Update the rejected state of events during resync (#13459)

Events can be un-rejected or newly-rejected during resync, so ensure we update
the database and caches when that happens.

4 files changed, 66 insertions, 9 deletions

diff --git a/changelog.d/13459.misc b/changelog.d/13459.misc
new file mode 100644
index 0000000000..e6082210a0
--- /dev/null
+++ b/changelog.d/13459.misc
@@ -0,0 +1 @@
+Faster joins: update the rejected state of events during de-partial-stating.
diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py
index e9ff6cfb34..b07d812ae2 100644
--- a/synapse/storage/databases/main/events_worker.py
+++ b/synapse/storage/databases/main/events_worker.py
@@ -2200,3 +2200,63 @@ 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
diff --git a/synapse/storage/databases/main/state.py b/synapse/storage/databases/main/state.py
index f70705a0af..0b10af0e58 100644
--- a/synapse/storage/databases/main/state.py
+++ b/synapse/storage/databases/main/state.py
@@ -430,6 +430,11 @@ class StateGroupWorkerStore(EventsWorkerStore, SQLBaseStore):
             updatevalues={"state_group": state_group},
         )
 
+        # the event may now be rejected where it was not before, or vice versa,
+        # in which case we need to update the rejected flags.
+        if bool(context.rejected) != (event.rejected_reason is not None):
+            self.mark_event_rejected_txn(txn, event.event_id, context.rejected)
+
         self.db_pool.simple_delete_one_txn(
             txn,
             table="partial_state_events",
diff --git a/synapse/storage/state.py b/synapse/storage/state.py
index af3bab2c15..0004d955b4 100644
--- a/synapse/storage/state.py
+++ b/synapse/storage/state.py
@@ -539,15 +539,6 @@ class StateFilter:
             is_mine_id: a callable which confirms if a given state_key matches a mxid
                of a local user
         """
-
-        # TODO(faster_joins): it's not entirely clear that this is safe. In particular,
-        #  there may be circumstances in which we return a piece of state that, once we
-        #  resync the state, we discover is invalid. For example: if it turns out that
-        #  the sender of a piece of state wasn't actually in the room, then clearly that
-        #  state shouldn't have been returned.
-        #  We should at least add some tests around this to see what happens.
-        #  https://github.com/matrix-org/synapse/issues/13006
-
         # if we haven't requested membership events, then it depends on the value of
         # 'include_others'
         if EventTypes.Member not in self.types: