Work to be done

The operator should be able to deploy all cinder services as Active-Active.

• Cinder-API already supports it, so there's nothing to do here
• Cinder-Backup already supports it, so only the cleanup improvement is relevant.
• Cinder-Schedule theoretically supports
  • When multiple services are running we have eventual consistency, so we should evaluate performance of running a single cinder-scheduler service like SAP does.
  • The cleanup improvement is relevant.
  • Converting the services to a strong consistency model would come from upstream Cinder itself and is not a priority.
• Cinder-Volume
  • Requires a DLM
    • Before having the DLM code implemented we should prevent Active-Active deployments of the service
    • Evaluate usage of the DB driver as DLM:
      • Using the same DB as Cinder for the DLM: Functionality, load, performance, robustness, etc.
      • If necessary deploying a different DB since we already have an operator to deploy DBs
    • Evaluate the memcached TooZ driver since the podification effort is already looking into deploying it via the new infra-operator
    • Evaluate the redis TooZ driver since there seems to be an effort to look into deploying it
    • Implement the DLM support in the operator when deploying in A/A as well as ensuring cinder.conf is properly configured
    • If possible be user friendly regarding notification of drivers that don't support Active-Active
  • Changing from A/P to A/A should be supported
  • The cleanup improvement is relevant.

Note: Cinder-volume Active-Active support is only referring to the podified part (running under OpenShift) not to Edge/DCN where cinder-volume is running on RHEL nodes, as that is tracked on a different story.

Definition of Done

Cinder API, Backup, and Scheduler services can have their replicas values increased and work as expected.

Cinder Volume service can be deployed as Active-Active with an appropriate DLM service that is managed and configured in cinder.conf by the cinder-operator itself.