oxygen914/kubeoncall/pod-crashloop-triage

Pod CrashLoopBackOff Triage

Diagnostic objective

Explain repeated container exits by separating application, configuration, dependency, probe, image, and resource evidence.

Required live evidence

Collect last state, reason, exit code, restart count, events, revision, current and previous logs, probes, image digest, requests/limits, node, and release timing. Compare a healthy replica.

Decision branches

• Exit code or stack trace is deterministic: correlate it with the affected revision and configuration source.
• Startup or liveness probe fails: verify real startup behavior and dependency reachability before blaming probe settings.
• OOMKilled is present: hand off to pod-oom-triage.
• Scheduling or node evidence leads the failure: hand off to the cluster-capacity or node Skill.

Safe next step

Preserve previous-log and revision evidence; propose any change as a separate governed task.

Prohibited actions

Do not loop-delete Pods, restart the workload, disable probes, expose Secret values, or patch live Pods from this Skill.

Recovery and escalation

Require stable restart count, Ready replicas, clean error trends, and healthy dependencies across two windows. Escalate when no healthy replica exists, previous logs are unavailable, or a critical dependency is overloaded.