[PVE-User] zvol vs qcow2 on zfs

[Roland]

Hello,

i'm currently planning a larger migration from xenserver to proxmox.

we want to use a proxmox cluster without shared storage, i.e. local
storage only. zfs is perfect for that.

whatever, we have found zvol (the proxmox default) is not optimal for
us, because of the following reasons:

- zvol cannot be replicated (on a "per dataset" basis) with pve-zsync.
we want to replicate our datasets to central backupserver.

- other replication tools (e.g. syncoid) won't handle zvols well, i.e.
when zvol is deleted on the source , it is not deleted on the target.
that is a problem when we "shuffle around" zvols between different
pools/datasets or servers. they would need extra scripting/handling on
the replication target.

- backing up zvols on the replicated server (for example with
borgbackup) is also not straightforward (because they are no files and
snapshots from a "backupsnap" aren't easily acessible, too)

- zvol has known performance issues , e.g. (
https://github.com/openzfs/zfs/issues/10095 )

is anybody using qcow2 on zfs in production at a larger scale or someone
wants to share his thoughts/experience with using qcow2 on zfs ?

regards
roland