Blameless retrospective · revised 30 Aug 2026

Publishing delay retrospective

Impact

Observed: 8% of publishes between 14:02–14:31 UTC exceeded 120 seconds; no documents or revisions were lost.

Evidence and analysis

Evidence: queue depth rose from 12 to 214 after a traffic spike.

Contributing cause: workers scaled on CPU, while the bottleneck was queue age.

Actions

ActionOwnerDue
Alert on oldest queued jobReliability4 Sep
Scale on queue agePlatform11 Sep