For databases & datastores in general, each Juju unit should have a volume attached to it where data is stored. When a unit gets lost or removed (say for a scale down event), the data should not be lost as we would like to reuse it later when a new unit joins (say during a later scale up event). So, ideally: a new unit joins, finds a detached volume, picks it up instead of creating a new empty one.
This is useful as it allows the replication process to be faster, by only copying the missing data between the primary node and the newly joined one as opposed to doing full copies.
Therefore, it would be useful to be able to choose to reuse volumes, an option could be to treat Persistent Volumes as a queue, where the oldest detached volume (if any) gets assigned to the newest created juju unit, and only if no old volume is still around a new PV gets created.
Currently, this is not the case. Sometimes an old PV gets reused, sometimes a new one gets created despite having old detached ones around.