diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py
index e9ff6cfb34..90e6d82058 100644
--- a/synapse/storage/databases/main/events_worker.py
+++ b/synapse/storage/databases/main/events_worker.py
@@ -54,6 +54,7 @@ from synapse.logging.context import (
current_context,
make_deferred_yieldable,
)
+from synapse.logging.tracing import start_active_span, tag_args, trace
from synapse.metrics.background_process_metrics import (
run_as_background_process,
wrap_as_background_process,
@@ -430,6 +431,8 @@ class EventsWorkerStore(SQLBaseStore):
return {e.event_id: e for e in events}
+ @trace
+ @tag_args
async def get_events_as_list(
self,
event_ids: Collection[str],
@@ -1090,23 +1093,42 @@ class EventsWorkerStore(SQLBaseStore):
"""
fetched_event_ids: Set[str] = set()
fetched_events: Dict[str, _EventRow] = {}
- events_to_fetch = event_ids
- while events_to_fetch:
- row_map = await self._enqueue_events(events_to_fetch)
+ async def _fetch_event_ids_and_get_outstanding_redactions(
+ event_ids_to_fetch: Collection[str],
+ ) -> Collection[str]:
+ """
+ Fetch all of the given event_ids and return any associated redaction event_ids
+ that we still need to fetch in the next iteration.
+ """
+ row_map = await self._enqueue_events(event_ids_to_fetch)
# we need to recursively fetch any redactions of those events
redaction_ids: Set[str] = set()
- for event_id in events_to_fetch:
+ for event_id in event_ids_to_fetch:
row = row_map.get(event_id)
fetched_event_ids.add(event_id)
if row:
fetched_events[event_id] = row
redaction_ids.update(row.redactions)
- events_to_fetch = redaction_ids.difference(fetched_event_ids)
- if events_to_fetch:
- logger.debug("Also fetching redaction events %s", events_to_fetch)
+ event_ids_to_fetch = redaction_ids.difference(fetched_event_ids)
+ return event_ids_to_fetch
+
+ # Grab the initial list of events requested
+ event_ids_to_fetch = await _fetch_event_ids_and_get_outstanding_redactions(
+ event_ids
+ )
+ # Then go and recursively find all of the associated redactions
+ with start_active_span("recursively fetching redactions"):
+ while event_ids_to_fetch:
+ logger.debug("Also fetching redaction events %s", event_ids_to_fetch)
+
+ event_ids_to_fetch = (
+ await _fetch_event_ids_and_get_outstanding_redactions(
+ event_ids_to_fetch
+ )
+ )
# build a map from event_id to EventBase
event_map: Dict[str, EventBase] = {}
@@ -1424,6 +1446,8 @@ class EventsWorkerStore(SQLBaseStore):
return {r["event_id"] for r in rows}
+ @trace
+ @tag_args
async def have_seen_events(
self, room_id: str, event_ids: Iterable[str]
) -> Set[str]:
@@ -2200,3 +2224,63 @@ class EventsWorkerStore(SQLBaseStore):
(room_id,),
)
return [row[0] for row in txn]
+
+ def mark_event_rejected_txn(
+ self,
+ txn: LoggingTransaction,
+ event_id: str,
+ rejection_reason: Optional[str],
+ ) -> None:
+ """Mark an event that was previously accepted as rejected, or vice versa
+
+ This can happen, for example, when resyncing state during a faster join.
+
+ Args:
+ txn:
+ event_id: ID of event to update
+ rejection_reason: reason it has been rejected, or None if it is now accepted
+ """
+ if rejection_reason is None:
+ logger.info(
+ "Marking previously-processed event %s as accepted",
+ event_id,
+ )
+ self.db_pool.simple_delete_txn(
+ txn,
+ "rejections",
+ keyvalues={"event_id": event_id},
+ )
+ else:
+ logger.info(
+ "Marking previously-processed event %s as rejected(%s)",
+ event_id,
+ rejection_reason,
+ )
+ self.db_pool.simple_upsert_txn(
+ txn,
+ table="rejections",
+ keyvalues={"event_id": event_id},
+ values={
+ "reason": rejection_reason,
+ "last_check": self._clock.time_msec(),
+ },
+ )
+ self.db_pool.simple_update_txn(
+ txn,
+ table="events",
+ keyvalues={"event_id": event_id},
+ updatevalues={"rejection_reason": rejection_reason},
+ )
+
+ self.invalidate_get_event_cache_after_txn(txn, event_id)
+
+ # TODO(faster_joins): invalidate the cache on workers. Ideally we'd just
+ # call '_send_invalidation_to_replication', but we actually need the other
+ # end to call _invalidate_local_get_event_cache() rather than (just)
+ # _get_event_cache.invalidate().
+ #
+ # One solution might be to (somehow) get the workers to call
+ # _invalidate_caches_for_event() (though that will invalidate more than
+ # strictly necessary).
+ #
+ # https://github.com/matrix-org/synapse/issues/12994
|
