Open Stack

Thanks for the detailed email on this. How about we add this to the
agenda for this weeks nova meeting?

One option would be to add a fixture to some higher level test class,
but perhaps someone has a better idea than that.

Michael

On Wed, Apr 1, 2015 at 8:54 PM, Markus Zoeller <mzoeller at de.ibm.com> wrote:
> Context:
> --------
> During the Kilo development cycle the "KVM/libvirt on system z" platform
> made some effort to be supported by the libvirt driver [1].
>
> Observation:
> ------------
> Our first tests in a prototype platform CI showed some false negatives
> because some unit tests don't seem to be fully platform independent.
> For example the result of test:
>
>     nova.tests.unit.virt.libvirt.test_driver.LibvirtConnTestCase.
>     test_get_guest_config_without_qga_through_image_meta [0.016369s]
>     ... FAILED
>
>     Captured traceback:
>     ~~~~~~~~~~~~~~~~~~~
>         Traceback (most recent call last):
>           File "nova/tests/unit/virt/libvirt/test_driver.py", line 3112,
>           in test_get_guest_config_without_qga_through_image_meta
>             vconfig.LibvirtConfigGuestSerial)
>             [...]
>             raise mismatch_error
>         testtools.matchers._impl.MismatchError:
>         '<nova.virt.libvirt.config.LibvirtConfigGuestConsole object>'
>         is not an instance of LibvirtConfigGuestSerial
>
> This mismatch makes fully sense if x86 is assumed as default underlying
> platform for unit test execution. The root cause (in this case) is that
> the call of "libvirt_utils.get_arch()" is not mocked and actually speaks
> to the underlying platform. On our system z CI this call returns "s390x"
> which hits platform switches in the code (a search for "arch.S390X" will
> show you all this platform specific code).
>
> A first test
> ------------
> A change of this specific test could look like this:
>
>     # git diff nova/tests/unit/virt/libvirt/test_driver.py
>     diff --git a/nova/tests/unit/virt/libvirt/test_driver.py
>                b/nova/tests/unit/virt/libvirt/test_driver.py
>     index 5fbe5e1..ebcc9ed 100644
>     --- a/nova/tests/unit/virt/libvirt/test_driver.py
>     +++ b/nova/tests/unit/virt/libvirt/test_driver.py
>     @@ -3091,7 +3091,9 @@ class LibvirtConnTestCase(test.NoDBTestCase):
>                                    image_meta,
>                                    disk_info)
>
>     -    def test_get_guest_config_without_qga_through_image_meta(self):
>     +    @mock.patch.object(libvirt_driver.libvirt_utils, 'get_arch',
>     +                       return_value=arch.X86_64)
>     +    def test_get_guest_config_without_qga_through_image_meta(self,
> mock_get_arch):
>              self.flags(virt_type='kvm', group='libvirt')
>
>              drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), True)
>
> Open questions
> --------------
> We currently discovered around 30 test cases (mostly the class
> "LibvirtConnTestCase") which could be treated that way, which seems to
> be cumbersome from my point of view. I'm looking for a way to express
> the assumption that x86 should be the default platform in the unit tests
> and prevent calls to the underlying system. This has to be rewritable if
> platform specific code like in [2] has to be tested.
>
> I'd like to discuss how that could be achieved in a maintainable way.
>
>
> References
> ----------
> [1] https://blueprints.launchpad.net/nova/+spec/libvirt-kvm-systemz
> [2] test_driver.py; test_get_guest_config_with_type_kvm_on_s390;
>
> https://github.com/openstack/nova/blob/master/nova/tests/unit/virt/libvirt/test_driver.py#L2592
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev



-- 
Rackspace Australia