Open Stack

Hi,

Clark responded with some technical questions and suggestions about
some first steps we should take before we proceed.  He also had some
general thoughts and concerns which I would like to elaborate on.

We operate the CI system as a single multi-cloud application.  Our unit
of fault-tolerance is the cloud itself.  When something goes wrong with
one of the public clouds we are using, we simply stop using that cloud
until it is corrected.  In our experience, cloud-level disruption is not
uncommon.  So it is important to us that the service we provide to the
project is able to be supplied by more than one cloud to maintain this
fault tolerance.  As it stands, of the two providers we currently use,
Rackspace is the only one that provides bare-metal servers.  I hope that
will change in the future, but it is the case today.

Having said that, it is not necessarily a showstopper.

As Clark mentioned, we have incorporated specialized hardware resources
into our system before.  The policy we adopted is that such resources
can not be used for gating, they can only be used for advisory testing
(ie, as in the "Check" pipeline), and we must create a new pipeline
specifically for those resources, so that existing "Check" and "Gate"
builds are not affected if there is a problem with the specialized
resources.  In effect, our policy is that we must be able to run tests
on multiple clouds in order for them to be in the normal check/gate
pipelines, but additional resources can be used in other pipelines as
long as they don't impact the ability to merge changes in a project --
because those resources are provided with a lower expectation of
availability.

Finally, it is worth considering the scope of what these resources would
be used for.  Our remit is to provide resources to the entire project,
so generally we would not manage resources for the use of one specific
project except on an experimental basis before opening it to the wider
project.  I hope that Rackspace would be happy to support OpenStack as a
whole with these resources as it has with other contributions.

I hope that clarifies a few things.  Thank you for starting this
conversation here with us and trying first to do this work upstream!  I
do think the way to proceed is to try to clarify the things Clark
mentioned in his reply and go from there.  I will add that if for some
reason this doesn't work out, there is the option of third-party
testing, where you could operate a test system yourself that reports on
changes.  But let's consider that a last resort.

Thanks,

Jim