[openstack-dev] Updating libvirt in gate jobs

Sean Dague sean at dague.net
Wed Mar 19 10:43:40 UTC 2014


On 03/18/2014 08:15 PM, Joe Gordon wrote:
> 
> 
> 
> On Tue, Mar 18, 2014 at 8:12 AM, Sean Dague <sean at dague.net
> <mailto:sean at dague.net>> wrote:
> 
>     On 03/18/2014 10:11 AM, Daniel P. Berrange wrote:
>     > On Tue, Mar 18, 2014 at 07:50:15AM -0400, Davanum Srinivas wrote:
>     >> Hi Team,
>     >>
>     >> We have 2 choices
>     >>
>     >> 1) Upgrade to libvirt 0.9.8+ (See [1] for details)
>     >> 2) Enable UCA and upgrade to libvirt 1.2.2+ (see [2] for details)
>     >>
>     >> For #1, we received a patched deb from @SergeHallyn/@JamesPage
>     and ran
>     >> tests on it in review https://review.openstack.org/#/c/79816/
>     >> For #2, @SergeHallyn/@JamesPage have updated UCA
>     >> ("precise-proposed/icehouse") repo and we ran tests on it in review
>     >> https://review.openstack.org/#/c/74889/
>     >>
>     >> For IceHouse, my recommendation is to request Ubuntu folks to
>     push the
>     >> patched 0.9.8+ version we validated to public repos, then we can can
>     >> install/run gate jobs with that version. This is probably the
>     smallest
>     >> risk of the 2 choices.
>     >
>     > If we've re-run the tests in that review enough times to be confident
>     > we've had a chance of exercising the race conditions, then using the
>     > patched 0.9.8 seems like a no-brainer. We know the current version in
>     > ubuntu repos is broken for us, so the sooner we address that the
>     better.
> 
> 
> 
> ++
>  
> 
>     >
>     >> As soon as Juno begins, we can switch 1.2.2+ on UCA and request
>     Ubuntu
>     >> folks to push the verified version where we can use it.
> 
>  
> ++
>  
> 
>     >
>     > This basically re-raises the question of /what/ we should be
>     testing in
>     > the gate, which was discussed on this list a few weeks ago, and
>     I'm not
>     > clear that there was a definite decision in that thread
>     >
>     >  
>     http://lists.openstack.org/pipermail/openstack-dev/2014-February/027734.html
>     >
>     > Testing the lowest vs highest is targetting two different scenarios
>     >
>     >   - Testing the lowest version demonstrates that OpenStack has not
>     >     broken its own code by introducing use of a new feature.
>     >
>     >   - Testing the highest version demonstrates that OpenStack has not
>     >     been broken by 3rd party code introducing a regression.
>     >
>     > I think it is in scope for openstack to be targetting both of these
>     > scenarios. For anything in-between though, it is upto the downstream
>     > vendors to test their precise combination of versions. Currently
>     though
>     > our testing policy for non-python bits is "whatever version ubuntu
>     ships",
>     > which may be neither the lowest or highest versions, just some
>     arbitrary
>     > version they wish to support. So this discussion is currently more
>     of a
>     > 'what ubuntu version should we test on' kind of decision
> 
>     I think testing 2 versions of libvirt in the gate is adding a matrix
>     dimension that we currently can't really support. We're just going to
>     have to pick one per release and be fine with it (at least for
>     icehouse).
> 
>     If people want other versions tested, please come in with 3rd party ci
>     on it.
> 
>     We can revisit the big test matrix at summit about the combinations
>     we're going to actually validate, because with the various limitations
>     we've got (concurrency limits, quota limits, upstream package limits,
>     kinds of tests we want to run) we're going to have to make a bunch of
>     compromises. Testing something new is going to require throwing existing
>     stuff out of the test path.
> 
> 
> I think this is definitely worth revisiting at the summit, but I think
> we should move Juno to Libvirt 1.2.2+ as soon as possible instead of
> gating on a 2 year old release, and at the summit we can sort out what
> the full test matrix can be.
> 
> As a side note tripleo uses libvirt from Saucy (1.1.1) so moving to
> latest libvirt would help support them.

Honestly, given that we've been trying to get a working UCA for 6
months, I'm really not thrilled by the idea of making UCA part of our
gate. Because it's clearly not at the same level of testing as the base
distro. I think this will be even more so with UCA post 14.04 release,
as that's designed as a transitional stage to get you to 14.04.

As has been demonstrated, Canonical's testing systems are clearly not
finding the same bugs we are finding in their underlying packages.

I think the libvirt 1.2+ plan should be moving Juno to 14.04 as soon as
we can get that stable. That will bring in a whole fresh OS, kernel,
etc. And we recenter our testing on that LTS going forward.

	-Sean

-- 
Sean Dague
Samsung Research America
sean at dague.net / sean.dague at samsung.com
http://dague.net

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 482 bytes
Desc: OpenPGP digital signature
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140319/9f8deb74/attachment.pgp>


More information about the OpenStack-dev mailing list