[libvirt PATCH] ci: print test suite logs on failure for Cirrus jobs

[Daniel P. Berrangé]

On Tue, Apr 26, 2022 at 12:07:40PM +0000, Andrea Bolognani wrote:
> On Tue, Apr 26, 2022 at 10:12:17AM +0100, Daniel P. Berrangé wrote:
> > We don't have access to the 'testlog.txt' file, so we need meson to
> > print the failures for any broken tests directly.
> >
> > Signed-off-by: Daniel P. Berrangé <berrange at redhat.com>
> > ---
> >  ci/cirrus/build.yml | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> Reviewed-by: Andrea Bolognani <abologna at redhat.com>
> 
> > The CI pipeline for macOS started failing a few days ago. It was not
> > triggered by any commit, as the pipeline immediately preceeding the
> > first failure used the same commit hash on master. The logs show
> > glib2 being updated from 2.72.0 to 2.72.1
> >
> > With this patch applied I can see the test logs
> >
> >    https://gitlab.com/berrange/libvirt/-/jobs/2376891003
> >
> > and all the failing tests are hitting:
> >
> > (process:50961): GLib-WARNING **: 01:56:14.162: poll(2) failed due to:
> > Bad file descriptor.
> >
> > so something todo with the QEMU monitor/event loop AFAIK, but not
> > sure what.
> 
> Looking at
> 
>   https://gitlab.gnome.org/GNOME/glib/-/releases/2.72.1
> 
> the interesting change seems to be
> 
>   * Fix detection of broken poll() function on macOS
> 
> which would correspond to
> 
>   https://gitlab.gnome.org/GNOME/glib/-/merge_requests/2571

Yep, I since filed a bug

https://gitlab.com/libvirt/libvirt/-/issues/303

The problem is that poll handles bad file descriptors by setting
the event POLLNVAL. ie poll() syscall succeeds.

With the glib BROKEN_POLL macro set, it emulates poll() using
select() and with bad file descriptors that fails hard with
EBADF.

IOW, I suspect there's a latent bug hiding in libvirt tests (or
real code) that is passing a bad FD to poll(), and we've been
lucky that it was harmless with POLLNVAL, and now we're seeing
the bug exposed.

> So it looks like this was very intentional, and motivated by the
> needs of QEMU of all projects... I wonder what they're doing
> differently from us?

Tracing the history back, poll() on macOS fails when used with
FDs associated with a device node under /dev/*.  Not likely
something libvirt does, but certainly  an issue for QEMU with
tap devices.

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|