Re: [PATCH 0/5] QEMU Gating CI

[Andrea Bolognani]

On Mon, 2020-04-27 at 01:24 -0400, Cleber Rosa wrote:
> On Fri, 24 Apr 2020 08:57:46 +0200
> Erik Skultety <address@hidden> wrote:
> > On Thu, Apr 23, 2020 at 11:28:21PM +0200, Philippe Mathieu-Daudé
> > wrote:
> > > Thanks to insist with that point Daniel. I'd rather see every
> > > configuration reproducible, so if we loose a hardware sponsor, we
> > > can find another one and start another runner.
> > > Also note, if it is not easy to reproduce a runner, it will be very
> > > hard to debug a reported build/test error.
> > 
> > (Thanks for bringing ^this point up Philippe)
> > 
> > ...However, what we've been actively working on in libvirt is to
> > extend the lcitool we have (which can spawn local test VMs) to the
> > point where we're able to generate machines that would be the
> > reproducible. Right now I'm playing with cloud-init integration with
> > lcitool (patches coming soon) that would allow us to use the same
> > machines locally as we'd want to have in, say, OpenStack and share
> > them as compressed images, so even when updated/managed by lcitool
> > locally, you'd get the same environment.
> 
> This is great, and it aligns with my previous point that reproducibility
> it's not *just* about the hardware, but about diligently documenting
> and automating the environments, be them mundane or super specialized.
> And IMO that should score some points when it comes to being
> promoted/demoted as a Gating machine/job.

I think there's room to extend and rework lcitool so that it can be
used for QEMU CI as well, and we should definitely look into that.

Right now it only really covers VMs and containers, but there's
already one situation (FreeBSD) where the expectation is that you'd
import an existing VM image and then apply CI-related customizations
on top, so it's not too much of a stretch to imagine doing the same
for a bare metal machine. We use Ansible, so as long as we can
connect to the machine via ssh we're pretty much good to go.

Installation of VMs we already perform in an unattended fashion using
preseed/kickstart, and it should be relatively straighforward to
adapt those configurations to also work on real hardware. This way
we'd both be able to rely on having a sane OS as the base, and
relieve the administrator from the duty of manually installing the
machines.

-- 
Andrea Bolognani / Red Hat / Virtualization