[libvirt] [PATCH RFC V2] guests: Add support for openSUSE

[Andrea Bolognani]

On Tue, 2019-12-03 at 11:05 +0100, Fabiano Fidêncio wrote:
> > Speaking of which: I've run into a fairly major problem with
> > the Leap 15.1 support enabled by this patch: while unattended
> > installation works as expected on my laptop (Fedora 31), when I try
> > the same steps on the CentOS CI production environment (CentOS 7.5)
> > it doesn't seem to find or recognize the autoinst.xml file: instead,
> > I get a "please make sure your installation medium is available"
> > message and the chance to input the information manually, which is
> > obviously not what we want :) I'll try to figure out what's going
> > wrong tomorrow.
> 
> So, the error you're facing is not because the autoinst.xml file is
> not recognised. It happens because the installation source is not
> passed.
> 
> virt-install 1.5.0, which is the version present on CentOS 7.5,
> doesn't seem to know the proper kernel URL argument to be passed to
> different distros. So far it worked because it uses "method=URL",
> which is what both CentOS 7 / Fedora (the ones we care about) use and
> Ubuntu / Debian are special with regard to this.
> 
> OpenSUSE, though, requires "install=URL" to be passed. As it doesn't
> happen, the installation asks for a specific installation source.

So that was the problem! Great detective work :)

> Okay, the way to go, IMHO, is patching lcitool to:
> - Have a kernel-url-argument as part of the distro definition;
> - Mount the extra-args based on that;
> 
> Mind that "method=" will always be appended but as long as we provide
> the "install=" to the kernel command line, we're fine.

It sounds like we can just provide install= unconditionally, same as
we currently do with ks=? We'd have to verify doing so doesn't cause
any unintended side-effects for existing guest types.

(Incidentally, didn't CentOS 8 also require passing the installation
URL explicitly based on your tests, or something like that? We need
to get around to adding that OS to CI too at some point O:-)

> I'll cook some patches in the afternoon and post them.
> 
> An important question ... can't we have / use a CentOS 8 machine?
> Would be lovely to not have to work those issues around on lcitool.

I'd love that as well, but I'm afraid that's out of our hands since
the CentOS CI machines are handled by the CentOS CI folks... We could
try asking nicely and see what happens :) But in the short term
having a workaround in lcitool is definitely the way to go.

-- 
Andrea Bolognani / Red Hat / Virtualization