OpenCloud
/
MatrixSynapse
mirror of https://github.com/matrix-org/synapse
1
0
Fork 0
Code Issues Releases Wiki Activity

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 d63814fd73, which in
turn reverted 36097e88c4. This restores
the latter.

* Fix StreamIdGenerator not handling unpersisted IDs

Spotted by @erikjohnston.

Closes #14456.

* Changelog

Co-authored-by: Nick Mills-Barrett <nick@fizzadar.com>
Co-authored-by: Erik Johnston <erik@matrix.org>
pull/14476/head
David Robertson 2022-11-16 22:16:46 +00:00 committed by GitHub
parent c15e9a0edb
commit 115f0eb233
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
14 changed files with 230 additions and 186 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
changelog.d/14376.misc Normal file
View File

@ -0,0 +1 @@
Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar).

1
changelog.d/14468.misc Normal file
View File

@ -0,0 +1 @@
Remove old stream ID tracking code. Contributed by Nick @Beeper (@fizzadar).

3
mypy.ini
View File

@ -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

13
synapse/replication/slave/__init__.py
View File

@ -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.

13
synapse/replication/slave/storage/__init__.py
View File

@ -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.

50
synapse/replication/slave/storage/_slaved_id_tracker.py
View File

@ -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()

30
synapse/storage/databases/main/account_data.py
View File

@ -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(

36
synapse/storage/databases/main/devices.py
View File

@ -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).

35
synapse/storage/databases/main/events_worker.py
View File

@ -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(

17
synapse/storage/databases/main/push_rule.py
View File

@ -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,

24
synapse/storage/databases/main/pusher.py
View File

@ -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",

18
synapse/storage/databases/main/receipts.py
View File

@ -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)

13
synapse/storage/util/id_generators.py
View File

@ -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

162
tests/storage/test_id_generators.py
View File

@ -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.
"""