Sorry I haven't been talking much about this here, but there have
been
discussions going in the back-channel around what we need to do for
putting up Gerrit and Jenkins.
Why all the fast moves? I think the point is to be nimble, do what we
know to do, and make things as breakable, fixable, and replaceable as
we can. We're getting things going in record time, as we need to be
that way for now.[0]
We'll have some breathing time mid-November to rethink how we are
approaching infrastructure. We have to recognize that many people come
to this community, and we want an infrastructure that is welcoming to
them. For example, I'm hoping more infrastructure and services will
come from some of the strategic companies involved, just as I hope to
see more from Red Hat IT.
So the back-channel discussions we've had so far settle on us using a
few VM images on EC2. Red Hat is providing these, we'll load them with
RHEL 6 + EPEL, and we'll go in to trim and harden along the lines of
the community services infrastructure[1] (CSI) I'd like us to work
with. Volunteers?
We'll need to work out the configuration details for Gerrit and
Jenkins. They are both new to me, but I'm not afraid. :) So ... what
should the architecture of these look like?
I'm thinking we have two instances:
*
jenkins.ovirt.org - a low-running, very tight image that can scale
to great heights when crunching tests. I'm not going to push an app
server here - small is good, I presume, but I wouldn't mind learning
an EE6 platform, such as JBoss AS 7. :) Still, that's our own
overhead we manage ...
from our experience we will need multiple machines set up as jenkins
slaves to run the expected workload.
jenkins has a plugin to stop/start the instances in EC2 per required
load (didn't test how good it is yet)
*
gerrit.ovirt.org git.ovirt.org - trimmed as a web server. It doesn't
need a database, right?
git.ovirt.org is just an alias.
we are using a lightly forked version today for improved emails to patch
mailing list until our patches are accepted (hopefully) in gerrit upstream.
the change allows patches in the mailing list to contain the patch
comments inline like in regular patch reviews.