Fix races in room stats (and other) updates. (#6187)

Hopefully this will fix the occasional failures we were seeing in the room directory.

The problem was that events are not necessarily persisted (and `current_state_delta_stream` updated) in the same order as their stream_id. So for instance current_state_delta 9 might be persisted *before* current_state_delta 8. Then, when the room stats saw stream_id 9, it assumed it had done everything up to 9, and never came back to do stream_id 8.

We can solve this easily by only processing up to the stream_id where we know all events have been persisted.

1 files changed, 12 insertions, 5 deletions

diff --git a/synapse/handlers/user_directory.py b/synapse/handlers/user_directory.py
index e53669e40d..624f05ab5b 100644
--- a/synapse/handlers/user_directory.py
+++ b/synapse/handlers/user_directory.py
@@ -138,21 +138,28 @@ class UserDirectoryHandler(StateDeltasHandler):
         # Loop round handling deltas until we're up to date
         while True:
             with Measure(self.clock, "user_dir_delta"):
-                deltas = yield self.store.get_current_state_deltas(self.pos)
-                if not deltas:
+                room_max_stream_ordering = self.store.get_room_max_stream_ordering()
+                if self.pos == room_max_stream_ordering:
                     return
 
+                logger.debug(
+                    "Processing user stats %s->%s", self.pos, room_max_stream_ordering
+                )
+                max_pos, deltas = yield self.store.get_current_state_deltas(
+                    self.pos, room_max_stream_ordering
+                )
+
                 logger.info("Handling %d state deltas", len(deltas))
                 yield self._handle_deltas(deltas)
 
-                self.pos = deltas[-1]["stream_id"]
+                self.pos = max_pos
 
                 # Expose current event processing position to prometheus
                 synapse.metrics.event_processing_positions.labels("user_dir").set(
-                    self.pos
+                    max_pos
                 )
 
-                yield self.store.update_user_directory_stream_pos(self.pos)
+                yield self.store.update_user_directory_stream_pos(max_pos)
 
     @defer.inlineCallbacks
     def _handle_deltas(self, deltas):