[Avocado-devel] Question: qemu virtio_console tests pass port name, instead of port id, to QMP?

[Cleber Rosa]

On 9/12/18 2:53 AM, Xiong, Huan wrote:
> Hi,
> 
> While running avocado-vt qemu virtio_console test, I observe this test failure:
> 
> type_specific.io-github-autotest-qemu.virtio_console.spread_linear.specifiable.virtserialport.with_vm.interrupted_transfer.replug_recv.short: FAIL
> 
> It failed because it passed device name, instead of device id, to device_del command:
> 
> 2018-09-12 01:26:51,201 qemu_monitor     L0334 DEBUG| (monitor avocado-vt-vm1.qmpmonitor1) Sending command 'device_del vs2' 
> 2018-09-12 01:26:51,209 qemu_monitor     L1690 DEBUG| Send command: {'execute': 'device_del vs2', 'id': 'qd1yhYb2'}
> 2018-09-12 01:26:51,215 virtio_console   L0889 ERROR| interrupted_loopback failed with exception: QMP command 'device_del vs2' failed    (arguments: None,    error message: {u'class': u'CommandNotFound', u'desc': u'The command device_del vs2 has not been found'})
> 
> Note "vs2" in above log is virtio serial port name defined in virtio_console.cfg. The device's id is a random string generated in VM.create(). See log below:
> 
> MALLOC_PERTURB_=1  /usr/bin/qemu-system-aarch64 \
>     ...
>     -chardev socket,path=/var/tmp/avocado_Qppix1/virtio_port-vs2-20180911-221535-jIZMYhJ2,nowait,id=idZPwyah,server \
>     -device virtserialport,id=idRvU3kn,name=vs2,chardev=idZPwyah \
>     ...
> 
> It seems to me the root cause of the issue is with the code in VM.create() in avocado-vt/virttest/qemu_vm.py. VM.create() first calls make_create_command() to create device information, which are later saved in self.devices, and generate QEMU command line and start QEMU process. Then when it continues to create VirtioConsole and VirtioSerial instances and add them to self.virtio_ports, it doesn't uses the device information saved in self.devices, but uses port id as port name instead. The above test retrieves port information from vm.virtio_ports and gets wrong id and hence QMP command failed (BTW, the 'CommandNotFound' error returned by QMP seems odd to me. I'd expect it be something like "Device not found". But I think that is a separate issue, which I'll look into).
> 
> So, from my understanding, line 2915-2938 (the code that creates VirtioConsole and VirtioSerial instances and add them to self.virtio_ports) is the culprit. However, that part of the code has existed since 2012. So...am I missing something? I really can't see how it could work with the tp-qemu's virio-console test's default cfg file.
> 
> Also note that I run the test on an aarch64 server running centos 7.4. But I don't think the test code in question is architecture specific.
> 
> Thanks,
> Ray
> 
> 

Hi Ray,

I think you shouldn't underestimate bugs resilience - they can indeed
live in some code bases for decades :).

I believe that the best approach is to put together a PR with this
explanation, and the proposed changes, of course.  Then people can check
the before/after by running the test you describe.

Let me know if you need help reviewing it.

Regards,

-- 
Cleber Rosa
[ Sr Software Engineer - Virtualization Team - Red Hat ]
[ Avocado Test Framework - avocado-framework.github.io ]
[  7ABB 96EB 8B46 B94D 5E0F  E9BB 657E 8D33 A5F2 09F3  ]