diff options
author | Richard van der Hoff <1389908+richvdh@users.noreply.github.com> | 2019-10-10 11:29:01 +0100 |
---|---|---|
committer | GitHub <noreply@github.com> | 2019-10-10 11:29:01 +0100 |
commit | a139420a3cfda6a4a4ee4750611b31dd71fc33f3 (patch) | |
tree | 9129aff6a5e73673054a1c9125fa977099a2b7db /synapse/storage | |
parent | Rewrite the user_filter migration again (#6184) (diff) | |
download | synapse-a139420a3cfda6a4a4ee4750611b31dd71fc33f3.tar.xz |
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.
Diffstat (limited to 'synapse/storage')
-rw-r--r-- | synapse/storage/state_deltas.py | 38 |
1 files changed, 29 insertions, 9 deletions
diff --git a/synapse/storage/state_deltas.py b/synapse/storage/state_deltas.py index 5fdb442104..28f33ec18f 100644 --- a/synapse/storage/state_deltas.py +++ b/synapse/storage/state_deltas.py @@ -21,7 +21,7 @@ logger = logging.getLogger(__name__) class StateDeltasStore(SQLBaseStore): - def get_current_state_deltas(self, prev_stream_id): + def get_current_state_deltas(self, prev_stream_id: int, max_stream_id: int): """Fetch a list of room state changes since the given stream id Each entry in the result contains the following fields: @@ -36,15 +36,27 @@ class StateDeltasStore(SQLBaseStore): Args: prev_stream_id (int): point to get changes since (exclusive) + max_stream_id (int): the point that we know has been correctly persisted + - ie, an upper limit to return changes from. Returns: - Deferred[list[dict]]: results + Deferred[tuple[int, list[dict]]: A tuple consisting of: + - the stream id which these results go up to + - list of current_state_delta_stream rows. If it is empty, we are + up to date. """ prev_stream_id = int(prev_stream_id) + + # check we're not going backwards + assert prev_stream_id <= max_stream_id + if not self._curr_state_delta_stream_cache.has_any_entity_changed( prev_stream_id ): - return [] + # if the CSDs haven't changed between prev_stream_id and now, we + # know for certain that they haven't changed between prev_stream_id and + # max_stream_id. + return max_stream_id, [] def get_current_state_deltas_txn(txn): # First we calculate the max stream id that will give us less than @@ -54,21 +66,29 @@ class StateDeltasStore(SQLBaseStore): sql = """ SELECT stream_id, count(*) FROM current_state_delta_stream - WHERE stream_id > ? + WHERE stream_id > ? AND stream_id <= ? GROUP BY stream_id ORDER BY stream_id ASC LIMIT 100 """ - txn.execute(sql, (prev_stream_id,)) + txn.execute(sql, (prev_stream_id, max_stream_id)) total = 0 - max_stream_id = prev_stream_id - for max_stream_id, count in txn: + + for stream_id, count in txn: total += count if total > 100: # We arbitarily limit to 100 entries to ensure we don't # select toooo many. + logger.debug( + "Clipping current_state_delta_stream rows to stream_id %i", + stream_id, + ) + clipped_stream_id = stream_id break + else: + # if there's no problem, we may as well go right up to the max_stream_id + clipped_stream_id = max_stream_id # Now actually get the deltas sql = """ @@ -77,8 +97,8 @@ class StateDeltasStore(SQLBaseStore): WHERE ? < stream_id AND stream_id <= ? ORDER BY stream_id ASC """ - txn.execute(sql, (prev_stream_id, max_stream_id)) - return self.cursor_to_dict(txn) + txn.execute(sql, (prev_stream_id, clipped_stream_id)) + return clipped_stream_id, self.cursor_to_dict(txn) return self.runInteraction( "get_current_state_deltas", get_current_state_deltas_txn |