Cinder-volume role does not work when deploy on node other than single-controller

Single-controller takes longer time to finish, so when cinder-volume tries to restart, it is most likely that the tables haven't been created in the db by cinder-manage db sync yet, but service restart still gives an ' exit 0 ' because the start failure is asynchronous, making chef-client believe the restart is successful.
Timing of installation can't be guaranteed for the moment, so we should ensure the service is truly started and abort the chef-client if failure is found and retry until tables are created.