Can k8s redeploy the pod when container CrashLoopBackOff error contine?

General Discussions
1

Typically, we use a container liveness prober to monitor container within a pod. If the prober returns a failure, kubectl restarts the container not the pod. If the container continues to have problems, it will enter the CrashLoopBackOff state. Even in this state, the container continues to retry, but the Pod is normal.

If a container problems occurs, can I terminate the Pod itself and force it to be redistributed to another node?

The goal is to give unhealthy container one more high availability opportunity to run on another node automatically before administrator intervention.

I think it would be possible by developing operator, but I’m also curious if there’s already a feature like this.

2

JaeHeon,

Using kubectl delete pod <podName> will basically do what you are asking for. Sometimes it will recreate on the same node but either way it will be recreated if it is not a static pod and is apart of a deployment/replicaset/statefulset. Ideally you want to want to use one of those resources to achieve the highly available portion you’re referring to. This also allows you to scale the number of pods up if needed.