oxygen914/kubeoncall/deployment-unavailable-triage

Deployment Unavailable Triage

Diagnostic objective

Explain why desired Deployment capacity is unavailable by separating rollout, scheduling, image, probe, and dependency evidence.

Required live evidence

Compare desired, updated, available and ready replicas; rollout conditions; ReplicaSet revisions; Pod events; probe results; image digest; and the release timeline. Inspect failing Pod logs only after identifying the affected revision.

Decision branches

• Updated replicas stall: inspect scheduling, image pull, quota, and admission events.
• Pods run but never become Ready: verify probe failures and dependency reachability.
• Old and new revisions overlap: check rollout strategy, disruption budget, and capacity.
• A Pod has CrashLoopBackOff or OOMKilled: hand off to the specific Pod Skill instead of treating it as the primary Deployment cause.

Safe next step

Use the runbook to assemble revision-level evidence and propose a separately governed rollback, scale, or configuration task.

Prohibited actions

Do not restart, scale, roll back, patch, or delete Pods from this diagnostic Skill.

Recovery and escalation

Require desired available replicas, completed rollout conditions, stable readiness, and recovered service indicators. Escalate when no healthy revision exists or the rollout affects a critical service.