When we started doing backy2 backups, the goal was recovering VMs that were hard to replace and our glance images in the event of a ceph cluster disaster (since we caused one by trying to involve the wrong VLANs during a network renumbering).
Over time, backy2 has proven that it likes to occasionally explode, and it has exposed that some of our openstack APIs occasionally just return random errors or loss of auth (which is another problem). However, we've now introduced cinder (T269511: Attachable block storage in cloud-vps), which is intended to be more-persistent storage than VM disks, without backing it up and have had to remove VMs from the system over time. Ceph is now quite stable, but recovering it would require a proper mirror at this point anyway, which we don't have plans to implement so far.
I'm suggesting we should stop supporting the backy2 backups since they are becoming increasingly unhelpful, and everything we support is extra work we could be putting elsewhere. It was just meant as a stopgap until we were sure we'd stabilized ceph, anyway.