Kubernetes Pod Debugger

CrashLoopBackOff means the container keeps starting and crashing. Kubernetes backs off between restarts (1s, 2s, 4s... up to 5 minutes). Common causes: application error on startup (bad config, missing env var), failed health check, out-of-memory kill (OOMKilled), or trying to bind to an already-used port. Check 'kubectl logs' and 'kubectl describe pod' for the exit code and last error.

Pending means Kubernetes can't schedule the pod. Common causes: insufficient CPU/memory on all nodes (FailedScheduling), PodDisruptionBudget blocking, taint/toleration mismatch (NoSchedule), NodeAffinity/NodeSelector not matching any node, or PVC not bound (no storage class). Check 'kubectl describe pod' for the Events section which usually identifies the exact reason.

ErrImagePull is the initial failure when Kubernetes can't pull a container image. ImagePullBackOff is the retry state with exponential backoff. Causes: typo in image name, private registry without imagePullSecrets, rate limiting (Docker Hub), or deleted image tag. Fix by checking the image name, adding imagePullSecrets for private registries, or using a specific tag instead of latest.

For a crashed/restarted container, use 'kubectl logs <pod> --previous' (-p) to get the logs from the last terminated container instance. Without --previous you get current instance logs (which may be empty if it just started). For multi-container pods, add '-c <container-name>' to target a specific container.

OOMKilled (exit code 137) means the container exceeded its memory limit and was killed by the Linux OOM killer. Fix: either increase the memory limit in your resource spec (limits.memory), or optimize your application's memory usage. Use 'kubectl top pods' to see actual memory usage and set your limit to at least 2× the observed peak. Check for memory leaks with profiling tools.

Use 'kubectl exec -it <pod-name> -- /bin/sh' (or /bin/bash for bash). For pods without a shell (distroless), use 'kubectl debug': it creates an ephemeral debug container with tools attached to the same pod. Always specify '-n <namespace>' if not in the default namespace.