cAdvisor metrics on OpenShift nodes sum CPU into `/` and `/system.slice` cgroups, but not memory

[smarterclayton]

container_cpu_usage_seconds_total is cumulative for cgroups in 3.5 and 3.6. container_memory_rss is not. Confusing, not necessarily broken.

[smarterclayton]

Not a 3.6 blocker, but both confusing and potentially not right

[sjenning]

possibly related opencontainers/runc#1376

[derekwaynecarr]

my fix for runc merged, we will have to bump runc in cadvisor, and use it to expose total_ values so we do not look at charges local to the cgroup but also its children.

[sjenning]

for the record, that is a bump to at least opencontainers/runc@7139b61

[sjenning]

@aveshagarwal PTAL

[smarterclayton]

I thought this was fixed in 3.7

…

On Fri, Dec 1, 2017 at 2:04 PM, Seth Jennings ***@***.***> wrote: @aveshagarwal <https://github.com/aveshagarwal> PTAL — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#14920 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABG_p9kFeVbuRv51TKvoQTKeBxMAuTQvks5s8E3SgaJpZM4OG9rQ> .

[sjenning]

Very true, fixed by #16419 and pulled in on the rebase for 3.8. Nothing to see here!