diff --git a/synapse/storage/databases/main/__init__.py b/synapse/storage/databases/main/__init__.py
index ccb3384db9..0cb12f4c61 100644
--- a/synapse/storage/databases/main/__init__.py
+++ b/synapse/storage/databases/main/__init__.py
@@ -160,14 +160,20 @@ class DataStore(
)
if isinstance(self.database_engine, PostgresEngine):
+ # We set the `writers` to an empty list here as we don't care about
+ # missing updates over restarts, as we'll not have anything in our
+ # caches to invalidate. (This reduces the amount of writes to the DB
+ # that happen).
self._cache_id_gen = MultiWriterIdGenerator(
db_conn,
database,
- instance_name="master",
+ stream_name="caches",
+ instance_name=hs.get_instance_name(),
table="cache_invalidation_stream_by_instance",
instance_column="instance_name",
id_column="stream_id",
sequence_name="cache_invalidation_stream_seq",
+ writers=[],
)
else:
self._cache_id_gen = None
diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py
index de9e8d1dc6..f95679ebc4 100644
--- a/synapse/storage/databases/main/events_worker.py
+++ b/synapse/storage/databases/main/events_worker.py
@@ -83,21 +83,25 @@ class EventsWorkerStore(SQLBaseStore):
self._stream_id_gen = MultiWriterIdGenerator(
db_conn=db_conn,
db=database,
+ stream_name="events",
instance_name=hs.get_instance_name(),
table="events",
instance_column="instance_name",
id_column="stream_ordering",
sequence_name="events_stream_seq",
+ writers=hs.config.worker.writers.events,
)
self._backfill_id_gen = MultiWriterIdGenerator(
db_conn=db_conn,
db=database,
+ stream_name="backfill",
instance_name=hs.get_instance_name(),
table="events",
instance_column="instance_name",
id_column="stream_ordering",
sequence_name="events_backfill_stream_seq",
positive=False,
+ writers=hs.config.worker.writers.events,
)
else:
# We shouldn't be running in worker mode with SQLite, but its useful
diff --git a/synapse/storage/databases/main/registration.py b/synapse/storage/databases/main/registration.py
index a06451b7f0..2ed696cc14 100644
--- a/synapse/storage/databases/main/registration.py
+++ b/synapse/storage/databases/main/registration.py
@@ -472,7 +472,7 @@ class RegistrationWorkerStore(SQLBaseStore):
async def get_user_by_external_id(
self, auth_provider: str, external_id: str
- ) -> str:
+ ) -> Optional[str]:
"""Look up a user by their external auth id
Args:
@@ -480,7 +480,7 @@ class RegistrationWorkerStore(SQLBaseStore):
external_id: id on that system
Returns:
- str|None: the mxid of the user, or None if they are not known
+ the mxid of the user, or None if they are not known
"""
return await self.db_pool.simple_select_one_onecol(
table="user_external_ids",
diff --git a/synapse/storage/databases/main/roommember.py b/synapse/storage/databases/main/roommember.py
index 4fa8767b01..86ffe2479e 100644
--- a/synapse/storage/databases/main/roommember.py
+++ b/synapse/storage/databases/main/roommember.py
@@ -13,7 +13,6 @@
# 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.
-
import logging
from typing import TYPE_CHECKING, Dict, FrozenSet, Iterable, List, Optional, Set
@@ -37,7 +36,7 @@ from synapse.storage.roommember import (
ProfileInfo,
RoomsForUser,
)
-from synapse.types import Collection, get_domain_from_id
+from synapse.types import Collection, PersistedEventPosition, get_domain_from_id
from synapse.util.async_helpers import Linearizer
from synapse.util.caches import intern_string
from synapse.util.caches.descriptors import _CacheContext, cached, cachedList
@@ -387,7 +386,7 @@ class RoomMemberWorkerStore(EventsWorkerStore):
# for rooms the server is participating in.
if self._current_state_events_membership_up_to_date:
sql = """
- SELECT room_id, e.stream_ordering
+ SELECT room_id, e.instance_name, e.stream_ordering
FROM current_state_events AS c
INNER JOIN events AS e USING (room_id, event_id)
WHERE
@@ -397,7 +396,7 @@ class RoomMemberWorkerStore(EventsWorkerStore):
"""
else:
sql = """
- SELECT room_id, e.stream_ordering
+ SELECT room_id, e.instance_name, e.stream_ordering
FROM current_state_events AS c
INNER JOIN room_memberships AS m USING (room_id, event_id)
INNER JOIN events AS e USING (room_id, event_id)
@@ -408,7 +407,12 @@ class RoomMemberWorkerStore(EventsWorkerStore):
"""
txn.execute(sql, (user_id, Membership.JOIN))
- return frozenset(GetRoomsForUserWithStreamOrdering(*row) for row in txn)
+ return frozenset(
+ GetRoomsForUserWithStreamOrdering(
+ room_id, PersistedEventPosition(instance, stream_id)
+ )
+ for room_id, instance, stream_id in txn
+ )
async def get_users_server_still_shares_room_with(
self, user_ids: Collection[str]
diff --git a/synapse/storage/databases/main/schema/delta/58/14events_instance_name.sql.postgres b/synapse/storage/databases/main/schema/delta/58/14events_instance_name.sql.postgres
index 97c1e6a0c5..c31f9af82a 100644
--- a/synapse/storage/databases/main/schema/delta/58/14events_instance_name.sql.postgres
+++ b/synapse/storage/databases/main/schema/delta/58/14events_instance_name.sql.postgres
@@ -21,6 +21,8 @@ SELECT setval('events_stream_seq', (
CREATE SEQUENCE IF NOT EXISTS events_backfill_stream_seq;
+-- If the server has never backfilled a room then doing `-MIN(...)` will give
+-- a negative result, hence why we do `GREATEST(...)`
SELECT setval('events_backfill_stream_seq', (
- SELECT COALESCE(-MIN(stream_ordering), 1) FROM events
+ SELECT GREATEST(COALESCE(-MIN(stream_ordering), 1), 1) FROM events
));
diff --git a/synapse/storage/databases/main/schema/delta/58/18stream_positions.sql b/synapse/storage/databases/main/schema/delta/58/18stream_positions.sql
new file mode 100644
index 0000000000..985fd949a2
--- /dev/null
+++ b/synapse/storage/databases/main/schema/delta/58/18stream_positions.sql
@@ -0,0 +1,22 @@
+/* Copyright 2020 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.
+ */
+
+CREATE TABLE stream_positions (
+ stream_name TEXT NOT NULL,
+ instance_name TEXT NOT NULL,
+ stream_id BIGINT NOT NULL
+);
+
+CREATE UNIQUE INDEX stream_positions_idx ON stream_positions(stream_name, instance_name);
diff --git a/synapse/storage/persist_events.py b/synapse/storage/persist_events.py
index d89f6ed128..603cd7d825 100644
--- a/synapse/storage/persist_events.py
+++ b/synapse/storage/persist_events.py
@@ -31,7 +31,7 @@ from synapse.logging.context import PreserveLoggingContext, make_deferred_yielda
from synapse.metrics.background_process_metrics import run_as_background_process
from synapse.storage.databases import Databases
from synapse.storage.databases.main.events import DeltaState
-from synapse.types import Collection, StateMap
+from synapse.types import Collection, PersistedEventPosition, RoomStreamToken, StateMap
from synapse.util.async_helpers import ObservableDeferred
from synapse.util.metrics import Measure
@@ -190,6 +190,7 @@ class EventsPersistenceStorage:
self.persist_events_store = stores.persist_events
self._clock = hs.get_clock()
+ self._instance_name = hs.get_instance_name()
self.is_mine_id = hs.is_mine_id
self._event_persist_queue = _EventPeristenceQueue()
self._state_resolution_handler = hs.get_state_resolution_handler()
@@ -198,7 +199,7 @@ class EventsPersistenceStorage:
self,
events_and_contexts: List[Tuple[EventBase, EventContext]],
backfilled: bool = False,
- ) -> int:
+ ) -> RoomStreamToken:
"""
Write events to the database
Args:
@@ -228,11 +229,11 @@ class EventsPersistenceStorage:
defer.gatherResults(deferreds, consumeErrors=True)
)
- return self.main_store.get_current_events_token()
+ return RoomStreamToken(None, self.main_store.get_current_events_token())
async def persist_event(
self, event: EventBase, context: EventContext, backfilled: bool = False
- ) -> Tuple[int, int]:
+ ) -> Tuple[PersistedEventPosition, RoomStreamToken]:
"""
Returns:
The stream ordering of `event`, and the stream ordering of the
@@ -247,7 +248,10 @@ class EventsPersistenceStorage:
await make_deferred_yieldable(deferred)
max_persisted_id = self.main_store.get_current_events_token()
- return (event.internal_metadata.stream_ordering, max_persisted_id)
+ event_stream_id = event.internal_metadata.stream_ordering
+
+ pos = PersistedEventPosition(self._instance_name, event_stream_id)
+ return pos, RoomStreamToken(None, max_persisted_id)
def _maybe_start_persisting(self, room_id: str):
async def persisting_queue(item):
diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py
index 8c4a83a840..f152f63321 100644
--- a/synapse/storage/roommember.py
+++ b/synapse/storage/roommember.py
@@ -25,7 +25,7 @@ RoomsForUser = namedtuple(
)
GetRoomsForUserWithStreamOrdering = namedtuple(
- "_GetRoomsForUserWithStreamOrdering", ("room_id", "stream_ordering")
+ "_GetRoomsForUserWithStreamOrdering", ("room_id", "event_pos")
)
diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py
index b0353ac2dc..4269eaf918 100644
--- a/synapse/storage/util/id_generators.py
+++ b/synapse/storage/util/id_generators.py
@@ -22,6 +22,7 @@ from typing import Dict, List, Optional, Set, Union
import attr
from typing_extensions import Deque
+from synapse.metrics.background_process_metrics import run_as_background_process
from synapse.storage.database import DatabasePool, LoggingTransaction
from synapse.storage.util.sequence import PostgresSequenceGenerator
@@ -184,12 +185,16 @@ class MultiWriterIdGenerator:
Args:
db_conn
db
+ stream_name: A name for the stream.
instance_name: The name of this instance.
table: Database table associated with stream.
instance_column: Column that stores the row's writer's instance name
id_column: Column that stores the stream ID.
sequence_name: The name of the postgres sequence used to generate new
IDs.
+ writers: A list of known writers to use to populate current positions
+ on startup. Can be empty if nothing uses `get_current_token` or
+ `get_positions` (e.g. caches stream).
positive: Whether the IDs are positive (true) or negative (false).
When using negative IDs we go backwards from -1 to -2, -3, etc.
"""
@@ -198,16 +203,20 @@ class MultiWriterIdGenerator:
self,
db_conn,
db: DatabasePool,
+ stream_name: str,
instance_name: str,
table: str,
instance_column: str,
id_column: str,
sequence_name: str,
+ writers: List[str],
positive: bool = True,
):
self._db = db
+ self._stream_name = stream_name
self._instance_name = instance_name
self._positive = positive
+ self._writers = writers
self._return_factor = 1 if positive else -1
# We lock as some functions may be called from DB threads.
@@ -216,9 +225,7 @@ class MultiWriterIdGenerator:
# Note: If we are a negative stream then we still store all the IDs as
# positive to make life easier for us, and simply negate the IDs when we
# return them.
- self._current_positions = self._load_current_ids(
- db_conn, table, instance_column, id_column
- )
+ self._current_positions = {} # type: Dict[str, int]
# Set of local IDs that we're still processing. The current position
# should be less than the minimum of this set (if not empty).
@@ -251,30 +258,84 @@ class MultiWriterIdGenerator:
self._sequence_gen = PostgresSequenceGenerator(sequence_name)
+ # This goes and fills out the above state from the database.
+ self._load_current_ids(db_conn, table, instance_column, id_column)
+
def _load_current_ids(
self, db_conn, table: str, instance_column: str, id_column: str
- ) -> Dict[str, int]:
- # If positive stream aggregate via MAX. For negative stream use MIN
- # *and* negate the result to get a positive number.
- sql = """
- SELECT %(instance)s, %(agg)s(%(id)s) FROM %(table)s
- GROUP BY %(instance)s
- """ % {
- "instance": instance_column,
- "id": id_column,
- "table": table,
- "agg": "MAX" if self._positive else "-MIN",
- }
-
+ ):
cur = db_conn.cursor()
- cur.execute(sql)
- # `cur` is an iterable over returned rows, which are 2-tuples.
- current_positions = dict(cur)
+ # Load the current positions of all writers for the stream.
+ if self._writers:
+ sql = """
+ SELECT instance_name, stream_id FROM stream_positions
+ WHERE stream_name = ?
+ """
+ sql = self._db.engine.convert_param_style(sql)
- cur.close()
+ cur.execute(sql, (self._stream_name,))
+
+ self._current_positions = {
+ instance: stream_id * self._return_factor
+ for instance, stream_id in cur
+ if instance in self._writers
+ }
+
+ # We set the `_persisted_upto_position` to be the minimum of all current
+ # positions. If empty we use the max stream ID from the DB table.
+ min_stream_id = min(self._current_positions.values(), default=None)
+
+ if min_stream_id is None:
+ # We add a GREATEST here to ensure that the result is always
+ # positive. (This can be a problem for e.g. backfill streams where
+ # the server has never backfilled).
+ sql = """
+ SELECT GREATEST(COALESCE(%(agg)s(%(id)s), 1), 1)
+ FROM %(table)s
+ """ % {
+ "id": id_column,
+ "table": table,
+ "agg": "MAX" if self._positive else "-MIN",
+ }
+ cur.execute(sql)
+ (stream_id,) = cur.fetchone()
+ self._persisted_upto_position = stream_id
+ else:
+ # If we have a min_stream_id then we pull out everything greater
+ # than it from the DB so that we can prefill
+ # `_known_persisted_positions` and get a more accurate
+ # `_persisted_upto_position`.
+ #
+ # We also check if any of the later rows are from this instance, in
+ # which case we use that for this instance's current position. This
+ # is to handle the case where we didn't finish persisting to the
+ # stream positions table before restart (or the stream position
+ # table otherwise got out of date).
+
+ sql = """
+ SELECT %(instance)s, %(id)s FROM %(table)s
+ WHERE ? %(cmp)s %(id)s
+ """ % {
+ "id": id_column,
+ "table": table,
+ "instance": instance_column,
+ "cmp": "<=" if self._positive else ">=",
+ }
+ sql = self._db.engine.convert_param_style(sql)
+ cur.execute(sql, (min_stream_id,))
+
+ self._persisted_upto_position = min_stream_id
+
+ with self._lock:
+ for (instance, stream_id,) in cur:
+ stream_id = self._return_factor * stream_id
+ self._add_persisted_position(stream_id)
- return current_positions
+ if instance == self._instance_name:
+ self._current_positions[instance] = stream_id
+
+ cur.close()
def _load_next_id_txn(self, txn) -> int:
return self._sequence_gen.get_next_id_txn(txn)
@@ -316,6 +377,21 @@ class MultiWriterIdGenerator:
txn.call_after(self._mark_id_as_finished, next_id)
txn.call_on_exception(self._mark_id_as_finished, next_id)
+ # Update the `stream_positions` table with newly updated stream
+ # ID (unless self._writers is not set in which case we don't
+ # bother, as nothing will read it).
+ #
+ # We only do this on the success path so that the persisted current
+ # position points to a persited row with the correct instance name.
+ if self._writers:
+ txn.call_after(
+ run_as_background_process,
+ "MultiWriterIdGenerator._update_table",
+ self._db.runInteraction,
+ "MultiWriterIdGenerator._update_table",
+ self._update_stream_positions_table_txn,
+ )
+
return self._return_factor * next_id
def _mark_id_as_finished(self, next_id: int):
@@ -447,6 +523,28 @@ class MultiWriterIdGenerator:
# do.
break
+ def _update_stream_positions_table_txn(self, txn):
+ """Update the `stream_positions` table with newly persisted position.
+ """
+
+ if not self._writers:
+ return
+
+ # We upsert the value, ensuring on conflict that we always increase the
+ # value (or decrease if stream goes backwards).
+ sql = """
+ INSERT INTO stream_positions (stream_name, instance_name, stream_id)
+ VALUES (?, ?, ?)
+ ON CONFLICT (stream_name, instance_name)
+ DO UPDATE SET
+ stream_id = %(agg)s(stream_positions.stream_id, EXCLUDED.stream_id)
+ """ % {
+ "agg": "GREATEST" if self._positive else "LEAST",
+ }
+
+ pos = (self.get_current_token_for_writer(self._instance_name),)
+ txn.execute(sql, (self._stream_name, self._instance_name, pos))
+
@attr.s(slots=True)
class _AsyncCtxManagerWrapper:
@@ -503,4 +601,16 @@ class _MultiWriterCtxManager:
if exc_type is not None:
return False
+ # Update the `stream_positions` table with newly updated stream
+ # ID (unless self._writers is not set in which case we don't
+ # bother, as nothing will read it).
+ #
+ # We only do this on the success path so that the persisted current
+ # position points to a persisted row with the correct instance name.
+ if self.id_gen._writers:
+ await self.id_gen._db.runInteraction(
+ "MultiWriterIdGenerator._update_table",
+ self.id_gen._update_stream_positions_table_txn,
+ )
+
return False
|
