From 2:09 pm UTC to 2:20 pm UTC, we experience increased delivery latency (p99 of ~5 minutes at its worst). A DB switchover, which is normally transparent and doesn't impact latency, left some workers in a crashed state and failed to recover automatically. We're looking into the root cause of why the workers crashed and did not recover.

We're seeing increased latency in event delivery and are investigating the root cause.