cgroupdriver=systemd?

[Mars White]

xpost: https://github.com/coreos/bugs/issues/1435#issuecomment-558838802 

Currently running Kubernetes 1.11.10, CoreOS 2247.7.0 (latest AWS AMI, at least). We're seeing issues where nodes are unstable at high load.

Based on everything I can find on upstream Kubernetes (https://kubernetes.io/docs/setup/production-environment/container-runtimes/#cgroup-drivers) and various project recommendations (like Kubeadm: https://github.com/kubernetes/kubeadm/issues/1394), it seems to be recommended to use `systemd` to control cgroups on systems that are already running systemd -- like CoreOS. However, CoreOS ships with the docker default `cgroupfs` manager.

Can anyone shed light on why this is the case? Have the cgroup driver configuration recommendations changed since this thread? How should I be overriding Docker and Kubelet configurations to configure cgroupdriver? 

Thanks in advance!

[Andrew Jeddeloh]

We keep the same cgroup driver for backwards compat. If it's causing
issues for your use case, you can try switching as described in the
issue you linked.

- Andrew

> --
> You received this message because you are subscribed to the Google Groups "CoreOS User" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/coreos-user/fafbf63f-9256-4ccd-86da-5817062b0bb7%40googlegroups.com.