Unnecessary pids directory created under cgroup v2 root #862
Description
I'm running kubernetes v1.22 with cri-o v1.22 on a cgroupv2-enabled node. When starting pods, cri-o still creates /sys/fs/cgroup/pids/kubepods.slice/xxxxxx, like what it may do for cgroupv1. Then kubelet reports lots of container /pids/kubepods.slice/xxxxxx not found errors in its logs.
It seems that cri-o uses github.com/containers/podman/v3/pkg/cgroups, which was moved to this repository as pkg/cgroups several days ago.
Comparing the code, in pkg/cgroups/cpus.go, it ignores the directory creation for cgroupv2:
func (c *cpuHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
return false, nil
}
return ctr.createCgroupDirectory(CPU)
}But in pkg/cgroups/pids.go, it does not:
func (c *pidHandler) Create(ctr *CgroupControl) (bool, error) {
return ctr.createCgroupDirectory(Pids)
}Is it intended to create the pids directory or not?