On 05/06/17 09:43, Sean Dague wrote:
On 06/01/2017 06:09 AM, Chris Dent wrote:
<snip>
It's clear from this thread and other conversations that the
management of tempest plugins is creating a multiplicity of issues
and confusions:
* Some projects are required to use plugins and some are not. This
creates classes of projects.
While this is true, there are also reasons for that. We decided to break
up the compute service into distinct parts years ago to help let each
part grow dedicated expertise (images, networking, block storage).
However, there is a ton of coupling here even though these are broken up.
My continued resistance to decomposing the QA side of those projects is
getting that integration testing right, and debugging it is hard,
because there are so many interactions required to have a working server
started. And Nova, Neutron, Cinder are the top three most active
projects in OpenStack. So the rate of change individually is quite high.
Forcing those services out into plugins because of the feeling that
Presumably that could be addressed by splitting the
Nova/Neutron/Cinder/Glance tests not used by the OpenStack Powered
trademark program (aka DefCore) into a combined base-compute Tempest
plugin though?
Or are you saying there's coupling between the Defcore and non-Defcore
tests that isn't mediated by tempest-lib? If that's the case then I'd be
concerned that we might run into this problem again in the future.
something doesn't look fair on paper is just generating more work to
create spherical elephants, instead of acknowledging the amount of work
the QA team has on it's shoulders, and letting it optimize for a better
experience by OpenStack users. Especially given limited resources.
-Sean
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: [email protected]?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
