Open Stack

On Tue, Apr 04, 2017 at 04:01:48PM -0400, Emilien Macchi wrote:
> After our weekly meeting of today, I found useful to share and discuss
> our roadmap for Container CI jobs in TripleO.
> They are ordered by priority from the highest to lowest:
> 
> 1. Swap ovb-nonha job with ovb-containers, enable introspection on the
> container job and shuffle other coverage (e.g ssl) to other jobs
> (HA?). It will help us to get coverage for ovb-containers scenario
> again, without consuming more rh1 resources and keep existing
> coverage.
> 2. Get multinode coverage of deployments - this should integrate with
> the scenarios we already have defined for non-container deployment.
> This is super important to cover all overcloud services, like we did
> with classic deployments. It should be non voting to start and then
> voting once it works. We should find a way to keep the same templates
> as we have now, and just include the docker environment. In other
> words, find a way to keep using:
> https://github.com/openstack/tripleo-heat-templates/blob/master/ci/environments/scenario001-multinode.yaml
> so we don't duplicate scenario environments.
> 3. Implement container upgrade job, which for Pike will be deploy a
> baremetal overcloud, then migrate on upgrade to containers. Use
> multinode jobs for this task. Start with a non-voting job and move to
> the gate once it work. I also suggest to use scenarios framework, so
> we keep good coverage.
> 4. After we implement the workflow for minor updates, have a job with
> tests container-to-container updates for minor (rolling) updates, this
> ideally should add some coverage to ensure no downtime of APIs and
> possibly checks for service restarts (ref recent bugs about bouncing
> services on minor updates)
> 5. Once Pike is released and Queens starts, let's work on container to
> containers upgrade job.
> 
> Any feedback or question is highly welcome,

+1, I think this makes sense and is well aligned with what we discussed in
the meeting.

I agree the priority is roughly in the order listed above, but provided we
have sufficient folks willing to help we can probably work on some of these
tasks in parallel, as really we need at least (1), (2) and (3) ASAP.

I've started looking at (4) but there is significant work required to
enable this as our current breakpoint based update workflow won't work, and
it looks like we also can't use the rolling update feature of
SoftwareDeploymentGroup, because we want each node to be fully updated
before moving on to the next.

Steve