Open Stack

On Thu, Mar 23, 2017 at 7:02 AM, Ken'ichi Ohmichi <ken1ohmichi at gmail.com>
wrote:

> 2017-03-22 14:32 GMT-07:00 Andrea Frittoli <andrea.frittoli at gmail.com>:
> >
> >
> > On Wed, Mar 22, 2017 at 8:31 PM Sean McGinnis <sean.mcginnis at gmx.com>
> wrote:
> >>
> >> On Wed, Mar 22, 2017 at 01:08:23PM -0700, Ken'ichi Ohmichi wrote:
> >> > Hi,
> >> >
> >> > Now we need to update Tempest for following Cinder API status.
> >> > I have an idea for restructuring and happy to see feedback about that.
> >> >
> >> > Now Cinder API status is
> >> >   V1: Deprecated
> >> >   V2: Deprecated
> >> >   V3: Current
> >> > V1 API tests have been removed from Tempest side already, so we just
> >> > need to concentrate on V2 and V3 now.
> >>
> >> >
> >> > **Gate jobs**
> >> > Most Cinder tests are implemented for V2 API on Tempest side and the
> >> > base microversion of V3 is the same as V2.
> >> > Then we can re-use V2 API tests for the base microversion of V3 API.
> >> > One idea is that we can have Cinder V3 API tests as the default on the
> >> > gate jobs and the V2 API tests as another job like the following
> >> > because the V2 API is deprecated.
> >> >
> >> >   gate-tempest-dsvm-neutron-full-ubuntu-xenial - (existing job):
> >> > testing Cinder V3 API
> >> >   gate-tempest-dsvm-py35-ubuntu-xenial - (existing job): testing
> Cinder
> >> > V3 API
> >> >   ...
> >> >   gate-tempest-dsvm-neutron-full-ubuntu-xenial-cinder-v2: (new job):
> >> > testing Cinder V2 API
> >> >
> >>
> > I guess this job would run against tempest and cinder only?
>
> A nice point, I think so.
>
> >> +1 I like this idea.
> >>
> >> > We had the same testing way for Nova V2 API and V2.1 API before, and
> >> > we could avoid copy&paste V2 test code for V2.1 API on Tempest.
> >> >
> >> > **Test Structure**
> >> > Current test structure is like:
> >> >   tempest/api/volume/  - V2 API tests
> >> >   tempest/api/volume/v2 - V2 API tests
> >> >   tempest/api/volume/v3 - V3 API tests
> >> > Yes, this is mess.
> >> > For re-using V2 API tests for V3 API, it would be better to remove
> >> > "v2" from V2 API tests for avoiding confusions.
> >> >
> >> > A new structure could be
> >> >   tempest/api/volume/  - All tests for V2 API and the base
> >> > microversion of V3 API
> >> >   tempest/api/volume/v3 - V3 API specific tests for newer
> microversions
> >> > or
> >> >   tempest/api/volume/  - All tests for V2 API and V3 API which
> >> > includes newer microversions
>

​+1, this looks better as no more version specific tests and all v2 tests
should run as it is on v3 base version.​



> >> >
> >> > As the reference, Nova API structure is like the later.
> >>
> >> I like the last one better as well.
> >>
> > My favourite option would be that that generates less churn in the code
> :)
> > One folder for everything means moving 4 or 5 modules only, so I think
> that
> > would
> > be a good option.
> >
> > I would prefer to avoid though having a lot of v3 test classes that
> inherit
> > from
> > v2 test classes, and just set _api_version = 3.
>
> Yeah, I agree :-)
>


​yea we should not have that.



>
> > As long as we can assume we will never run v2 and v3 in the same job, we
> > could
> > have cinder_api_version as a configuration setting, to determine which
> > cinder
> > endpoint to hit when running the volume tests.
>
> Or it would be enough to have the existing "catalog_type",
> "min_microversion" and "max_microversion" only without api_v1/v2/v3 to
> control the target API version, because of the above separated gate
> jobs.
>
>
​Yes, so devstack does set different catalog for v2 and v3 [0]​. Based on
those catalog_type configured on tempest config(we already have that for
volume API config) auth can select the right endpoints to make the API call.

All existing tests can be run for both API without any extra class or
change. This way we can get rid of 'api_version' in all volumes clients and
have them as single copy for v2 and v3 base.
Further v3 microversion tests can be implemented as accordingly by sending
the microversion header on API request. and devstack can tell Temepst not
to set microversion if catalog_type volume_v2 is being asked to run the
tests.

As you mentioned it can be same way we handle compute v2, v2.1 and +
microversion tests.



> > Apart from the volume tests, if we split the gate jobs into standard one
> > running v3
> > plus and extra v2 one, we should make sure that all tests that use the
> > volume API
> > use a consistent version of the volume API. Nova as well should be
> > configured if
> > possible to use the same volume API version.
>
> This also is a nice point.
> Nova team also has a plan to use cinder v3 as the default in Pike.
> We have consistent direction now.
>
> Thanks
>
> __________________________________________________________________________
> 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
>


​.. 0
http://git.openstack.org/cgit/openstack-dev/devstack/tree/lib/cinder#n403

Thanks
gmann​
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20170323/9a95a009/attachment.html>