Matrix-Mirrors/anonymousland-synapse
1
0
Fork 0
mirror of https://git.anonymousland.org/anonymousland/synapse.git synced 2025-05-07 16:35:05 -04:00
Code Issues Projects Releases Packages Wiki Activity

Add experimental support for sharding event persister. (#8170)

Browse source 
This is *not* ready for production yet. Caveats:

1. We should write some tests...
2. The stream token that we use for events can get stalled at the minimum position of all writers. This means that new events may not be processed and e.g. sent down sync streams if a writer isn't writing or is slow.
This commit is contained in:
Erik Johnston 2020-09-02 15:48:37 +01:00 • committed by GitHub
parent b257c788c0
commit 82c1ee1c22
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
18 changed files with 206 additions and 77 deletions
Download patch file Download diff file Expand all files Collapse all files

66
synapse/storage/databases/main/events_worker.py
View file

@ -42,7 +42,8 @@ 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
from synapse.storage.database import DatabasePool
from synapse.storage.util.id_generators import StreamIdGenerator
from synapse.storage.engines import PostgresEngine
from synapse.storage.util.id_generators import MultiWriterIdGenerator, StreamIdGenerator
from synapse.types import Collection, get_domain_from_id
from synapse.util.caches.descriptors import Cache, cached
from synapse.util.iterutils import batch_iter
@ -78,27 +79,54 @@ class EventsWorkerStore(SQLBaseStore):
def __init__(self, database: DatabasePool, db_conn, hs):
super(EventsWorkerStore, self).__init__(database, db_conn, hs)
if hs.config.worker.writers.events == hs.get_instance_name():
# We are the process in charge of generating stream ids for events,
# so instantiate ID generators based on the database
self._stream_id_gen = StreamIdGenerator(
db_conn, "events", "stream_ordering",
if isinstance(database.engine, PostgresEngine):
# If we're using Postgres than we can use `MultiWriterIdGenerator`
# regardless of whether this process writes to the streams or not.
self._stream_id_gen = MultiWriterIdGenerator(
db_conn=db_conn,
db=database,
instance_name=hs.get_instance_name(),
table="events",
instance_column="instance_name",
id_column="stream_ordering",
sequence_name="events_stream_seq",
)
self._backfill_id_gen = StreamIdGenerator(
db_conn,
"events",
"stream_ordering",
step=-1,
extra_tables=[("ex_outlier_stream", "event_stream_ordering")],
self._backfill_id_gen = MultiWriterIdGenerator(
db_conn=db_conn,
db=database,
instance_name=hs.get_instance_name(),
table="events",
instance_column="instance_name",
id_column="stream_ordering",
sequence_name="events_backfill_stream_seq",
positive=False,
)
else:
# Another process is in charge of persisting events and generating
# stream IDs: rely on the replication streams to let us know which
# IDs we can process.
self._stream_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering")
self._backfill_id_gen = SlavedIdTracker(
db_conn, "events", "stream_ordering", step=-1
)
# We shouldn't be running in worker mode with SQLite, but its useful
# to support it for unit tests.
#
# If this process is the writer than we need to use
# `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets
# updated over replication. (Multiple writers are not supported for
# SQLite).
if hs.get_instance_name() in hs.config.worker.writers.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._get_event_cache = Cache(
"*getEvent*",