Currently when multiple services share a host, the
restart order is random. This is due to an unordered
dict being used to facilitate the mapping of services
to their groups, names and other options.
Based on [1], this patch implements changes to the role
to ensure that services on the same host are restarted
in the correct order when the software/config changes.
[1] https://docs.openstack.org/developer/cinder/upgrade.html#minimal-downtime-upgrade-procedure
Conflicts:
>------handlers/main.yml
>------tasks/cinder_init_systemd.yml
Added appropriate changes to:
>------tasks/cinder_init_upstart.yml
>------templates/cinder-upstart-init.j2
Change-Id: If5729671cb69f928df660ec2d9ba83fe3f567946
(cherry picked from commit b98100fe0c)
b1a50d8a54