Kubernetes · 2 steps

CrashLoopBackOff

Read the previous container’s logs and its exit code — those two facts resolve most crash loops before you touch anything else.

Every step

Read the logs of the container that died

The running container is a fresh one that has not failed yet. Its logs are nearly empty. The evidence is in the previous instance, and `--previous` is the flag that gets it.

$ kubectl logs <pod> --previous

$ kubectl logs <pod> --previous -c <container> # when there are several

$ kubectl describe pod <pod> | tail -30

Look for: The last lines before it stopped, and in describe: `Last State`, `Reason` and `Exit Code`.

Check whether a probe is killing it

A liveness probe that is stricter than the application’s startup time creates an unbreakable loop: the container is killed before it ever becomes ready, forever.

$ kubectl describe pod <pod> | grep -A5 -E "Liveness|Readiness|Startup"

$ kubectl get events --field-selector involvedObject.name=<pod> --sort-by=.lastTimestamp

$ kubectl exec <pod> -- curl -sv localhost:<port>/healthz # if it stays up long enough

Look for: Events saying "Liveness probe failed", and the initialDelaySeconds against how long the application really takes to start.