_id": "$some_event"}, + {"stream_id": 6, "event_id": "$some_event"}, + ], + (self.other_room_id, "m.read", self.user_id): [ + {"stream_id": 7, "event_id": "$some_event"} + ], + }, + expected_unique_receipts={ + (self.room_id, "m.read", self.user_id): {"stream_id": 6}, + (self.other_room_id, "m.read", self.user_id): {"stream_id": 7}, + }, + ) + + def test_background_receipts_graph_unique_index(self): + """Test that the background update to uniqueify non-thread receipts in + `receipts_graph` works properly. + """ + self._test_background_receipts_unique_index( + "receipts_graph_unique_index", + "receipts_graph_unique_index", + "receipts_graph", + receipts={ + (self.room_id, "m.read", self.user_id): [ + { + "event_ids": '["$some_event"]', + }, + { + "event_ids": '["$some_event"]', + }, + ], + (self.other_room_id, "m.read", self.user_id): [ + { + "event_ids": '["$some_event"]', + } + ], + }, + expected_unique_receipts={ + (self.room_id, "m.read", self.user_id): None, + (self.other_room_id, "m.read", self.user_id): { + "event_ids": '["$some_event"]' + }, + }, + ) -- cgit 1.5.1 From 115f0eb2334b13665e5c112bd87f95ea393c9047 Mon Sep 17 00:00:00 2001 From: David Robertson Date: Wed, 16 Nov 2022 22:16:46 +0000 Subject: Reintroduce #14376, with bugfix for monoliths (#14468) * Add tests for StreamIdGenerator * Drive-by: annotate all defs * Revert "Revert "Remove slaved id tracker (#14376)" (#14463)" This reverts commit d63814fd736fed5d3d45ff3af5e6d3bfae50c439, which in turn reverted 36097e88c4da51fce6556a58c49bd675f4cf20ab. This restores the latter. * Fix StreamIdGenerator not handling unpersisted IDs Spotted by @erikjohnston. Closes #14456. * Changelog Co-authored-by: Nick Mills-Barrett Co-authored-by: Erik Johnston --- changelog.d/14376.misc | 1 + changelog.d/14468.misc | 1 + mypy.ini | 3 + synapse/replication/slave/__init__.py | 13 -- synapse/replication/slave/storage/__init__.py | 13 -- .../slave/storage/_slaved_id_tracker.py | 50 ------- synapse/storage/databases/main/account_data.py | 30 ++-- synapse/storage/databases/main/devices.py | 36 ++--- synapse/storage/databases/main/events_worker.py | 35 ++--- synapse/storage/databases/main/push_rule.py | 17 +-- synapse/storage/databases/main/pusher.py | 24 ++- synapse/storage/databases/main/receipts.py | 18 +-- synapse/storage/util/id_generators.py | 13 +- tests/storage/test_id_generators.py | 162 +++++++++++++++++++-- 14 files changed, 230 insertions(+), 186 deletions(-) create mode 100644 changelog.d/14376.misc create mode 100644 changelog.d/14468.misc delete mode 100644 synapse/replication/slave/__init__.py delete mode 100644 synapse/replication/slave/storage/__init__.py delete mode 100644 synapse/replication/slave/storage/_slaved_id_tracker.py (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/changelog.d/14376.misc b/changelog.d/14376.misc new file mode 100644 index 0000000000..2ca326fea6 --- /dev/null +++ b/changelog.d/14376.misc @@ -0,0 +1 @@ +Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar). diff --git a/changelog.d/14468.misc b/changelog.d/14468.misc new file mode 100644 index 0000000000..2ca326fea6 --- /dev/null +++ b/changelog.d/14468.misc @@ -0,0 +1 @@ +Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar). diff --git a/mypy.ini b/mypy.ini index 8f1141a239..53512b2584 100644 --- a/mypy.ini +++ b/mypy.ini @@ -117,6 +117,9 @@ disallow_untyped_defs = True [mypy-tests.state.test_profile] disallow_untyped_defs = True +[mypy-tests.storage.test_id_generators] +disallow_untyped_defs = True + [mypy-tests.storage.test_profile] disallow_untyped_defs = True diff --git a/synapse/replication/slave/__init__.py b/synapse/replication/slave/__init__.py deleted file mode 100644 index f43a360a80..0000000000 --- a/synapse/replication/slave/__init__.py +++ /dev/null @@ -1,13 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. diff --git a/synapse/replication/slave/storage/__init__.py b/synapse/replication/slave/storage/__init__.py deleted file mode 100644 index f43a360a80..0000000000 --- a/synapse/replication/slave/storage/__init__.py +++ /dev/null @@ -1,13 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. diff --git a/synapse/replication/slave/storage/_slaved_id_tracker.py b/synapse/replication/slave/storage/_slaved_id_tracker.py deleted file mode 100644 index 8f3f953ed4..0000000000 --- a/synapse/replication/slave/storage/_slaved_id_tracker.py +++ /dev/null @@ -1,50 +0,0 @@ -# Copyright 2016 OpenMarket Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -from typing import List, Optional, Tuple - -from synapse.storage.database import LoggingDatabaseConnection -from synapse.storage.util.id_generators import AbstractStreamIdTracker, _load_current_id - - -class SlavedIdTracker(AbstractStreamIdTracker): - """Tracks the "current" stream ID of a stream with a single writer. - - See `AbstractStreamIdTracker` for more details. - - Note that this class does not work correctly when there are multiple - writers. - """ - - def __init__( - self, - db_conn: LoggingDatabaseConnection, - table: str, - column: str, - extra_tables: Optional[List[Tuple[str, str]]] = None, - step: int = 1, - ): - self.step = step - self._current = _load_current_id(db_conn, table, column, step) - if extra_tables: - for table, column in extra_tables: - self.advance(None, _load_current_id(db_conn, table, column)) - - def advance(self, instance_name: Optional[str], new_id: int) -> None: - self._current = (max if self.step > 0 else min)(self._current, new_id) - - def get_current_token(self) -> int: - return self._current - - def get_current_token_for_writer(self, instance_name: str) -> int: - return self.get_current_token() diff --git a/synapse/storage/databases/main/account_data.py b/synapse/storage/databases/main/account_data.py index c38b8a9e5a..282687ebce 100644 --- a/synapse/storage/databases/main/account_data.py +++ b/synapse/storage/databases/main/account_data.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.api.constants import AccountDataTypes -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import AccountDataStream, TagAccountDataStream from synapse.storage._base import db_to_json from synapse.storage.database import ( @@ -68,12 +67,11 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # 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 + self._can_write_to_account_data = ( + self._instance_name in hs.config.worker.writers.account_data + ) 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, @@ -95,21 +93,13 @@ class AccountDataWorkerStore(PushRulesWorkerStore, CacheInvalidationWorkerStore) # `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")], - ) + self._account_data_id_gen = StreamIdGenerator( + db_conn, + "room_account_data", + "stream_id", + extra_tables=[("room_tags_revisions", "stream_id")], + is_writer=self._instance_name in hs.config.worker.writers.account_data, + ) account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py index e114c733d1..57230df5ae 100644 --- a/synapse/storage/databases/main/devices.py +++ b/synapse/storage/databases/main/devices.py @@ -38,7 +38,6 @@ from synapse.logging.opentracing import ( whitelisted_homeserver, ) from synapse.metrics.background_process_metrics import wrap_as_background_process -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams._base import DeviceListsStream, UserSignatureStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -86,28 +85,19 @@ class DeviceWorkerStore(RoomMemberWorkerStore, EndToEndKeyWorkerStore): ): super().__init__(database, db_conn, hs) - if hs.config.worker.worker_app is None: - self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - 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"), - ], - ) - else: - 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"), - ], - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._device_list_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + 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"), + ], + is_writer=hs.config.worker.worker_app is None, + ) # Type-ignore: _device_list_id_gen is mixed in from either DataStore (as a # StreamIdGenerator) or SlavedDataStore (as a SlavedIdTracker). diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index 8a104f7e93..01e935edef 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -59,7 +59,6 @@ from synapse.metrics.background_process_metrics import ( run_as_background_process, wrap_as_background_process, ) -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import BackfillStream from synapse.replication.tcp.streams.events import EventsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause @@ -213,26 +212,20 @@ class EventsWorkerStore(SQLBaseStore): # `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.events: - 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: - self._stream_id_gen = SlavedIdTracker( - db_conn, "events", "stream_ordering" - ) - self._backfill_id_gen = SlavedIdTracker( - db_conn, "events", "stream_ordering", step=-1 - ) + self._stream_id_gen = StreamIdGenerator( + db_conn, + "events", + "stream_ordering", + is_writer=hs.get_instance_name() in hs.config.worker.writers.events, + ) + self._backfill_id_gen = StreamIdGenerator( + db_conn, + "events", + "stream_ordering", + step=-1, + extra_tables=[("ex_outlier_stream", "event_stream_ordering")], + is_writer=hs.get_instance_name() in hs.config.worker.writers.events, + ) events_max = self._stream_id_gen.get_current_token() curr_state_delta_prefill, min_curr_state_delta_id = self.db_pool.get_cache_dict( diff --git a/synapse/storage/databases/main/push_rule.py b/synapse/storage/databases/main/push_rule.py index 8ae10f6127..12ad44dbb3 100644 --- a/synapse/storage/databases/main/push_rule.py +++ b/synapse/storage/databases/main/push_rule.py @@ -30,7 +30,6 @@ from typing import ( from synapse.api.errors import StoreError from synapse.config.homeserver import ExperimentalConfig -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushRulesStream from synapse.storage._base import SQLBaseStore from synapse.storage.database import ( @@ -111,14 +110,14 @@ class PushRulesWorkerStore( ): 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" - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._push_rules_stream_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + db_conn, + "push_rules_stream", + "stream_id", + is_writer=hs.config.worker.worker_app is None, + ) push_rules_prefill, push_rules_id = self.db_pool.get_cache_dict( db_conn, diff --git a/synapse/storage/databases/main/pusher.py b/synapse/storage/databases/main/pusher.py index 4a01562d45..fee37b9ce4 100644 --- a/synapse/storage/databases/main/pusher.py +++ b/synapse/storage/databases/main/pusher.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.push import PusherConfig, ThrottleParams -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import PushersStream from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import ( @@ -59,20 +58,15 @@ class PusherWorkerStore(SQLBaseStore): ): super().__init__(database, db_conn, hs) - if hs.config.worker.worker_app is None: - self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( - db_conn, - "pushers", - "id", - extra_tables=[("deleted_pushers", "stream_id")], - ) - else: - self._pushers_id_gen = SlavedIdTracker( - db_conn, - "pushers", - "id", - extra_tables=[("deleted_pushers", "stream_id")], - ) + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. + self._pushers_id_gen: AbstractStreamIdTracker = StreamIdGenerator( + db_conn, + "pushers", + "id", + extra_tables=[("deleted_pushers", "stream_id")], + is_writer=hs.config.worker.worker_app is None, + ) self.db_pool.updates.register_background_update_handler( "remove_deactivated_pushers", diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index fbf27497ec..a580e4bdda 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -27,7 +27,6 @@ from typing import ( ) from synapse.api.constants import EduTypes -from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker from synapse.replication.tcp.streams import ReceiptsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import ( @@ -61,6 +60,9 @@ class ReceiptsWorkerStore(SQLBaseStore): hs: "HomeServer", ): self._instance_name = hs.get_instance_name() + + # In the worker store this is an ID tracker which we overwrite in the non-worker + # class below that is used on the main process. self._receipts_id_gen: AbstractStreamIdTracker if isinstance(database.engine, PostgresEngine): @@ -87,14 +89,12 @@ class ReceiptsWorkerStore(SQLBaseStore): # `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" - ) + self._receipts_id_gen = StreamIdGenerator( + db_conn, + "receipts_linearized", + "stream_id", + is_writer=hs.get_instance_name() in hs.config.worker.writers.receipts, + ) super().__init__(database, db_conn, hs) diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py index 2dfe4c0b66..0d7108f01b 100644 --- a/synapse/storage/util/id_generators.py +++ b/synapse/storage/util/id_generators.py @@ -186,11 +186,13 @@ class StreamIdGenerator(AbstractStreamIdGenerator): column: str, extra_tables: Iterable[Tuple[str, str]] = (), step: int = 1, + is_writer: bool = True, ) -> None: assert step != 0 self._lock = threading.Lock() self._step: int = step self._current: int = _load_current_id(db_conn, table, column, step) + self._is_writer = is_writer for table, column in extra_tables: self._current = (max if step > 0 else min)( self._current, _load_current_id(db_conn, table, column, step) @@ -204,9 +206,11 @@ class StreamIdGenerator(AbstractStreamIdGenerator): self._unfinished_ids: OrderedDict[int, int] = OrderedDict() def advance(self, instance_name: str, new_id: int) -> None: - # `StreamIdGenerator` should only be used when there is a single writer, - # so replication should never happen. - raise Exception("Replication is not supported by StreamIdGenerator") + # Advance should never be called on a writer instance, only over replication + if self._is_writer: + raise Exception("Replication is not supported by writer StreamIdGenerator") + + self._current = (max if self._step > 0 else min)(self._current, new_id) def get_next(self) -> AsyncContextManager[int]: with self._lock: @@ -249,6 +253,9 @@ class StreamIdGenerator(AbstractStreamIdGenerator): return _AsyncCtxManagerWrapper(manager()) def get_current_token(self) -> int: + if not self._is_writer: + return self._current + with self._lock: if self._unfinished_ids: return next(iter(self._unfinished_ids)) - self._step diff --git a/tests/storage/test_id_generators.py b/tests/storage/test_id_generators.py index 2d8d1f860f..d6a2b8d274 100644 --- a/tests/storage/test_id_generators.py +++ b/tests/storage/test_id_generators.py @@ -16,15 +16,157 @@ from typing import List, Optional from twisted.test.proto_helpers import MemoryReactor from synapse.server import HomeServer -from synapse.storage.database import DatabasePool, LoggingTransaction +from synapse.storage.database import ( + DatabasePool, + LoggingDatabaseConnection, + LoggingTransaction, +) from synapse.storage.engines import IncorrectDatabaseSetup -from synapse.storage.util.id_generators import MultiWriterIdGenerator +from synapse.storage.types import Cursor +from synapse.storage.util.id_generators import MultiWriterIdGenerator, StreamIdGenerator from synapse.util import Clock from tests.unittest import HomeserverTestCase from tests.utils import USE_POSTGRES_FOR_TESTS +class StreamIdGeneratorTestCase(HomeserverTestCase): + def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None: + self.store = hs.get_datastores().main + self.db_pool: DatabasePool = self.store.db_pool + + self.get_success(self.db_pool.runInteraction("_setup_db", self._setup_db)) + + def _setup_db(self, txn: LoggingTransaction) -> None: + txn.execute( + """ + CREATE TABLE foobar ( + stream_id BIGINT NOT NULL, + data TEXT + ); + """ + ) + txn.execute("INSERT INTO foobar VALUES (123, 'hello world');") + + def _create_id_generator(self) -> StreamIdGenerator: + def _create(conn: LoggingDatabaseConnection) -> StreamIdGenerator: + return StreamIdGenerator( + db_conn=conn, + table="foobar", + column="stream_id", + ) + + return self.get_success_or_raise(self.db_pool.runWithConnection(_create)) + + def test_initial_value(self) -> None: + """Check that we read the current token from the DB.""" + id_gen = self._create_id_generator() + self.assertEqual(id_gen.get_current_token(), 123) + + def test_single_gen_next(self) -> None: + """Check that we correctly increment the current token from the DB.""" + id_gen = self._create_id_generator() + + async def test_gen_next() -> None: + async with id_gen.get_next() as next_id: + # We haven't persisted `next_id` yet; current token is still 123 + self.assertEqual(id_gen.get_current_token(), 123) + # But we did learn what the next value is + self.assertEqual(next_id, 124) + + # Once the context manager closes we assume that the `next_id` has been + # written to the DB. + self.assertEqual(id_gen.get_current_token(), 124) + + self.get_success(test_gen_next()) + + def test_multiple_gen_nexts(self) -> None: + """Check that we handle overlapping calls to gen_next sensibly.""" + id_gen = self._create_id_generator() + + async def test_gen_next() -> None: + ctx1 = id_gen.get_next() + ctx2 = id_gen.get_next() + ctx3 = id_gen.get_next() + + # Request three new stream IDs. + self.assertEqual(await ctx1.__aenter__(), 124) + self.assertEqual(await ctx2.__aenter__(), 125) + self.assertEqual(await ctx3.__aenter__(), 126) + + # None are persisted: current token unchanged. + self.assertEqual(id_gen.get_current_token(), 123) + + # Persist each in turn. + await ctx1.__aexit__(None, None, None) + self.assertEqual(id_gen.get_current_token(), 124) + await ctx2.__aexit__(None, None, None) + self.assertEqual(id_gen.get_current_token(), 125) + await ctx3.__aexit__(None, None, None) + self.assertEqual(id_gen.get_current_token(), 126) + + self.get_success(test_gen_next()) + + def test_multiple_gen_nexts_closed_in_different_order(self) -> None: + """Check that we handle overlapping calls to gen_next, even when their IDs + created and persisted in different orders.""" + id_gen = self._create_id_generator() + + async def test_gen_next() -> None: + ctx1 = id_gen.get_next() + ctx2 = id_gen.get_next() + ctx3 = id_gen.get_next() + + # Request three new stream IDs. + self.assertEqual(await ctx1.__aenter__(), 124) + self.assertEqual(await ctx2.__aenter__(), 125) + self.assertEqual(await ctx3.__aenter__(), 126) + + # None are persisted: current token unchanged. + self.assertEqual(id_gen.get_current_token(), 123) + + # Persist them in a different order, starting with 126 from ctx3. + await ctx3.__aexit__(None, None, None) + # We haven't persisted 124 from ctx1 yet---current token is still 123. + self.assertEqual(id_gen.get_current_token(), 123) + + # Now persist 124 from ctx1. + await ctx1.__aexit__(None, None, None) + # Current token is then 124, waiting for 125 to be persisted. + self.assertEqual(id_gen.get_current_token(), 124) + + # Finally persist 125 from ctx2. + await ctx2.__aexit__(None, None, None) + # Current token is then 126 (skipping over 125). + self.assertEqual(id_gen.get_current_token(), 126) + + self.get_success(test_gen_next()) + + def test_gen_next_while_still_waiting_for_persistence(self) -> None: + """Check that we handle overlapping calls to gen_next.""" + id_gen = self._create_id_generator() + + async def test_gen_next() -> None: + ctx1 = id_gen.get_next() + ctx2 = id_gen.get_next() + ctx3 = id_gen.get_next() + + # Request two new stream IDs. + self.assertEqual(await ctx1.__aenter__(), 124) + self.assertEqual(await ctx2.__aenter__(), 125) + + # Persist ctx2 first. + await ctx2.__aexit__(None, None, None) + # Still waiting on ctx1's ID to be persisted. + self.assertEqual(id_gen.get_current_token(), 123) + + # Now request a third stream ID. It should be 126 (the smallest ID that + # we've not yet handed out.) + self.assertEqual(await ctx3.__aenter__(), 126) + + self.get_success(test_gen_next()) + + class MultiWriterIdGeneratorTestCase(HomeserverTestCase): if not USE_POSTGRES_FOR_TESTS: skip = "Requires Postgres" @@ -48,9 +190,9 @@ class MultiWriterIdGeneratorTestCase(HomeserverTestCase): ) def _create_id_generator( - self, instance_name="master", writers: Optional[List[str]] = None + self, instance_name: str = "master", writers: Optional[List[str]] = None ) -> MultiWriterIdGenerator: - def _create(conn): + def _create(conn: LoggingDatabaseConnection) -> MultiWriterIdGenerator: return MultiWriterIdGenerator( conn, self.db_pool, @@ -446,7 +588,7 @@ class MultiWriterIdGeneratorTestCase(HomeserverTestCase): self._insert_row_with_id("master", 3) # Now we add a row *without* updating the stream ID - def _insert(txn): + def _insert(txn: Cursor) -> None: txn.execute("INSERT INTO foobar VALUES (26, 'master')") self.get_success(self.db_pool.runInteraction("_insert", _insert)) @@ -481,9 +623,9 @@ class BackwardsMultiWriterIdGeneratorTestCase(HomeserverTestCase): ) def _create_id_generator( - self, instance_name="master", writers: Optional[List[str]] = None + self, instance_name: str = "master", writers: Optional[List[str]] = None ) -> MultiWriterIdGenerator: - def _create(conn): + def _create(conn: LoggingDatabaseConnection) -> MultiWriterIdGenerator: return MultiWriterIdGenerator( conn, self.db_pool, @@ -617,9 +759,9 @@ class MultiTableMultiWriterIdGeneratorTestCase(HomeserverTestCase): ) def _create_id_generator( - self, instance_name="master", writers: Optional[List[str]] = None + self, instance_name: str = "master", writers: Optional[List[str]] = None ) -> MultiWriterIdGenerator: - def _create(conn): + def _create(conn: LoggingDatabaseConnection) -> MultiWriterIdGenerator: return MultiWriterIdGenerator( conn, self.db_pool, @@ -641,7 +783,7 @@ class MultiTableMultiWriterIdGeneratorTestCase(HomeserverTestCase): instance_name: str, number: int, update_stream_table: bool = True, - ): + ) -> None: """Insert N rows as the given instance, inserting with stream IDs pulled from the postgres sequence. """ -- cgit 1.5.1 From 373c485d8c7f39206bac60c6ef313b4a1978bbc0 Mon Sep 17 00:00:00 2001 From: Sean Quah <8349537+squahtx@users.noreply.github.com> Date: Fri, 9 Dec 2022 23:02:11 +0000 Subject: Handle half-created indices in receipts index background update (#14650) When Synapse is terminated while running the background update to create the `receipts_graph` or `receipts_linearized` indexes, the indexes may be successfully created (or marked as invalid on postgres) while the background update remains unfinished. When Synapse next starts up, the background update will fail because the index already exists, or exists but is invalid on postgres. Use the existing code to create indices in background updates, since it handles these edge cases. Signed-off-by: Sean Quah --- changelog.d/14650.bugfix | 2 ++ synapse/storage/background_updates.py | 55 +++++++++++++++++++++++++----- synapse/storage/databases/main/receipts.py | 51 +++++++-------------------- 3 files changed, 60 insertions(+), 48 deletions(-) create mode 100644 changelog.d/14650.bugfix (limited to 'synapse/storage/databases/main/receipts.py') diff --git a/changelog.d/14650.bugfix b/changelog.d/14650.bugfix new file mode 100644 index 0000000000..5e18641bf7 --- /dev/null +++ b/changelog.d/14650.bugfix @@ -0,0 +1,2 @@ +Fix a bug introduced in Synapse 1.72.0 where the background updates to add non-thread unique indexes on receipts would fail if they were previously interrupted. + diff --git a/synapse/storage/background_updates.py b/synapse/storage/background_updates.py index 2056ecb2c3..a99aea8926 100644 --- a/synapse/storage/background_updates.py +++ b/synapse/storage/background_updates.py @@ -544,6 +544,48 @@ class BackgroundUpdater: The named index will be dropped upon completion of the new index. """ + async def updater(progress: JsonDict, batch_size: int) -> int: + await self.create_index_in_background( + index_name=index_name, + table=table, + columns=columns, + where_clause=where_clause, + unique=unique, + psql_only=psql_only, + replaces_index=replaces_index, + ) + await self._end_background_update(update_name) + return 1 + + self._background_update_handlers[update_name] = _BackgroundUpdateHandler( + updater, oneshot=True + ) + + async def create_index_in_background( + self, + index_name: str, + table: str, + columns: Iterable[str], + where_clause: Optional[str] = None, + unique: bool = False, + psql_only: bool = False, + replaces_index: Optional[str] = None, + ) -> None: + """Add an index in the background. + + Args: + update_name: update_name to register for + index_name: name of index to add + table: table to add index to + columns: columns/expressions to include in index + where_clause: A WHERE clause to specify a partial unique index. + unique: true to make a UNIQUE index + psql_only: true to only create this index on psql databases (useful + for virtual sqlite tables) + replaces_index: The name of an index that this index replaces. + The named index will be dropped upon completion of the new index. + """ + def create_index_psql(conn: Connection) -> None: conn.rollback() # postgres insists on autocommit for the index @@ -618,16 +660,11 @@ class BackgroundUpdater: else: runner = create_index_sqlite - async def updater(progress: JsonDict, batch_size: int) -> int: - if runner is not None: - logger.info("Adding index %s to %s", index_name, table) - await self.db_pool.runWithConnection(runner) - await self._end_background_update(update_name) - return 1 + if runner is None: + return - self._background_update_handlers[update_name] = _BackgroundUpdateHandler( - updater, oneshot=True - ) + logger.info("Adding index %s to %s", index_name, table) + await self.db_pool.runWithConnection(runner) async def _end_background_update(self, update_name: str) -> None: """Removes a completed background update task from the queue. diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index a580e4bdda..e06725f69c 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -924,39 +924,6 @@ class ReceiptsBackgroundUpdateStore(SQLBaseStore): return batch_size - async def _create_receipts_index(self, index_name: str, table: str) -> None: - """Adds a unique index on `(room_id, receipt_type, user_id)` to the given - receipts table, for non-thread receipts.""" - - def _create_index(conn: LoggingDatabaseConnection) -> None: - conn.rollback() - - # we have to set autocommit, because postgres refuses to - # CREATE INDEX CONCURRENTLY without it. - if isinstance(self.database_engine, PostgresEngine): - conn.set_session(autocommit=True) - - try: - c = conn.cursor() - - # Now that the duplicates are gone, we can create the index. - concurrently = ( - "CONCURRENTLY" - if isinstance(self.database_engine, PostgresEngine) - else "" - ) - sql = f""" - CREATE UNIQUE INDEX {concurrently} {index_name} - ON {table}(room_id, receipt_type, user_id) - WHERE thread_id IS NULL - """ - c.execute(sql) - finally: - if isinstance(self.database_engine, PostgresEngine): - conn.set_session(autocommit=False) - - await self.db_pool.runWithConnection(_create_index) - async def _background_receipts_linearized_unique_index( self, progress: dict, batch_size: int ) -> int: @@ -999,9 +966,12 @@ class ReceiptsBackgroundUpdateStore(SQLBaseStore): _remote_duplicate_receipts_txn, ) - await self._create_receipts_index( - "receipts_linearized_unique_index", - "receipts_linearized", + await self.db_pool.updates.create_index_in_background( + index_name="receipts_linearized_unique_index", + table="receipts_linearized", + columns=["room_id", "receipt_type", "user_id"], + where_clause="thread_id IS NULL", + unique=True, ) await self.db_pool.updates._end_background_update( @@ -1050,9 +1020,12 @@ class ReceiptsBackgroundUpdateStore(SQLBaseStore): _remote_duplicate_receipts_txn, ) - await self._create_receipts_index( - "receipts_graph_unique_index", - "receipts_graph", + await self.db_pool.updates.create_index_in_background( + index_name="receipts_graph_unique_index", + table="receipts_graph", + columns=["room_id", "receipt_type", "user_id"], + where_clause="thread_id IS NULL", + unique=True, ) await self.db_pool.updates._end_background_update( -- cgit 1.5.1