That could actually be another approach, which I think we discussed previously; instead of relying on the cluster to provide a reliable storage provider, just require some random default storage class and when Juju creates a model (or lazily), deploy CephFS into the cluster and then use that for whatever storage Juju needs. We get the guarantee of full feature support, while keeping the pre-requirements to a minimum.
Actually, I'm not certain that LXD is a limitation for CephFS, since you can deploy it into Kubernetes: http:// docs.ceph. com/docs/ mimic/start/ kube-helm/
That could actually be another approach, which I think we discussed previously; instead of relying on the cluster to provide a reliable storage provider, just require some random default storage class and when Juju creates a model (or lazily), deploy CephFS into the cluster and then use that for whatever storage Juju needs. We get the guarantee of full feature support, while keeping the pre-requirements to a minimum.