Bug 1190284

Description of problem:

The RHEL OSP installer by default has nova compute talk to the Cinder API via the Public URL defined in Keystone.  In several deployments the public network is isolated and the compute nodes cannot talk to it.  The result is that cinder volumes would be created, but could not be attached to the instance.  /var/log/nova/nova-compute.log indicates the connection error.  

The OpenStack security guide (http://docs.openstack.org/security-guide/content/ch021_paste-and-middleware.html) recommends that nova communicate over internal URL rather than public.  This is achieved by modifying /etc/nova/nova.conf and setting the following parameter to internal: 

cinder_catalog_info='volume:cinder:internalURL'

The quickstack compute classes should be updated to use internal rather than public.  

Version-Release number of selected component (if applicable):
RHEL OSP 5 / RHEL OSP 6 

How reproducible:
Communication will fail in any deployment where compute nodes are isolated from the Public URL from Cinder.  

Steps to Reproduce:
1. Boot a nova instance 
2. Attach a cinder volume 


Actual results:
Cinder volume does not attach due to no communication to publicURL 

Expected results:
Nova communicates over internal URLs rather than public URLs.  

Additional info:

I believe for the Neutron host group this could be set in quickstack::compute_common just by adding a new configuration value like this: 

nova_config {
  'DEFAULT/cinder_catalog_info': value => 'volume:cinder:internalURL';
}

Alternatively this could be modified to allow a configurable parameter rather than hard coding the value.