test_basic_scenario fails in the gate with a 500 error

[Matt Riedemann]

I've seen this fail a few times now and no one has reported a bug for it.

http://paste.openstack.org/show/44154/

Here is one tempest review that hit it:

http://logs.openstack.org/46/39346/12/gate/gate-tempest-devstack-vm-postgres-full/da6c383/console.html

And a nova one:

http://logs.openstack.org/24/41324/4/gate/gate-tempest-devstack-vm-postgres-full/3b72b25/console.html.gz

[Kieran Spear (kspear)]

I get a 500 by logging in on Horizon commit 11dff26cec32a7c77c2c19a31a46fbcb32324843 and then switching to commit 7e7965f875c01ef7e23c7910ea87765f861b571f. It's strange that this would be happening on the gate though, since I wouldn't expect sessions to persist between jobs...

Proposed a change to revert 11dff26cec32a7c77c2c19a31a46fbcb32324843 to see if it fixes things:

https://review.openstack.org/#/c/42033/

[Xavier Queralt (xqueralt-deactivatedaccount)]

I was able to hunt down the origin of this issue to the fact that we patch django auth middleware's get_user method too late.

It looks like the AuthenticationMiddleware called before loading openstack_auth's urls module (where the patching happens) and this leads to the first request being associated with the original get_user method and crashing if accessing to the user attribute.

This is where the middleware gets patched now:
https://github.com/openstack/django_openstack_auth/blob/master/openstack_auth/urls.py#L5

If instead it is patched from the settings module, the issue is fixed. I don't know if that's the correct place to patch it though.

[Eric Peterson (ericpeterson-l)]

This does seem to work better / correctly. Good idea Xavier!

https://review.openstack.org/#/c/42240/