test: Also wait for virt-install to finish on centos-10

[mvollmer]

No description provided.

[mvollmer]

Aha, TF centos-10-stream is green.

[mvollmer]

The (guessed) fix makes it reliably green: https://cockpit-logs.us-east-1.linodeobjects.com/pull-1765-67f81cd5-20240813-112925-centos-10/log.html

[martinpitt]

If that works, I'm all for it! I had to make it conditional back then as waiting timed out on several OSes. I've never fully understood how virt-install works -- sometimes it leaves the VM running, sometimes not, I found it rather unpredictable. Perhaps by now all OSes have a new enough virt-install.

Thanks!

[martinpitt]

Yeah, what I feared -- this needs to stay conditionalized. Perhaps there's a better criterion, like the virt-install version? But for now just extending the OS list seems ok.

[mvollmer]

Yeah, what I feared -- this needs to stay conditionalized. Perhaps there's a better criterion, like the virt-install version? But for now just extending the OS list seems ok.

Ah, I see, bummer. So the virt-install process itself sticks around forever? I have one (1) closer look and then I make this conditional.

[mvollmer]

On some OSes the created VM crashes in the
kernel and virt-install never exits, on others, the BIOS can't find a bootable
device but virt-install does exit. shrug

Maybe we should make it so that virt-install can actually produce a working VM.

[mvollmer]

the BIOS can't find a bootable device but virt-install does exit.

The might be the state after a "successful" installation of our fake Fedora 28 OS. So virt-install is indeed done at this point.

[mvollmer]

Oh, the flakes...

[martinpitt]

I don't have proof that this affects rhel-10-0 (or not), as the last gating run in real RHEl 10 all failed due to SELinux. But it almost surely needs the same treatment. I triggered a rhel-10-0 bots run.

[martinpitt]

The rhel-10-0 run confirms this. I suggest to write this as

self.machine.image.startswith(('opensuse-tumbleweed', 'centos-10', 'rhel-10'))

so that it also catches future rhel-10-* images.

[mvollmer]

I have done that, and also increased the global timeout. I think opensuse runs into it.

[martinpitt]

Thanks for your investigations! The comment clarifies it a little indeed! (For our poor future selves -- this will come back and bite us in the ... a place where it hurts)

[mvollmer]

The comment clarifies it a little indeed!

The installation is not expected to succeed, but I don't know how exactly it is expected to fail. On some OSes, a kernel is clearly started, so we must be downloading something? I think this is worth figuring out and robustifying, eventually.

[martinpitt]

I reviewed the failures:

• The virtqemud crashes in c10 and r10 are very persistent. We need to report these and hopefully find a CLI reproducer. But unrelated to this PR.
• r10 testCopyStorageMigration is a known naughty in f40, let's copy it -- but that'll be a bots PR.
• d-testing also fails testCreateDownloadAnOS -- given that it's also a "new" distro similar to tumbleweed and R10, my gut feeling is that we should extend the hack from this PR there. But let's do that in a follow-up.

So in summary, I think this is fine to land, it makes things incrementally better.

[martinpitt]

Above follow-ups: #1767 and cockpit-project/bots#6735