Use a chain cover index to efficiently calculate auth chain difference

synapse/storage/databases/main/room.py

@@ -1179,6 +1179,18 @@ async def upsert_room_on_join(self, room_id: str, room_version: RoomVersion):
         Called when we join a room over federation, and overwrites any room version
         currently in the table.
         """
+        # It's possible that we already have events for the room in our DB
+        # without a corresponding room entry. If we do then we don't want to
+        # mark the room as having an auth chain cover index.
+        max_ordering = await self.db_pool.simple_select_one_onecol(
+            table="events",
+            keyvalues={"room_id": room_id},
+            retcol="MAX(stream_ordering)",
+            allow_none=True,
+            desc="upsert_room_on_join",
+        )
+        has_auth_chain_index = max_ordering is None
+
         await self.db_pool.simple_upsert(
             desc="upsert_room_on_join",
             table="rooms",
@@ -1187,7 +1199,7 @@ async def upsert_room_on_join(self, room_id: str, room_version: RoomVersion):
             insertion_values={
                 "is_public": False,
                 "creator": "",
-                "has_auth_chain_index": True,
+                "has_auth_chain_index": has_auth_chain_index,
             },
             # rooms has a unique constraint on room_id, so no need to lock when doing an
             # emulated upsert.
@@ -1252,6 +1264,18 @@ async def maybe_store_room_on_outlier_membership(
         When we receive an invite or any other event over federation that may relate to a room
         we are not in, store the version of the room if we don't already know the room version.
         """
+        # It's possible that we already have events for the room in our DB
+        # without a corresponding room entry. If we do then we don't want to
+        # mark the room as having an auth chain cover index.
+        max_ordering = await self.db_pool.simple_select_one_onecol(
+            table="events",
+            keyvalues={"room_id": room_id},
+            retcol="MAX(stream_ordering)",
+            allow_none=True,
+            desc="maybe_store_room_on_outlier_membership",
+        )
+        has_auth_chain_index = max_ordering is None
+
         await self.db_pool.simple_upsert(
             desc="maybe_store_room_on_outlier_membership",
             table="rooms",
@@ -1261,7 +1285,7 @@ async def maybe_store_room_on_outlier_membership(
                 "room_version": room_version.identifier,
                 "is_public": False,
                 "creator": "",
-                "has_auth_chain_index": True,
+                "has_auth_chain_index": has_auth_chain_index,
             },
             # rooms has a unique constraint on room_id, so no need to lock when doing an
             # emulated upsert.