k8s Flashcards
← Back to blog
kubectlPod DebuggingStudy TipsDecision Trainer

describe or logs? Training Your First kubectl Decision

The hardest part of troubleshooting is not knowing commands — it is picking the right one first. A simple decision tree for Pod status under pressure.

1 min read

Every Kubernetes course teaches kubectl describe and kubectl logs. Almost none drill when to use which — yet that is what breaks you in incidents and on CKAD/CKA tasks.

The decision tree

Pod status?
│
├─ Pending ──────────────► describe pod (FailedScheduling Events)
│                          logs usually empty / error
│
├─ ContainerCreating ────► describe pod (FailedMount, sandbox errors)
│
├─ ImagePullBackOff ─────► describe pod (registry/tag/auth in Events)
│
├─ CrashLoopBackOff ─────► describe pod (exit code, probes, OOM)
│                          then logs --previous
│
├─ Running but not Ready ► describe pod (readiness probe Events)
│                          then logs (current container)
│
└─ Running and Ready ────► logs / exec (application-level issue)

Why describe wins early

kubectl describe pod bundles:

  • Status and conditions
  • Container states and last termination reason
  • Events (scheduler, kubelet, probe failures)

One command answers where in the lifecycle the failure lives.

When logs are first

Logs shine after the container has run at least once:

  • CrashLoopBackOff → kubectl logs <pod> --previous for the crash you missed
  • Running app errors → current logs
  • Init container failure → kubectl logs <pod> -c <init-container-name>

logs on a Pending pod wastes seconds — there is no container output.

Trap answers examiners love

TrapWhy it hurts
`kubectl delete pod` firstController recreates; root cause unchanged
`kubectl logs` on PendingNo logs exist
Restart kubelet immediatelyApp/config issue masked as infra
Scale Deployment to zeroHides problem; does not fix it

Decision Trainer grades these explicitly — best, acceptable, bad, dangerous.

Practice deliberately

Read the status line in scenario context, pick your first step, then compare to the debrief. Pod Debugging is free on k8s Flashcards — five minutes per session beats re-reading a command cheat sheet.

Related: CrashLoopBackOff guide · Pending pods