Steps to reproduce:
1. Install glance, cinder, nova and keystone services on Stable/Juno branch.
2. Update the branch to master, do 'sudo python setup.py install’ on all the services and do a db sync on all the services.
3. Restart all the services.
4. The nova and cinder services restart properly. In fact, I am able to do all sanity testing operations. Keystone though intermittently stops working. And doing cinder-list or nova-list gives either http://paste.openstack.org/show/yR4bFyzry9Lrhfp6nPWg/ or gives the listing of the volumes and instances on the host.
5. What I concluded was that Upgrading from Juno to Kilo has not been documented anywhere yet. I think I might be missing some secret sauce for keystone to start working correctly and stop giving intermittent errors. So, is there an official documentation like the one that exists for Icehouse to Kilo upgrade (http://docs.openstack.org/openstack-ops/content/upgrade-icehouse-juno.html) ? If not, then am I missing something in /etc/keystone/keystone.conf ?
AFAIK, the only changes you'd *need* to make during an upgrade of Keystone are simply related to removing XML translation middleware, which was dropped this cycle:
https:/
I was not able to reproduce your failure, although I'm having to make many assumptions about your deployment process & configuration. I used a completely default keystone.conf from juno, the default pipelines from juno and master, and ran using eventlet (which is deprecated in master). I used `python setup.py install` on stable/juno and again on master, and db_sync'd before & after. Keystone handles basic authentication flows just fine post-upgrade.
Can you make specific requests to keystone to produce the "Can't load plugin: sqlalchemy.