kubelet service not compatible with cgroups v2

I have a snap-based microk8s installation on Fedora. After upgrading to Fedora 31 Beta, kubelet service failed to start:

Sep 17 23:34:14 k8s.sakura-it.pl microk8s.daemon-kubelet[5736]: I0917 23:34:14.837307    5736 server.go:425] Version: v1.15.3
Sep 17 23:34:14 k8s.sakura-it.pl microk8s.daemon-kubelet[5736]: I0917 23:34:14.837804    5736 plugins.go:103] No cloud provider specified.
Sep 17 23:34:14 k8s.sakura-it.pl systemd[1]: run-rac03045094fb44c29a3af5719f0b1a94.scope: Succeeded.
Sep 17 23:34:14 k8s.sakura-it.pl microk8s.daemon-kubelet[5736]: F0917 23:34:14.857370    5736 server.go:273] failed to run Kubelet: mountpoint for cpu not found

This seems to be related to a change introduced in Fedora, that made cgroups v2 default: https://fedoraproject.org/wiki/Changes/CGroupsV2

Setting systemd.unified_cgroup_hierarchy=0 kernel parameter (essentially, going back to cgroups v1) works the problem around and now kubelet starts again.

I’d include the result of microk8s.inspect, it wasn’t for the fact that inspect script fails due to #631 .

[Richard87]

Editing /var/snap/microk8s/current/args/containerd-template.toml updating runtime_type to io.containerd.runc.v2 works!

     [plugins.cri.containerd.default_runtime]
        runtime_type = "io.containerd.runc.v2"
        runtime_engine = ""
        runtime_root = ""

[jglick]

https://github.com/ubuntu/microk8s/issues/651#issuecomment-718799924 works for me as well to solve

  Warning  FailedCreatePodSandBox  5m29s (x26 over 11m)  kubelet            Failed to create pod sandbox: rpc error: code = Unknown desc = failed to create containerd task: cgroups: cgroup mountpoint does not exist: unknown

after starting Microk8s for the first time in a while after some OS/kernel upgrades. Reference: https://github.com/ubuntu/microk8s/blob/c22142ae3c2bd3702b891e68d53704eb153c4efd/microk8s-resources/wrappers/run-containerd-with-args#L43-L45