[openstack-dev] [Cinder] [stable] [all] Changing stable policy for drivers

Sean McGinnis sean.mcginnis at gmx.com
Mon Aug 8 14:41:36 UTC 2016

On Sat, Aug 06, 2016 at 08:00:06PM -0400, Ben Swartzlander wrote:
> On 08/06/2016 06:11 PM, Jeremy Stanley wrote:
> >On 2016-08-06 17:51:02 -0400 (-0400), Ben Swartzlander wrote:
> >[...]
> >>when it's no longer to run dsvm jobs on them (because those jobs
> >>WILL eventually break as infra stops maintaining support for very
> >>old releases) then we simply remove those jobs and rely on vendor
> >>CI + minimal upstream tests (pep8, unit tests).
> >
> >This suggestion has been resisted in the past as it's not up to our
> >community's QA standards, and implying there is "support" when we
> >can no longer test that changes don't cause breakage is effectively
> >dishonest. In the past we've held that if a branch is no longer
> >testable, then there's not much reason to collaborate on code
> >reviewing proposed backports in the first place. If we're reducing
> >these branches to merely a holding place for "fixes" that "might
> >work" it doesn't sound particularly beneficial.
> Well this was the whole point, and the reason I suggested using a
> different branch other than stable/release. Keeping the branches
> open for driver bugfix backports is only valuable if we can go 5
> releases back.

We can debate how far back they should go, but I think initially we can
limit it to less that 5. I haven't done anywhere near a thorough survey,
but my impression is most distros so far have tried to stay with three
releases of the current one.

> I agree the level of QA we can do gets less as releases get older,
> and nobody expects the Infra team to keep devstack-gate running on
> such old releases. However vendors and distros DO support such old
> releases and the proposal to create these branches is largely to
> simplify the distributions of bugfixes from vendors to customers and
> distros.
> Compare this proposal to the status quo, which is that several
> vendors effectively maintain forks of Cinder on github or other
> public repos just to have a place to distribute bugfixes on old
> releases. Distros either need to know about these repos or do the
> backports from master themselves when taking bugfixes into old
> releases.

I think this is a key point. The need for this is to have a common and
known place for these driver fixes to be backported and made available
to whomever needs them. This isn't necessarily saying we are going to
thoroughly test and guarantee that these backports are 100% OK.

>From my understanding, part of what has failed with keeping stable
branches around longer has been the lack of involvement from distros and
vendors to keep those branches up to date. After the EOL date, these
backported driver fixes would be 100% on the vendors to have tested
their drivers and made sure they work. We as a community would just be
provider a place for those updates to go and a known place for those
that need the fixes to find them.

> -Ben
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

More information about the OpenStack-dev mailing list