Re: [PATCH for-10.0] tests/functional: Convert the intel_iommu avocado test

["Philippe Mathieu-Daudé" <philmd@linaro.org>]

On 9/12/24 10:09, Daniel P. Berrangé wrote:
> On Mon, Dec 09, 2024 at 09:26:21AM +0100, Thomas Huth wrote:
>> On 09/12/2024 09.12, Eric Auger wrote:
>>> Hi Thomas,
>>>
>>> On 12/6/24 19:17, Thomas Huth wrote:
>>>> Convert the intel_iommu test to the new functional framework.
>>>> This test needs some changes since we neither support the old 'LinuxTest'
>>>> class in the functional framework yet, nor a way to use SSH for running
>>>> commands in the guest. So we now directly download a Fedora kernel and
>>>> initrd and set up the serial console for executing the commands and for
>>>> looking for the results.
>>>>
>>>> Signed-off-by: Thomas Huth <thuth@redhat.com>
>>>> ---
>>>>    MAINTAINERS                                   |   1 +
>>>>    tests/functional/meson.build                  |   1 +
>>>>    .../test_intel_iommu.py}                      | 119 ++++++++----------
>>>>    3 files changed, 51 insertions(+), 70 deletions(-)
>>>>    rename tests/{avocado/intel_iommu.py => functional/test_intel_iommu.py} (41%)
>>>>    mode change 100644 => 100755


>>>> -        self.launch_and_wait()
>>>> -        self.ssh_command('cat /proc/cmdline')
>>>> -        self.ssh_command('dmesg | grep -e DMAR -e IOMMU')
>>>> -        self.ssh_command('find /sys/kernel/iommu_groups/ -type l')
>>>> -        self.ssh_command('dnf -y install numactl-devel')
>>> I understand you cannot use ssh yet but the bulk of the test was the
>>> execution of the dnf install meaning we lose the main substance of it
>>> through the conversion.
>>
>> Ah, I see, so this was exercising the virtio-net device with the IOMMU ...
>> and I already wondered why there was this "dnf install" at the end without
>> doing anything with  the numactl-devel package ... (a comment would have
>> been helpful here)
> 
> FYI, I find 'dnf instal' to be a *highly* undesirable thing todo in
> our test functional. Its performance is highly non-deterministic
> depending on what mirror you happen to get sent to, such that it could
> easily push us over the timeouts. It is also susceptible to periodic
> broken mirrors, and instability around time of Fefdora EOL. I can't
> remember if it was this test case, or a different one, but I've seen
> problems before in avocado with 'dnf install'.
> 
> If we want to test working networking, then can we arrange for something
> more simple & targetted to run, with better worst case performance.

Could we use 2 virtio-net interfaces inter-connected and stress with
https://linux.die.net/man/1/ab ?