Comment # 3 on bug 1095131 from Richard Brown

(In reply to Thorsten Kukuk from comment #2)
> (In reply to Maximilian Meister from comment #1)
> > @richard any idea what could be the difference here? or should we test it
> > again on kubic with 1.10.4? last test was done with 1.10.3 IIRC
> 
> SLE12 SP3 (CaaSP until v3) has /var/lib/kubelet as subvolume
> Kubic (CaaSP from v4) has /var as subvolume and /var/lib/kubelet is a
> directory inside this subvolume.
> 
> I bet that this is what confuses kubernetes.

Indeed - my guestimate suggests that
https://github.com/google/cadvisor/pull/1668 only works if /var/lib/kubelet is
it's own subvolume

It's only a guestimate because I really don't understand how go's 'stat' works,
so I'm little blind as to how that fix worked in the past.

But one thing we can say for sure is that it doesn't work on Kubic and the
difference in the subvolume layout is the biggest change that is likely to
trigger any difference in logic for volume/partition ID detection.

That change isn't just present in Kubic - we can expect similar behaviour in
any SLE 15 based CaaSP also (eg. CaaSP v4)

So I'd recommend running any conformance tests for 1.10.x against both SLE
12/CaaSP v3 and SLE 15/Kubic/CaaSP v4 - assuming both are being targetted for
k8s 1.10 releases.

Bumping up the severity and priority on the grounds of Kubic/CaaSP v4 without
kubernetes is as useful as a submarine with a sunroof or an inflatable
dartboard ;)

I'd recommend the bug be considered equally important for CaaSPv4 until it's
proven that it doesn't exist there.