Reduce auth chains fetched during v2 state res. (#6952)

* commit '2b37eabca':
  Reduce auth chains fetched during v2 state res. (#6952)
  update changes.md

CHANGES.md

@@ -7,12 +7,12 @@ Features
 - Admin API to add or modify threepids of user accounts. ([\#6769](https://github.com/matrix-org/synapse/issues/6769))
 - Limit the number of events that can be requested by the backfill federation API to 100. ([\#6864](https://github.com/matrix-org/synapse/issues/6864))
 - Add ability to run some group APIs on workers. ([\#6866](https://github.com/matrix-org/synapse/issues/6866))
-- Reject device display names over 100 characters in length. ([\#6882](https://github.com/matrix-org/synapse/issues/6882))
+- Reject device display names over 100 characters in length to prevent abuse. ([\#6882](https://github.com/matrix-org/synapse/issues/6882))
 - Add ability to route federation user device queries to workers. ([\#6873](https://github.com/matrix-org/synapse/issues/6873))
 - The result of a user directory search can now be filtered via the spam checker. ([\#6888](https://github.com/matrix-org/synapse/issues/6888))
-- Implement `GET /_matrix/client/r0/rooms/{roomId}/aliases` endpoint as per [MSC2432](https://github.com/matrix-org/matrix-doc/pull/2432). ([\#6939](https://github.com/matrix-org/synapse/issues/6939), [\#6948](https://github.com/matrix-org/synapse/issues/6948), [\#6949](https://github.com/matrix-org/synapse/issues/6949))
+- Implement new `GET /_matrix/client/unstable/org.matrix.msc2432/rooms/{roomId}/aliases` endpoint as per [MSC2432](https://github.com/matrix-org/matrix-doc/pull/2432). ([\#6939](https://github.com/matrix-org/synapse/issues/6939), [\#6948](https://github.com/matrix-org/synapse/issues/6948), [\#6949](https://github.com/matrix-org/synapse/issues/6949))
 - Stop sending `m.room.alias` events wheng adding / removing aliases. Check `alt_aliases` in the latest `m.room.canonical_alias` event when deleting an alias. ([\#6904](https://github.com/matrix-org/synapse/issues/6904))
-
+- Change the default power levels of invites, tombstones and server ACLs for new rooms. ([\#6834](https://github.com/matrix-org/synapse/issues/6834))
 
 Bugfixes
 --------
@@ -41,6 +41,7 @@ Improved Documentation
 - Clean up and update docs on setting up federation. ([\#6940](https://github.com/matrix-org/synapse/issues/6940))
 - Add a warning about indentation to generated configuration files. ([\#6920](https://github.com/matrix-org/synapse/issues/6920))
 - Databases created using the compose file in contrib/docker will now always have correct encoding and locale settings. Contributed by Fridtjof Mund. ([\#6921](https://github.com/matrix-org/synapse/issues/6921))
+- Update pip install directions in readme to avoid error when using zsh. ([\#6855](https://github.com/matrix-org/synapse/issues/6855))
 
 
 Deprecations and Removals
@@ -54,15 +55,13 @@ Internal Changes
 
 - Add type hints to `SyncHandler`. ([\#6821](https://github.com/matrix-org/synapse/issues/6821))
 - Refactoring work in preparation for changing the event redaction algorithm. ([\#6823](https://github.com/matrix-org/synapse/issues/6823), [\#6827](https://github.com/matrix-org/synapse/issues/6827), [\#6854](https://github.com/matrix-org/synapse/issues/6854), [\#6856](https://github.com/matrix-org/synapse/issues/6856), [\#6857](https://github.com/matrix-org/synapse/issues/6857), [\#6858](https://github.com/matrix-org/synapse/issues/6858))
-- Change the default power levels of invites, tombstones and server ACLs for new rooms. ([\#6834](https://github.com/matrix-org/synapse/issues/6834))
 - Fix stacktraces when using `ObservableDeferred` and async/await. ([\#6836](https://github.com/matrix-org/synapse/issues/6836))
 - Port much of `synapse.handlers.federation` to async/await. ([\#6837](https://github.com/matrix-org/synapse/issues/6837), [\#6840](https://github.com/matrix-org/synapse/issues/6840))
 - Populate `rooms.room_version` database column at startup, rather than in a background update. ([\#6847](https://github.com/matrix-org/synapse/issues/6847))
-- Update pip install directions in readme to avoid error when using zsh. ([\#6855](https://github.com/matrix-org/synapse/issues/6855))
 - Reduce amount we log at `INFO` level. ([\#6833](https://github.com/matrix-org/synapse/issues/6833), [\#6862](https://github.com/matrix-org/synapse/issues/6862))
 - Remove unused `get_room_stats_state` method. ([\#6869](https://github.com/matrix-org/synapse/issues/6869))
 - Add typing to `synapse.federation.sender` and port to async/await. ([\#6871](https://github.com/matrix-org/synapse/issues/6871))
-- Refactor _EventInternalMetadata object to improve type safety. ([\#6872](https://github.com/matrix-org/synapse/issues/6872))
+- Refactor `_EventInternalMetadata` object to improve type safety. ([\#6872](https://github.com/matrix-org/synapse/issues/6872))
 - Add an additional entry to the SyTest blacklist for worker mode. ([\#6883](https://github.com/matrix-org/synapse/issues/6883))
 - Fix the use of sed in the linting scripts when using BSD sed. ([\#6887](https://github.com/matrix-org/synapse/issues/6887))
 - Add type hints to the spam checker module. ([\#6915](https://github.com/matrix-org/synapse/issues/6915))

changelog.d/6952.misc

@@ -0,0 +1 @@
+Improve perf of v2 state res for large rooms.

synapse/state/__init__.py

@@ -16,7 +16,7 @@
 
 import logging
 from collections import namedtuple
-from typing import Dict, Iterable, List, Optional
+from typing import Dict, Iterable, List, Optional, Set
 
 from six import iteritems, itervalues
 
@@ -662,7 +662,7 @@ class StateResolutionStore(object):
             allow_rejected=allow_rejected,
         )
 
-    def get_auth_chain(self, event_ids):
+    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).
 
@@ -674,11 +674,16 @@ class StateResolutionStore(object):
                presence of rejected events
 
         Args:
-            event_ids (list): The event IDs of the events to fetch the auth
-                chain for. Must be state events.
+            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.
+
 
         Returns:
             Deferred[list[str]]: List of event IDs of the auth chain.
         """
 
-        return self.store.get_auth_chain_ids(event_ids, include_given=True)
+        return self.store.get_auth_chain_ids(
+            event_ids, include_given=True, ignore_events=ignore_events,
+        )

synapse/state/v2.py

@@ -248,7 +248,7 @@ def _get_auth_chain_difference(state_sets, event_map, state_res_store):
             and eid not in common
         )
 
-        auth_chain = yield state_res_store.get_auth_chain(auth_ids)
+        auth_chain = yield state_res_store.get_auth_chain(auth_ids, common)
         auth_ids.update(auth_chain)
 
         auth_sets.append(auth_ids)

synapse/storage/data_stores/main/event_federation.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 import itertools
 import logging
+from typing import List, Optional, Set
 
 from six.moves import range
 from six.moves.queue import Empty, PriorityQueue
@@ -46,21 +47,37 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
             event_ids, include_given=include_given
         ).addCallback(self.get_events_as_list)
 
-    def get_auth_chain_ids(self, event_ids, include_given=False):
+    def get_auth_chain_ids(
+        self,
+        event_ids: List[str],
+        include_given: bool = False,
+        ignore_events: Optional[Set[str]] = None,
+    ):
         """Get auth events for given event_ids. The events *must* be state events.
 
         Args:
-            event_ids (list): state events
-            include_given (bool): include the given events in result
+            event_ids: state events
+            include_given: include the given events in result
+            ignore_events: Set of events to exclude from the returned auth
+                chain. This is useful if the caller will just discard the
+                given events anyway, and saves us from figuring out their auth
+                chains if not required.
 
         Returns:
             list of event_ids
         """
         return self.db.runInteraction(
-            "get_auth_chain_ids", self._get_auth_chain_ids_txn, event_ids, include_given
+            "get_auth_chain_ids",
+            self._get_auth_chain_ids_txn,
+            event_ids,
+            include_given,
+            ignore_events,
         )
 
-    def _get_auth_chain_ids_txn(self, txn, event_ids, include_given):
+    def _get_auth_chain_ids_txn(self, txn, event_ids, include_given, ignore_events):
+        if ignore_events is None:
+            ignore_events = set()
+
         if include_given:
             results = set(event_ids)
         else:
@@ -80,6 +97,7 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas
                 txn.execute(base_sql + clause, list(args))
                 new_front.update([r[0] for r in txn])
 
+            new_front -= ignore_events
             new_front -= results
 
             front = new_front

tests/state/test_v2.py

@@ -603,7 +603,7 @@ class TestStateResolutionStore(object):
 
         return {eid: self.event_map[eid] for eid in event_ids if eid in self.event_map}
 
-    def get_auth_chain(self, event_ids):
+    def get_auth_chain(self, event_ids, ignore_events):
         """Gets the full auth chain for a set of events (including rejected
         events).
 
@@ -617,6 +617,8 @@ class TestStateResolutionStore(object):
         Args:
             event_ids (list): 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.
 
         Returns:
             Deferred[list[str]]: List of event IDs of the auth chain.
@@ -627,7 +629,7 @@ class TestStateResolutionStore(object):
         stack = list(event_ids)
         while stack:
             event_id = stack.pop()
-            if event_id in result:
+            if event_id in result or event_id in ignore_events:
                 continue
 
             result.add(event_id)