This incident has been resolved. The issue seems to have been due to a short period where replication lag was higher than usual. The issue resolved itself after a few minutes and the replicas caught up. There will be planned works to improve observability and controls around replication lag and pooling

We are currently investigating this issue.