Kubernetes / Pods, Containers & Application Health
Investigate liveness-probe restarts without hiding real hangs
An overly aggressive liveness probe can restart a healthy but busy application, while a real deadlock may need recovery. Read the actual probe failures.
Environment and inputs: Kubernetes · kubectl with supported client/server versions and delegated RBAC. Examples use namespace app unless noted. Check context, namespace, APIs and provider support first.
Start with this check
Confirm the environment and replace sample objects, namespaces, images, paths and hosts with approved targets. Share only relevant, sanitized output. Follow the workload owner’s process for changes.
kubectl -n app describe pod web-0Read the result
Probe events, restart counts and termination details help correlate the pattern. Readiness and liveness have different consequences.
Resolve and verify
Compare response time under representative load with probe timeout and thresholds. Repair the endpoint or application bottleneck and test a controlled rollout.
Before changing production
Do not disable all probes or make liveness depend on every remote backend; that can amplify outages.