There is a race condition between the keystone and glance-api containers with regards to the user-create command in both keystone/start.sh and glance-api/start.sh. The race starts because both calls will reference the admin tenant and admin role, which crux will try to create if not present. Crux should not be run concurrently. Since it is being run concurrently, you can end up with two requests to create the admin tenant which will fail, causing the corresponding start.sh to exit and terminate the container.
The way I ran into this was I was using a 'docker run -d', which runs all the containers in the background. Often keystone would work, but the glance-api container would fail being unable to connect to keystone. Other times, both keystone and glance-api would fail because if glance won the race it was unable to find keystone.
The solution is to let the keystone script create the admin tenant and role. We need a properly working wait_for command that will verify a few things.
1) ensure the admin tenant exists
2) ensure the admin role was created by calling an action like 'endpoint-list', which you must be admin to do
The reason this wasn't seen with kubernetes is because kube restarts the containers over and over again until they work.
Fix proposed to branch: master
Review: https:/