Frontend application on a1 system (mostly impacted) was unstable due to a query which hit database master instead of slave.
This has caused no downtime, but made some requests take longer than usual.
Also, it took us a bit to identify the issue as it had weird manifestations.
This might have been caused by enabling some feature, as all systems were stable up until:
- 04:00 UTC
- 00:00 EDT
- 06:00 CEST
https://www.youtube.com/watch?v=Vhh_GeBPOhs have sorted it.