service is included in V3 catalog even with no endpoints

[Dave Chen]

This is actually the following up of the comment in this patch (https://review.openstack.org/#/c/144860/19/keystone/catalog/backends/sql.py).

In current source, we just keep the service in the catalog with no endpoints, we should skip the services as Henry's comments.

[David Stanek (dstanek)]

Dave, if this is something that you are working on feel free to mark it as In Progress.

[David Charles Kennedy (dkennedy-p)]

My patch set to move the endpoint filtering into the default driver might address this for the v3 catalog where a filtered catalog is returned. The only time that services without endpoints may be returned would be where the catalog is empty due to filtering and keystone is configured to return an unfiltered catalog.

https://review.openstack.org/#/c/167675/5

I only mention this to avoid a potential duplication of effort.

[Dave Chen (wei-d-chen)]

@David, I will watch that patch, thanks for the information.

[David Charles Kennedy (dkennedy-p)]

I'm a bit puzzled about this. I can make this work with my patch set for endpoint filtering by simply using .join for the query (which does an inner join between Service and Endpoint rather than adding the .options(joined_load) which does a left outer join. No endpoints = no service. But doesn't that remove the value of CONF.endpoint_filter.return_all_endpoints_if_no_filter = True? In the case of the left outer join if the filtered list is empty then the unfiltered service list would only contain services for which there are no endpoints assigned to the project - in which Henry's comment suggests we shouldn't return them anyway.

I presume that configuration property is there for a reason - though I can't fathom what that might be.

[Dave Chen (wei-d-chen)]

@David Charles Kennedy, acutally, this bug existed in both SQL and templates.

If this bug bloats your patch (https://review.openstack.org/#/c/167675/), maybe you can consider to address it in a separated patch.

[Dave Chen (wei-d-chen)]

The issue may happened after URL formatting as long as all of the URLs has none suitable replacement.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/176383

[Sean Perry (sean-perry-a)]

I am currently working on this in order to get #1405574 resolved.

@dkennedy, may I reassign?

[David Charles Kennedy (dkennedy-p)]

By all means @sean-perry as I have limited bandwidth for this right now. The patch set I submitted is functional but it has been suggested that I re-write to implement a builder for the catalogs, which makes sense. It's also dependent on another patch but you might like to take a look:
https://review.openstack.org/#/c/176383/

[Steve Martinelli (stevemar)]

unassigning due to inactivity

[Morgan Fainberg (mdrnstm)]

This is being marked as "wont fix" since it is a behavior we currently have. Changing it would be incompatible.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Morgan Fainberg (<email address hidden>) on branch: master
Review: https://review.openstack.org/176383
Reason: administrative abandoning.