Better default behavior if no glance server specified

[Lorin Hochstein]

Currently, if the admin does not specify a glance server in nova.conf (e.g., by using glance_api_servers or glance_host), then the compute host will default to using its own IP address ($my_ip). In a production deployment, this default behavior is almost never going to be correct. It's also difficult to figure out what's going on: I forgot to set this and I had a hard time tracking down why it was that the nova-compute service was trying to connect to a non-existent glance service on the same host.

If the admin doesn't specify a glance API server, the ideal solution would be to have nova-compute ask the service catalog (keystone) for the proper endpoint. If this isn't possible, then I would recommend that the default behavior should be for the nova-compute service to terminate with an error to the log such as:

Error: No glance server was specified in nova.conf. Please specify the glance endpoint using the "glance_api_servers" or "glance_host" config option.

[Sean Dague (sdague)]

I agree that this is actually a completely wrong default