Precompute joined hosts and store in Redis (#9198)

2 files changed, 47 insertions, 0 deletions

diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py
index fd8de8696d..b6dc7f99b6 100644
--- a/synapse/handlers/federation.py
+++ b/synapse/handlers/federation.py
@@ -2093,6 +2093,11 @@ class FederationHandler(BaseHandler):
         if event.type == EventTypes.GuestAccess and not context.rejected:
             await self.maybe_kick_guest_users(event)
 
+        # If we are going to send this event over federation we precaclculate
+        # the joined hosts.
+        if event.internal_metadata.get_send_on_behalf_of():
+            await self.event_creation_handler.cache_joined_hosts_for_event(event)
+
         return context
 
     async def _check_for_soft_fail(
diff --git a/synapse/handlers/message.py b/synapse/handlers/message.py
index 9dfeab09cd..e2a7d567fa 100644
--- a/synapse/handlers/message.py
+++ b/synapse/handlers/message.py
@@ -432,6 +432,8 @@ class EventCreationHandler:
 
         self._ephemeral_events_enabled = hs.config.enable_ephemeral_messages
 
+        self._external_cache = hs.get_external_cache()
+
     async def create_event(
         self,
         requester: Requester,
@@ -939,6 +941,8 @@ class EventCreationHandler:
 
         await self.action_generator.handle_push_actions_for_event(event, context)
 
+        await self.cache_joined_hosts_for_event(event)
+
         try:
             # If we're a worker we need to hit out to the master.
             writer_instance = self._events_shard_config.get_instance(event.room_id)
@@ -978,6 +982,44 @@ class EventCreationHandler:
             await self.store.remove_push_actions_from_staging(event.event_id)
             raise
 
+    async def cache_joined_hosts_for_event(self, event: EventBase) -> None:
+        """Precalculate the joined hosts at the event, when using Redis, so that
+        external federation senders don't have to recalculate it themselves.
+        """
+
+        if not self._external_cache.is_enabled():
+            return
+
+        # We actually store two mappings, event ID -> prev state group,
+        # state group -> joined hosts, which is much more space efficient
+        # than event ID -> joined hosts.
+        #
+        # Note: We have to cache event ID -> prev state group, as we don't
+        # store that in the DB.
+        #
+        # Note: We always set the state group -> joined hosts cache, even if
+        # we already set it, so that the expiry time is reset.
+
+        state_entry = await self.state.resolve_state_groups_for_events(
+            event.room_id, event_ids=event.prev_event_ids()
+        )
+
+        if state_entry.state_group:
+            joined_hosts = await self.store.get_joined_hosts(event.room_id, state_entry)
+
+            await self._external_cache.set(
+                "event_to_prev_state_group",
+                event.event_id,
+                state_entry.state_group,
+                expiry_ms=60 * 60 * 1000,
+            )
+            await self._external_cache.set(
+                "get_joined_hosts",
+                str(state_entry.state_group),
+                list(joined_hosts),
+                expiry_ms=60 * 60 * 1000,
+            )
+
     async def _validate_canonical_alias(
         self, directory_handler, room_alias_str: str, expected_room_id: str
     ) -> None: