Open Stack

Ihar Hrachyshka <ihrachys at redhat.com> wrote:

> Hi,
>
> we have this latent bug where we pass objects for subnetpools into  
> extension functions. (That was a mistake made when switching to using  
> objects in base db plugin for the resource.) Now, we want to revert to  
> passing db models as we did before, so that our plugin api is consistent  
> and reflects what we had before.
>
> The patch that would achieve that is at:  
> https://review.openstack.org/#/c/348279/
>
> The patch is still a slight WIP, and also has concerns from @zzzeek  
> around using populate_existing, same concerns as found in Kevin’s patch  
> at: https://review.openstack.org/#/c/364301/
>
> While we are still shaping the patches, it would be wise to land some of  
> dependencies that are just cosmetic changes to objects API and test  
> framework. Specifically, we’ll need:
>
> https://review.openstack.org/#/c/348271/
> https://review.openstack.org/#/c/366052/
>
> Those pass CI right now, and would be actually valuable irrelevant to the  
> result of the discussion with @zzzeek.
>
> Thanks for reviews,
> Ihar

OK, a thing to clarify since I was asked: I believe it’s material for  
Newton, because we should not release it with plugin API changed, even if  
we haven’t hit the issue anywhere in real world (or it just was not  
reported).

Also, we’ve found a way to avoid using populate_existing, that is reflected  
in the latest PS for: https://review.openstack.org/#/c/348279/

Note that there are still several deps for the patch that should go in first:
- https://review.openstack.org/#/c/348271/ cleaning up from_db_object  
signature;
- https://review.openstack.org/#/c/366052/ refactoring object unit tests to  
use db models instead of dicts;
- https://review.openstack.org/#/c/367268/ fixing RBAC metaclass for  
objects to NOT trigger DeclarativeObject.__init__ twice.

Thanks for attention,
Ihar