iotests failing on gitlab CI check-system-fedora job

iotests failing on gitlab CI check-system-fedora job

[Peter Maydell]

The check-system-fedora job in the gitlab CI seems to have
started reliably failing on iotests 040 041 127 256 267:
 https://gitlab.com/qemu-project/qemu/-/jobs/1106977551

Could somebody have a look at what's happening, please?
(This is probably a regression that's got into master because
I stopped using "did gitlab CI pass" as a gate because gitlab
was running massively too slowly to be usable for that.)

thanks
-- PMM

Re: iotests failing on gitlab CI check-system-fedora job

[Paolo Bonzini]

On 17/03/21 23:23, Peter Maydell wrote:
> The check-system-fedora job in the gitlab CI seems to have
> started reliably failing on iotests 040 041 127 256 267:
>   https://gitlab.com/qemu-project/qemu/-/jobs/1106977551
> 
> Could somebody have a look at what's happening, please?
> (This is probably a regression that's got into master because
> I stopped using "did gitlab CI pass" as a gate because gitlab
> was running massively too slowly to be usable for that.)

It's caused by adding virtio devices to the m68k target.  Probably it 
can be fixed by moving i386-softmmu to Fedora and m68k-softmmu to Debian.

But really, most of the tests we run are not ever going to fail in 
isolation.  Do we really gain anything by building all or almost all 
targets for each cross (as opposed to just one 32-bit and one 64-bit 
target), or by running check-block many times on similar or identical 
configurations?

Paolo

Re: iotests failing on gitlab CI check-system-fedora job

[Peter Maydell]

On Wed, 17 Mar 2021 at 23:26, Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 17/03/21 23:23, Peter Maydell wrote:
> > The check-system-fedora job in the gitlab CI seems to have
> > started reliably failing on iotests 040 041 127 256 267:
> >   https://gitlab.com/qemu-project/qemu/-/jobs/1106977551
> >
> > Could somebody have a look at what's happening, please?
> > (This is probably a regression that's got into master because
> > I stopped using "did gitlab CI pass" as a gate because gitlab
> > was running massively too slowly to be usable for that.)
>
> It's caused by adding virtio devices to the m68k target.  Probably it
> can be fixed by moving i386-softmmu to Fedora and m68k-softmmu to Debian.

Why are the iotests not specifying what machine type they run
on if they care about the machine type ?

thanks
-- PMM

Re: iotests failing on gitlab CI check-system-fedora job

[Paolo Bonzini]

On 18/03/21 11:07, Peter Maydell wrote:
> On Wed, 17 Mar 2021 at 23:26, Paolo Bonzini <pbonzini@redhat.com> wrote:
>>
>> On 17/03/21 23:23, Peter Maydell wrote:
>>> The check-system-fedora job in the gitlab CI seems to have
>>> started reliably failing on iotests 040 041 127 256 267:
>>>    https://gitlab.com/qemu-project/qemu/-/jobs/1106977551
>>>
>>> Could somebody have a look at what's happening, please?
>>> (This is probably a regression that's got into master because
>>> I stopped using "did gitlab CI pass" as a gate because gitlab
>>> was running massively too slowly to be usable for that.)
>>
>> It's caused by adding virtio devices to the m68k target.  Probably it
>> can be fixed by moving i386-softmmu to Fedora and m68k-softmmu to Debian.
> 
> Why are the iotests not specifying what machine type they run
> on if they care about the machine type ?

Because the intersection of people building only "uncommon" targets and 
people running "check-block" is usually empty.  Adjusting 
tests/qemu-iotests/testenv.py should be quite easy though.

Paolo