After Container Crashed

[zhoub]

Hi,

Our Pod(program) in container might need a lot of memory to process data, and during our tests once the Pod(program) requested too much memory that exceeded the Node(physical machine) could offer, the k3s cluster enters a strange mode:

1. Any operation to the problematic Development didn't have any response
2. Command kubectl get nodes returned no metric server
3. Couldn't delete any ReplicaSet

The only workable solution is to restart k3s server, then k3s would

1. Clean up all legacy ReplicaSet with Pod 0/0
2. Problematic Development could restart

And after a while, everything would be back to normal.

Using

k3s version v1.22.2+k3s1 (10bca343)
go version go1.16.8

Is this a bug at k3s which could be fixed or improved ?

Thanks in advance !

[dweomer]

Is this a bug at k3s which could be fixed or improved ?

Nope. It's a bug in your deployment. Consider placing resource requests+limits on your containers to prevent them from making your node unresponsive. See https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/#resource-requests-and-limits-of-pod-and-container