Description of problem: Feature exception for Nova compute on ppc64le support As part of our Multi-Arch project, we are required to ship Nova compute support on Power8 little-endian per the IBM SoW. To facilitate this, we have been working with the RHOS RelDel DFG to ensure all OSP packages are building on our architecture. That work has gone well, however our container build service (OSBS) is not multi-arch aware and will not allow us to ship containerized services on this architecture. This feature request is to allow bare metal compute role on ppc64le. Simplifying assumptions: 1. The undercloud will stay x86_64 only. 2. Nova is the OpenStack project that provides compute management. This is actually several services nova-api, nova-conductor, placement and nova-cpu/nova-compute. *Only* the nova-cpu service is being considered for ppc64le at this time. 3. Power8 in ppc64le mode only is the only supported platform for RHOS 12. Proposed solution: - Update rhosp-director-images to be multi-arch aware. This work is something we can leverage should we expand the RHOS offering to further architectures. - Update tripleo+ironic to understand how to deploy to multiple architectures - Update tripleo-heat-templates to have 2 compute roles, Compute (default containerized install) and ComputeAlt (custom role for RPM/puppet based install). There are open questions here on naming. ‘ComputeBM’ describes what it is now but when we upgrade this to containers the role name seems strange as those node will no longer be ‘baremetal’. This does need to be a separate role but there is potential for it to be generated at install time (with openstack roles generate) rather at packaging time. - Update the docs to call out where things need to be done differently if installing for multi-arch - The initial nova feature set could be base CRUD (Create, Read, Update, Delete) operations. SPICE consoles are known not to work on power so we’ll need to enable both spice and VNC or if this fails we need to ensure that spice is disabled on the ppc64le nodes. Live migration is highly desirable but is untested upstream (on power) so we can’t commit it to that today. - x86_64 and ppc64le compute nodes will be in separate HostAggregates and images tagged with appropriate metadata ‘hw_architecture={x86_64,ppc64le}’. Puppet-openstack has support for this already, but needs to be wired up in tripleo.