← ClaudeAtlas

k8s-debuglisted

Systematically debug Kubernetes workloads — pods stuck Pending, CrashLoopBackOff, OOMKilled, ImagePullBackOff, failing readiness probes, Services with no endpoints, DNS and network problems. Use when the user mentions kubectl, pods, deployments, a cluster, or a workload that won't start or serve traffic.
polytheseus/sre-skills · ★ 0 · DevOps & Infrastructure · score 72
Install: claude install-skill polytheseus/sre-skills
# Kubernetes debugging Work from the outside in: **workload → pod → container → node → network**. At each layer, read the status and events before forming a theory. ## Ground rules - Use read-only commands (`get`, `describe`, `logs`, `top`, `events`, `auth can-i`) freely. Before anything that changes the cluster (`delete`, `rollout restart`, `scale`, `edit`, `apply`, `cordon`/`drain`), show the command and its effect, and get the user's go-ahead. - Confirm the context and namespace first. Debugging the wrong cluster is a classic way to lose an hour or to break something else: ```bash kubectl config current-context kubectl get ns ``` - Always pass `-n <ns>` explicitly instead of relying on the default namespace. ## 1. Workload overview ```bash kubectl get deploy,sts,ds,rs,pods -n <ns> -o wide kubectl describe deploy/<name> -n <ns> # conditions, replica counts, events kubectl rollout status deploy/<name> -n <ns> --timeout=10s kubectl get events -n <ns> --sort-by=.lastTimestamp | tail -30 ``` If `READY` is lower than desired, take one failing pod and go to step 2. If the rollout is stuck, compare the new and old ReplicaSets: the new one is usually failing for one of the reasons below. ## 2. Pod status → cause Run `kubectl describe pod/<pod> -n <ns>` and read **Events** at the bottom, plus `State` / `Last State` / `Reason` / `Exit Code` per container. | Symptom | Likely causes | Check | |---|---|---| | `Pending` | Insufficient CPU/memory, unsatisfiab