Presently, the reference architecture uses Ceph for storage (via Cinder CSI or Manila CSI). All data is mounted on a single volume. We would like to investigate how we could support attaching multiple volumes with different performance characteristics to a instance (/worker or master node) or use a combination of volumes and local ephemeral storage. This would allow us to e.g. move etcd storage to a "fast" volume and leave everything else on regular volume(s).

This is likely platform scoped. While the work we do is going to be OpenStack-only, other platforms (such as bare metal) could choose to copy what we are doing.

Related to "disk isolation for etcd" RFE: https://issues.redhat.com/browse/RFE-3133