Reintroduce #17291.

synapse/storage/databases/main/event_federation.py

@@ -148,6 +148,10 @@ def __init__(
             500000, "_event_auth_cache", size_callback=len
         )
 
+        # Flag used by unit tests to disable fallback when there is no chain cover
+        # index.
+        self.tests_allow_no_chain_cover_index = True
+
         self._clock.looping_call(self._get_stats_for_federation_staging, 30 * 1000)
 
         if isinstance(self.database_engine, PostgresEngine):
@@ -220,8 +224,10 @@ async def get_auth_chain_ids(
                 )
             except _NoChainCoverIndex:
                 # For whatever reason we don't actually have a chain cover index
-                # for the events in question, so we fall back to the old method.
-                pass
+                # for the events in question, so we fall back to the old method
+                # (except in tests)
+                if not self.tests_allow_no_chain_cover_index:
+                    raise
 
         return await self.db_pool.runInteraction(
             "get_auth_chain_ids",
@@ -271,7 +277,7 @@ def _get_auth_chain_ids_using_cover_index_txn(
         if events_missing_chain_info:
             # This can happen due to e.g. downgrade/upgrade of the server. We
             # raise an exception and fall back to the previous algorithm.
-            logger.info(
+            logger.error(
                 "Unexpectedly found that events don't have chain IDs in room %s: %s",
                 room_id,
                 events_missing_chain_info,
@@ -482,8 +488,10 @@ async def get_auth_chain_difference(
                 )
             except _NoChainCoverIndex:
                 # For whatever reason we don't actually have a chain cover index
-                # for the events in question, so we fall back to the old method.
-                pass
+                # for the events in question, so we fall back to the old method
+                # (except in tests)
+                if not self.tests_allow_no_chain_cover_index:
+                    raise
 
         return await self.db_pool.runInteraction(
             "get_auth_chain_difference",
@@ -710,7 +718,7 @@ def _fixup_auth_chain_difference_sets(
         if events_missing_chain_info - event_to_auth_ids.keys():
             # Uh oh, we somehow haven't correctly done the chain cover index,
             # bail and fall back to the old method.
-            logger.info(
+            logger.error(
                 "Unexpectedly found that events don't have chain IDs in room %s: %s",
                 room_id,
                 events_missing_chain_info - event_to_auth_ids.keys(),

synapse/storage/databases/main/events.py

@@ -303,13 +303,13 @@ def calculate_chain_cover_index_for_events_txn(
             retcols=("room_id", "has_auth_chain_index"),
             allow_none=True,
         )
-        if row is None:
+        if row is None or row[1] is False:
             return {}
 
-        # Filter out already persisted events.
+        # Filter out events that we've already calculated.
         rows = self.db_pool.simple_select_many_txn(
             txn,
-            table="events",
+            table="event_auth_chains",
             column="event_id",
             iterable=[e.event_id for e in state_events],
             keyvalues={},
@@ -319,7 +319,7 @@ def calculate_chain_cover_index_for_events_txn(
         state_events = [
             event
             for event in state_events
-            if event.event_id in already_persisted_events
+            if event.event_id not in already_persisted_events
         ]
 
         if not state_events:
@@ -600,6 +600,9 @@ def _persist_event_auth_chain_txn(
         events: List[EventBase],
         new_event_links: Dict[str, NewEventChainLinks],
     ) -> None:
+        if new_event_links:
+            self._persist_chain_cover_index(txn, self.db_pool, new_event_links)
+
         # We only care about state events, so this if there are no state events.
         if not any(e.is_state() for e in events):
             return
@@ -622,9 +625,6 @@ def _persist_event_auth_chain_txn(
             ],
         )
 
-        if new_event_links:
-            self._persist_chain_cover_index(txn, self.db_pool, new_event_links)
-
     @classmethod
     def _add_chain_cover_index(
         cls,

tests/unittest.py

@@ -344,6 +344,8 @@ def setUp(self) -> None:
         self._hs_args = {"clock": self.clock, "reactor": self.reactor}
         self.hs = self.make_homeserver(self.reactor, self.clock)
 
+        self.hs.get_datastores().main.tests_allow_no_chain_cover_index = False
+
         # Honour the `use_frozen_dicts` config option. We have to do this
         # manually because this is taken care of in the app `start` code, which
         # we don't run. Plus we want to reset it on tearDown.