Idle transaction locks after deploy

On-call incidentMedium30 minStripe style

Your task

It's 09:14 and your payments API p99 latency jumped from 80ms to 6s, with a rising trickle of 500s. The app tier dashboards show CPU at 30% and memory flat, nothing looks saturated there. The Postgres dashboard shows active connections pinned at the 200 max for the primary, a deep ClientRead/Lock wait spike, and pg_stat_activity full of sessions in idle in transaction. A deploy went out at 09:05 that added a new 'order summary' endpoint. Walk me through how you triage this and what you do in the next 10 minutes versus the next week.

It covers

  • Connection pool
  • Database
  • Saturation