Comment 3 for bug 1401437

Looks to me like the changes referenced by Jaime address the issues on the nova side. As for cinderclient... It is great that nova switched to using cinderclient.client.SessionClient, but there remain a bunch of issues with cinderclient.client.HTTPClient which will affect other things that have not made that switch. For HTTPClient cases, we lack a way to pass an auth_token in on the __init__, which led to nova's horrible hack of setting it after the fact (before it switched to using sessions). HTTPClient should also stop requiring auth_url be passed in when retries is None or 0. It's also concerning that HTTPClient's authenticate() method doesn't support keystone v3. Since https://review.openstack.org/#/c/95305/ purports to add keystone v3 support to cinderclient, but didn't address that method, I assume whoever wrote that expects SessionClient to be used instead of HTTPClient when using keystone v3. That seems like an invalid assumption, though, or at least something that should be checked and a nice error returned in the event that keystone v3 was tried.