Liveness probe failed
The kubelet ran the liveness probe, it failed failureThreshold times in a row, and the kubelet restarted the container.
Diagnose in order
Section titled “Diagnose in order”- What does the probe check, and how patiently?
kubectl get pod <pod> -o jsonpath='{.spec.containers[*].livenessProbe}'. NotetimeoutSeconds(default 1),periodSeconds(default 10), andfailureThreshold(default 3). - When does it fail: at startup, under load, or at random? Startup failures point at a missing startup probe. Load failures point at a timeout that’s too short or a health check that does real work.
- Call the endpoint yourself from inside the pod and time it, if the image has a shell:
kubectl exec <pod> -- sh -c 'time wget -qO- localhost:8080/healthz'.
Usual causes
Section titled “Usual causes”| Symptom | Cause |
|---|---|
| Fails only during startup | No startup probe, so liveness starts before the app is ready |
context deadline exceeded under load or during GC |
timeoutSeconds: 1 is too short |
| All replicas restart together | The health check calls a shared dependency (database, cache) that blipped |
exec probe fails with executable file not found |
The probe runs a tool that isn’t in the image (common with distroless) |
Add a startup probe for slow starters, set a realistic timeout, and make the liveness endpoint check only that the process can respond. Never check dependencies in liveness.
Read next: Probes and resource limits.