On Tue, 15 Oct 2024 at 12:42, Thomas Huth <[email protected]> wrote:
>
> On 15/10/2024 13.37, Peter Maydell wrote:
> > The ide-test occasionally times out: on the system I run
> > vm-build-openbsd on, it usually takes about 18 seconds, but
> > occasionally hits the 60s timeout, likely when the host machine is
> > under heavy load.  I have also seen this test hit its time limit on
> > the s390x CI runner.
> >
> > Double the timeout for this test so that it won't hit its timeout
> > even when the host is running more slowly than usual.
> >
> > Signed-off-by: Peter Maydell <[email protected]>
> > ---
> >   tests/qtest/meson.build | 1 +
> >   1 file changed, 1 insertion(+)
> >
> > diff --git a/tests/qtest/meson.build b/tests/qtest/meson.build
> > index b207e386965..e8be8b3942d 100644
> > --- a/tests/qtest/meson.build
> > +++ b/tests/qtest/meson.build
> > @@ -4,6 +4,7 @@ slow_qtests = {
> >     'bios-tables-test' : 910,
> >     'cdrom-test' : 610,
> >     'device-introspect-test' : 720,
> > +  'ide-test' : 120,
> >     'migration-test' : 480,
> >     'npcm7xx_pwm-test': 300,
> >     'npcm7xx_watchdog_timer-test': 120,
>
> Reviewed-by: Thomas Huth <[email protected]>
>
> PS: It's a bit weird that we see more and more timeout problems recently ...
> I think I've also seen some local ones on my laptop when my system was under
> load, but not really reproducible. I wonder whether we have a performance
> degradation somewhere in the code...?

If it is, it's one where it's intermittent whether it hits or not
(e.g. I grepped a bunch of my local logs for this test and there's
a lot of 18s and a 30s and a couple of 60s) and it tends to hit for
multiple tests in a single CI run when it does hit. e.g.

https://gitlab.com/qemu-project/qemu/-/jobs/8065512834

https://gitlab.com/qemu-project/qemu/-/jobs/8069311713

are a failure and a pass of the same git hash for "clang-system".
The 'make check' stage on the failure took just under 40 minutes;
the same 'make check' on the passing job took less than 3 !
The 'make' stage also had a similar gap in timing between the
two, and I don't think that's entirely because of ccache caching
effects because the stats at the bottom of the passing job
say the hitrate was >90%.

-- PMM

Reply via email to