Based on [1], this patch implements changes to the playbook
which executes the cinder deployment in a play per host
group, serialised to ensure that:
1. The services are changed in the right order.
2. The services remain available at all times during
an upgrade.
3. Online data migrations are actioned once all versions
of software are at the same levels.
4. If services are sharing a host/container then the
role execution will not execute twice on the same
host.
[1] https://docs.openstack.org/developer/cinder/upgrade.html#minimal-downtime-upgrade-procedure
Depends-On: Id95cae40f736ea2c84200955fccdb44ea3bc1dd8
Depends-On: If5729671cb69f928df660ec2d9ba83fe3f567946
Depends-On: I9aacda78f92355374af3f4ab24d2d9a9b47491ed
Change-Id: I0bdb51ce0d8b3b9a145d29ef6808e1fe595924e2
ef12bf04ed