What it's like
> checkout is throwing 5xx since about 14:02, can you look?
Reading before guessing. Error rate on checkout went from 0.2% to 31% at 14:02:40. Deploy checkout v2.31 finished at 14:01:55; nothing else changed in that window. New error in the logs since then: PG connection pool exhausted (max=20).
Impact: about a third of checkouts failing, all regions.
Likely cause: v2.31 (pool size config changed from 50 to 20).
Proposed mitigation (needs your OK):
kubectl rollout undo deployment/checkout -n shop
Blast radius: checkout pods only, rolling, no downtime. Undo: kubectl rollout undo again. I'll draft the status update while you decide.
An illustrative session. The skills ask before any command that changes production.
Free skills
- incident-triage
- The first 15 minutes of an incident: impact, what changed, mitigation, status updates, timeline.
- k8s-debug
- Pending, CrashLoopBackOff, OOMKilled, ImagePullBackOff, not-Ready pods, Services with no endpoints.
- ci-failure
- Red CI to a verdict (regression, flaky or infra) with evidence, then the fix.
- postmortem
- Blameless postmortems with owned, verifiable action items.
- terraform-plan-review
- Destroys, replacements, IAM widening and public exposure, with a go/no-go before you apply.
- log-forensics
- The first bad minute, new and spiking messages, and request correlation across services.
Pro pack
Eight deeper skills for teams running production: slo-design, alert-audit, cloud-cost-audit, postgres-incident, linux-perf, runbook-writer, oncall-handoff and dockerfile-review. Pay once and get lifetime updates.
Solo: $29 / £25 / €27 Team (up to 10): $99 / £79 / €89
Checkout and VAT/sales tax are handled by Polar, the merchant of record. The download is delivered instantly. Not happy within 14 days? Reply to your receipt for a refund.