This is the Principal-level incident response question. Show structure, breadth, and depth — not just commands.

Systematic diagnosis methodology

Phase 1 — Scope the failure:

kubectl rollout status deployment/myapp # stuck? rolling back?

kubectl get pods -l app=myapp # pod states

kubectl get events --sort-by=.lastTimestamp -n <ns> | tail -20

Phase 2 — Inspect the failing pod:

kubectl describe pod <failing-pod> # Events section is gold

kubectl logs <failing-pod> --previous # logs from crashed container

kubectl logs <failing-pod> -c init-container-name # init container logs

Phase 3 — Identify the root cause category:

ImagePullBackOff → registry auth, wrong tag, image doesn't exist

CrashLoopBackOff → app crash: check logs, exit code, OOM (137)

Pending → scheduling: resources, node selectors, taints, PVC unbound

OOMKilled (exit 137) → increase memory limit, fix memory leak

Readiness failing → app not ready: check probe endpoint, startup time

Phase 4 — Immediate relief:

kubectl rollout undo deployment/myapp # rollback if new version

kubectl scale deployment/myapp --replicas=5 # scale up if degraded

Phase 5 — Fix and verify:

Fix root cause → commit → CI pipeline → re-deploy → verify rollout

Common patterns: ImagePullBackOff with "unauthorized" = Workload Identity not configured or imagePullSecret missing. CrashLoopBackOff with exit 1 = application error (check logs). CrashLoopBackOff with exit 137 = OOM kill (increase memory limit). Pending with "0/3 nodes are available" = insufficient resources (scale node pool) or taint mismatch. Readiness probe failing = app slow to start (add startupProbe) or dependency unavailable (check DB connectivity).

Principal-level deliverable: After every significant deployment failure, produce a postmortem: what failed, why it wasn't caught in staging, what the detection gap was (missing alert/test), and what process change prevents recurrence. Deploy failures that reach production represent gaps in testing or staging environment fidelity — fix the pipeline, not just the immediate issue.