[Pkg-libvirt-maintainers] Bug#848247: Bug#848247: Collection of fixes for libvirt smoke tests

[Christian Ehrhardt]

On Thu, Dec 15, 2016 at 6:26 PM, Guido Günther <agx at sigxcpu.org> wrote:

> Thanks for these. I've pulled parts in but I thins some are not needed:
>

Thanks for all the applied's removing those and continuing discussion on
the remaining ones.


> >
> > d/t/control, d/t/smoke-qemu-session, d/t/smoke-lxc: fix up smoke tests
> >
> > ERR: error: Failed to connect socket to '/var/run/libvirt/libvirt-sock':
> No
> >      such file or directory
> > FIX: install libvirt-daemon-system
>
> No, we run as non-root so (so we can use qemu:///session)
>

Yeah - I think I'm good with that.
Initially when in the autopkgtest env I think I missed to export that
properly.
I later fixed it up for my tests on the lxc tests and things were ok.

One thing I also found without "needs-root" was that the
`virt-host-validate qemu` failed on /dev/kvm being root/root and thereby
not accessible from non-root.
Is /dev/kvm set up differently by default on Debian so that it is
accessible to you?

Other than the discussion I think I need adapt and rerun when I re-merge to
your next release with the other fixes accepted here in place.
I'd come back to this bug with more detail then if I still would need
anything of the needs-root/libvirt-daemon-system.

> ERR: arch can be <arch>x86_64</arch>
> > FIX: modify grep "arch name='x86_64'" -> "arch.*x86_64""
>
> Which versions did you see this with?
>

In terms of qemu still an older one as merging that will follow libvirt, so
atm a combination of libvirt 2.5-experimental and qemu 2.6.1.
Both with Ubuntu Delta, but I'd doubt that would change this part of the
behavior.
But if it is the qemu version - then for all sorts of backporting activity
(if ever) it might be useful to have the more relaxed regexp I suggested.


> > ERR: potential race, guest dies on most non bare metal platforms after a
> few
> >      seconds
> > FIX: well, not a fix but add FIXME comment on 2nd level virtualization
> >      issues
>
> I'm not sure how this helps here.
>

Well it does not "help", it was just a reminder for myself when going
through the code.
I'm perfectly fine not applying that one for now and just forgetting it.

A real fix would be a guest that does not insist on all the features and
would boot up just fine (I haven't found kernel parameters to do so and
building an extra guest for that seems over-engineering).


> ERR: potential race on the check of lxc log after start
> > FIX: lacking a better fix a sleep and sync to closen the remaining
> > race window
>
> Hmm...did you see this? If so we should rather check for running again
> with virsh and then check the log.
>

No I haven't seen it in this test, but I had an almost equal test a while
ago (start, check for running) which worked on x86.
But then when the tests spread to other architectures it sometimes failed
on ppc64el and almost always on s390x.

We could convert it to a retry loop that does
while ! check-running
   sleep
   if > retry FAIL

The same would probably apply



-- 
Christian Ehrhardt
Software Engineer, Ubuntu Server
Canonical Ltd
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-libvirt-maintainers/attachments/20161216/f0175c3e/attachment-0001.html>