On Mon, Mar 09, 2020 at 10:32:34PM -0400, Cleber Rosa wrote:
> 
> 
> ----- Original Message -----
> > From: "David Gibson" <david@gibson.dropbear.id.au>
> > To: "Alex Bennée" <alex.bennee@linaro.org>
> > Cc: qemu-devel@nongnu.org, "Wainer dos Santos Moschetta" <wainersm@redhat.com>, "Kamil Rytarowski"
> > <kamil@netbsd.org>, "Hervé Poussineau" <hpoussin@reactos.org>, "Cleber Rosa" <crosa@redhat.com>,
> > qemu-ppc@nongnu.org, "Philippe Mathieu-Daudé" <philmd@redhat.com>
> > Sent: Monday, March 9, 2020 9:41:16 PM
> > Subject: Re: [PATCH] tests/acceptance/ppc_prep_40p: Use cdn.netbsd.org hostname
> > 
> > On Fri, Feb 28, 2020 at 04:10:19PM +0000, Alex Bennée wrote:
> > > 
> > > Philippe Mathieu-Daudé <philmd@redhat.com> writes:
> > > 
> > > > Use NetBSD content delivery network to get faster downloads.
> > > 
> > > Even with this patch I get failures on my big dev box:
> > > 
> > >  (48/67)
> > >  tests/acceptance/ppc_prep_40p.py:IbmPrep40pMachine.test_openbios_and_netbsd:
> > >  INTERRUPTED: Failed to fetch NetBSD-7.1.2-prep.iso.\nRunner error
> > >  occurred: Timeout reached\nOriginal status: ERROR\n{'name':
> > >  '48-tests/acceptance/ppc_prep_40p.py:IbmPrep40pMachine.test_openbios_and_netbsd',
> > >  'logdir': '/home/alex/lsrc/qemu.git/builds/all/tests/results/jo... (60.31
> > >  s)
> > > 
> > > I think ultimately a whole ISO download is just too much for an
> > > acceptance test.
> > 
> > I tend to agree.  Here in a network-remote part of the world, these
> > always seem to cause timeouts and other problems, in a bunch of the
> > testcases.
> > 
> > Those are testing useful things though, so I'd really like to see the
> > downloads split out into some sort of preparation step that can be
> > done just once, rather than part of the test proper.
> 
> We have added functionality in the latest Avocado that will let us
> easily set a "cancel this test if the ISO has not being previously
> downloaded", or "cancel if it fails to be downloaded during the
> test".

That certainly sounds like an improvement.  I still don't love the
idea that the test results will vary based on something outside of the
tested code, even if a cancel is definitely better than a fail.

> +Willian can explain how it works, and if found to be suitable, and work
> on a patch.

Ok.  That would be good to know - will there be some command to run in
advance to attempt to download all the necessary images?  Would that
run without a timeout - or at least a much longer timeout than the
testcase itself has.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson