Production Readiness Review

Our review covers 40+ items across categories: observability (metrics, logs, traces, dashboards), alerting (symptom-based alerts, runbooks), failure handling (circuit breakers, retries, timeouts, graceful degradation), scaling (HPA configuration, load testing results, capacity headroom), security (authentication, authorization, secret management), and operations (deployment pipeline, rollback procedure, on-call documentation).

We map the service's dependency tree and identify single points of failure, undocumented assumptions, and untested failure modes. What happens when the primary database fails over? How does the service behave when a downstream API is slow? Is there a thundering herd risk on cold start? Each risk is rated by likelihood and impact, with mitigation recommendations prioritized accordingly.

No service goes to production without load test results. We verify the service handles expected peak traffic with acceptable latency, that auto-scaling triggers correctly, and that resource limits are set appropriately. If load tests have not been run, we run them. If they were run months ago, we re-run against the current codebase. Stale performance data is worse than no data.

The review results in one of three outcomes: approved (ready for production), conditionally approved (launch with specific follow-up items within 14 days), or not approved (blocking issues must be resolved). Conditional approvals are tracked in the issue tracker with deadlines. We have found that 70% of services need at least one iteration before achieving full production readiness — and catching issues in review is far cheaper than catching them in production.