Prevent back-off in kubernetes crash loop

问题:

I have a pod with some terrible, buggy software in it. One reason Kubernetes is great is that it'll just restart the software when it crashes, which is awesome.

Kubernetes was designed for good software, not terrible software, so it does an exponential backoff while restarting pods. This means I have to wait five minutes between crashes before my pods are restarted.

Is there any way to cap the kubernetes backoff strategy? I'd like to change it to not wait longer than thirty seconds before starting up the pod again.

回答1:

Unfortunately, the max back off time for container restarts is not tunable for the node reliability (i.e., too many container restarts can overwhelm the node). If you absolutely want to change it in your cluster, you will need to modify the max backoff time in the code, compile your own kubelet binary, and distribute it onto your nodes.