Is it possible to blacklist /kubepods.slice ?

[zbialik]

We're already using kubelet's default exposed cadvisor metrics to monitor pod/container usages.

We want to monitor other usages for other cgroups under /system.slice + /kube.slice (for container runtime + kubelet), but this isn't exposed by kubelet right now.

I see cadvisor exposes the other cgroup metrics properly, but I can't seem to get it configured to ONLY expose those under /system.slice + /kube.slice. It appears like cadvisor will always expose the cgroups under /kubepods.slice (perhaps due to a reliance on CRI?).

I tried using the whitelist argument (--raw_cgroup_prefix_whitelist=/system.slice) but continue to get metrics for /kubepods.slice. Is this supposed to happen?

Current workaround is to replace our scraped kubelet-cadvisor metrics with new ones from a cadvisor DaemonSet, but I'd prefer to just use our cadvisor DaemonSet for the system-specific cgroups, not those already covered by kubelet.

[deepakzac]

hi @zbialik I'm running into the same exact issue. Did you ever resolve it?