osdir.com


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[tripleo][ci] RFC, supported releases of TripleO


On Tue, Oct 6, 2020 at 4:14 PM Marios Andreou <marios at redhat.com> wrote:

>
>
> On Tue, Oct 6, 2020 at 3:54 PM Jeremy Stanley <fungi at yuggoth.org> wrote:
>
>> On 2020-10-06 11:15:57 +0300 (+0300), Marios Andreou wrote:
>> [...]
>> > main reason for keeping queens is because it is part of the fast
>> > forward upgrade (ffu) ... i.e. newton->queens for ffu 1 and then
>> > queens->train for ffu 2. So indeed the backports will go as you
>> > described - to train then queens
>> [...]
>>
>> Previous discussions highlighted the need for all stable release
>> branches newer than a particular branch to have at least the same
>> level of support or higher. This expectation was encoded into the
>> Stable Branches chapter of the Project Team Guide:
>>
>>
>> https://docs.openstack.org/project-team-guide/stable-branches.html#processes
>>
>> Further, fast forward upgrades aren't designed the way you're
>> suggesting. You're expected to install each version of the software,
>> so to go from Newton to Queens you need to install Ocata and Pike
>> along the way to do the necessary upgrade steps, and then between
>> Queens and Train you need to upgrade through Rocky and Stein. What's
>> unique about FFU is simply that you don't need to start any
>> services from the intermediate branch deployments, so the upgrades
>> are performed "offline" so to speak.
>>
>
> yes you are correct - I am at least a little familiar with ffu I used to
> be part of the tripleo upgrades squad around the time of queens->train "ffu
> 1".
>

nit... ^^^ newton to queens for ffu1 .... queens to train is ffu2

So indeed, you may need to merge branch specific upgrades tasks or fixes
> into the intermediate branches and likely this is why we have not tagged
> older branches (like ocata and pike) as EOL
>
> I think there are two things in this thread. The first from weshay
> original mail, which doesn't clarify what 'removing stable/rocky' means -
> but I believe it means 'removing the ci jobs for stable/rocky' ...
>
> I mentioned EOL and that branches are not tagged as such - to which you
> replied. I now understand both that we should not do that, and likely *why*
> ocata/pike aren't marked eol like some older branches.
>
> Granted, no TripleO deliverables have the stable:follows-policy tag,
>> so as long as you're only referring to which branches of TripleO
>> repositories are switching to EOL you can probably do whatever you
>
> want (assuming the TC doesn't object). If TripleO's upgrade
>> orchestration in stable/train is able to perform the Rocky and Stein
>> intermediate deployments, it would presumably work. The service
>> projects don't have the same luxury however, and so can only EOL a
>
> particular stable branch if all older branches are already EOL.
>>
>
> thanks for the pointers and clarification. I apologise for the confusion -
> as I wrote above, I was mistaken about marking branches eol. We will not be
> doing this.
>
> This thread is about removing the upstream ci/rdo periodic jobs for those
> branches - rocky and stein to be specific
>
>
so... thinking about it some more just now... i think we may need to keep
some minimal ci for both rocky and stein.. in the same way we keep a
smaller subset of our 'normal' jobs for queens. The reason is what fungi
reminded me in his reply above... we may need to merge fixes into both
rocky and stein - for example undercloud upgrade tasks or any other kind of
upgrade task that will be used as part of ffu?

Unless all the ffu upgrade logic lives in the target (train in the ffu2
q->t) branch - but I doubt that for example the upgrade is at least
upgraded sequentially i.e. execute the n upgrade tasks, then n+1 and so on
to the target.

I will point the upgrades squad at this to comment and confirm about the
q-->t ffu2

thanks



> thanks, marios
>
>
>
>> --
>> Jeremy Stanley
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20201006/59d60e24/attachment-0001.html>