Open Stack

On 4 October 2013 11:00, Dan Prince <dprince at redhat.com> wrote:
> Hi Robert,
>
> In general I buy the vision laid out in this email. I think "Starting with the customer story" will keep us on the right track as to what features to implement, working on most important stuff, etc. A CD tripleO setup sounds just grand. For me though most of what you've laid out here is like kicking it up two notches (not just one). Not saying I don't think we should work towards this... just wondering if it is the most important thing at the moment. What I mean is... with all the instability in keeping TripleO working on a weekly basis would we be better served putting our resources on the CI front first. And once we have stability... then we kick things up another notch or two? Or perhaps we do both of these in parallel?
>
> I like the idea of multiple lines of defense... but given limited resources I wonder if some simpler CI doesn't trump CD at this point.

Thats a really good question. I think if we analyze tripleo, we see 3
major stories in-play and non-gating-CI'd, and not-actually-CD'd.
 - deploy an overcloud
 - deploy an undercloud
 - bootstrap a seed

Some issues, like a nova-client bug, will break all three, most issues
we hit either break heat related things, so over-and-under, or break
baremetal, so seed-and-under.

I guess my thinking on this is that by:
 - reducing this to one story - CD an overcloud using a manual static undercloud
 - rigorously fixing the story *and* putting in place things that make
the story break - CI/design changes/whatever

We will at least stabilise the end user valuable aspect. Once thats
there and doing it's thing, we bring in the next story, so what we're
maintaining looks like:
 - CD an overcloud using a CD'd undercloud done using a manual static seed
 - rigorously fix this story and put in place things to stop it
breaking - CI/design changes/whatever

The net effect of this engineering wise is that we're focusing on less
battles and we'll drive things like CI for each story through to
completion *before* we move onto the next story : breaking the
deadlock we've had with
test-all-the-things-all-the-time-and-gosh-it's-hard.

That said, https://etherpad.openstack.org/tripleo-test-cluster is
aiming straight at a full end to end test story for tripleo at the
moment, but we may want to modify it to get earlier delivery of CI -
once we're at the point of having it run things at all, of course.

-Rob

-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud