cc/ @roberthbailey @fabioy @wonderfly, the 3 failed GKE e2e test cases in our Jenkins are all due to this issue.

@dchen1107 No assignee? And should we target this for milestone 1.2?

I will work on it this week.

No need to target for milestone 1.2. New Kubelet metrics API is still in alpha, and the old ones are not deprecated yet. Also this is a cleanup task, and it is not the root cause of your issue: #23956

UPDATE: The test itself is hard-coded with docker's cgroup path. We should fix that as soon as possible.

The root cause is that cgroup "docker-daemon" is hard-coded in the kubelet_perf.go. But as #23956 (comment), this cgroup should not be created when systemd manages docker. Clean up this part should be enough for our testing.

this also impacts rktnetes, working around it by forcing it to think its not on gce/gke so it only uses / cgroup as don't have any other other static cgroups it is looking for, but in reality doesn't help much, just going to mark them as bad tests right now for us

Ok, since we decided to have a patch release including #23996, we need to address this very test issue for 1.2 too. cc/ @roberthbailey

this also impacts rktnetes, working around it by forcing it to think its not on gce/gke so it only uses / cgroup as don't have any other other static cgroups it is looking for, but in reality doesn't help much, just going to mark them as bad tests right now for us

For rkt and other container runtimes, I'll upload a separate PR to not check the runtime stats at all. The PR base the decision on the container runtime type reported by the node status. Any objections? @sjpotter @derekwaynecarr

#24003 was merged long ago. Closing the issue.