Improve get auth chain difference algorithm. (#7095)

It was originally implemented by pulling the full auth chain of all
state sets out of the database and doing set comparison. However, that
can take a lot work if the state and auth chains are large.

Instead, lets try and fetch the auth chains at the same time and
calculate the difference on the fly, allowing us to bail early if all
the auth chains converge. Assuming that the auth chains do converge more
often than not, this should improve performance. Hopefully.

synapse/state/__init__.py

@@ -662,28 +662,16 @@ class StateResolutionStore(object):
             allow_rejected=allow_rejected,
         )
 
-    def get_auth_chain(self, event_ids: List[str], ignore_events: Set[str]):
-        """Gets the full auth chain for a set of events (including rejected
-        events).
-
-        Includes the given event IDs in the result.
-
-        Note that:
-            1. All events must be state events.
-            2. For v1 rooms this may not have the full auth chain in the
-               presence of rejected events
-
-        Args:
-            event_ids: The event IDs of the events to fetch the auth chain for.
-                Must be state events.
-            ignore_events: Set of events to exclude from the returned auth
-                chain.
+    def get_auth_chain_difference(self, state_sets: List[Set[str]]):
+        """Given sets of state events figure out the auth chain difference (as
+        per state res v2 algorithm).
 
+        This equivalent to fetching the full auth chain for each set of state
+        and returning the events that don't appear in each and every auth
+        chain.
 
         Returns:
-            Deferred[list[str]]: List of event IDs of the auth chain.
+            Deferred[Set[str]]: Set of event IDs.
         """
 
-        return self.store.get_auth_chain_ids(
-            event_ids, include_given=True, ignore_events=ignore_events,
-        )
+        return self.store.get_auth_chain_difference(state_sets)

synapse/state/v2.py

@@ -227,36 +227,12 @@ def _get_auth_chain_difference(state_sets, event_map, state_res_store):
     Returns:
         Deferred[set[str]]: Set of event IDs
     """
-    common = set(itervalues(state_sets[0])).intersection(
-        *(itervalues(s) for s in state_sets[1:])
+
+    difference = yield state_res_store.get_auth_chain_difference(
+        [set(state_set.values()) for state_set in state_sets]
     )
 
-    auth_sets = []
-    for state_set in state_sets:
-        auth_ids = {
-            eid
-            for key, eid in iteritems(state_set)
-            if (
-                key[0] in (EventTypes.Member, EventTypes.ThirdPartyInvite)
-                or key
-                in (
-                    (EventTypes.PowerLevels, ""),
-                    (EventTypes.Create, ""),
-                    (EventTypes.JoinRules, ""),
-                )
-            )
-            and eid not in common
-        }
-
-        auth_chain = yield state_res_store.get_auth_chain(auth_ids, common)
-        auth_ids.update(auth_chain)
-
-        auth_sets.append(auth_ids)
-
-    intersection = set(auth_sets[0]).intersection(*auth_sets[1:])
-    union = set().union(*auth_sets)
-
-    return union - intersection
+    return difference
 
 
 def _seperate(state_sets):