mirrors/synapse
1
0
Fork 0
mirror of https://github.com/element-hq/synapse.git synced 2025-12-05 01:10:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix CPU and database spinning when retrying sending events to servers whilst at the same time purging those events. (#18499)

Browse Source 
Fixes: #18491

Fix hotlooping due to skipped PDUs if there is still no progress to be
made.
This could bite if the event was purged since being skipped during
catch-up.

Signed-off-by: Olivier 'reivilibre <oliverw@matrix.org>
This commit is contained in:
reivilibre
2025-07-15 11:01:41 +00:00
committed by GitHub
parent 945e22303c
commit 97d2738eef
2 changed files with 14 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
changelog.d/18499.bugfix Normal file
View File

@@ -0,0 +1 @@
Fix CPU and database spinning when retrying sending events to servers whilst at the same time purging those events.

14
synapse/federation/sender/per_destination_queue.py
View File

@@ -129,6 +129,8 @@ class PerDestinationQueue:
# The stream_ordering of the most recent PDU that was discarded due to
# being in catch-up mode.
# Can be set to zero if no PDU has been discarded since the last time
# we queried for new PDUs during catch-up.
self._catchup_last_skipped: int = 0
# Cache of the last successfully-transmitted stream ordering for this
@@ -462,8 +464,18 @@ class PerDestinationQueue:
# of a race condition, so we check that no new events have been
# skipped due to us being in catch-up mode
if self._catchup_last_skipped > last_successful_stream_ordering:
if (
self._catchup_last_skipped != 0
and self._catchup_last_skipped > last_successful_stream_ordering
):
# another event has been skipped because we were in catch-up mode
# As an exception to this case: we can hit this branch if the
# room has been purged whilst we have been looping.
# In that case we avoid hot-looping by resetting the 'catch-up skipped
# PDU' flag.
# Then if there is still no progress to be made at the next iteration,
# we can exit catch-up mode.
self._catchup_last_skipped = 0
continue
# we are done catching up!