Cri-o: applying cgroup configuration for process caused \"No such device or address\""

[runcom]

Starting a local kube cluster with cri-o master throws this when starting kube-dns:

  1m    2s      8       kubelet, 127.0.0.1      spec.containers{kubedns}        Warning Failed          Error: container create failed: container_linux.go:265: starting container process caused "process_linux.go:264: applying cgroup configuration for process caused \"No such device or address\""

I thought it was because of https://github.com/kubernetes-incubator/cri-o/pull/812 but after reverting that the issue still persists. I tested with systemd cgroup manager + fedora 26 + systemd 233-6, everything else is stock CRI-O (runc, conmon), SELinux is disabled.

It only happens to the kube-dns container, I can start any other container/pod just fine. (in the kube-dns pod, sidecar and dnsmasq containers start up just fine, it's just kube-dns...)

The same happens with cgroupfs fwiw

Not sure what's going on...

[runcom]

@mrunalp PTAL

[runcom]

Seems like it was just runc, using latest runc master works fine

[huzhengchuan]

I meet the same question. How to fix it. @runcom

[Hades32]

@runcom same happening here with Docker 17.09 on Ubuntu 17.04

[runcom]

just upgrade runc

[Hades32]

I don`t have runc installed at all. I just have followed the tutorial for "kubeadm". I only decided to use the newest docker instead of 17.03...

[rhatdan]

Well CRI-O will definitely not work without runc installed. Unless you override with a different OCI Runtime implementation.

[Hades32]

Not sure what problem I really had, but for anybody coming here (via google) for the same reason:
You should NOT follow the tutorial (https://kubernetes.io/docs/setup/independent/install-kubeadm/) and instead use cgroupfs instead of systemd.
Compiling runc from master didn´t change anything for me...

[MOZGIII]

Here are some snippets/examples to resolve this issue: https://gist.github.com/MOZGIII/22bf4eb811ff5d4e0bbe36444422b6d3