PVE migration – Samba file server shadow copy: which technology should be used (vm with zfs vs. lxc with zfs vs. vm with ext4/lvm)?

PRJ

New Member
Oct 28, 2025
2
0
1
Hello,

we are currently in the process of migrating our vmware esxi to Proxmox.

One of the next steps is migrating the samba fileserver, since it is ~3TB large, I want to set up a new vm/lxc on the pve node and transfer the data with rsync so I can switch to the new system more or less “on the fly”.

With the new fileserver I also want to introduce samba shadow copy (previous versions for files in windows). Now comes the tricky part, I can’t really decide which base my new fileserver should be
  • VM with zfs
    • But then I have zfs on top of zfs, since my pve node has some zfs pools. (RAM usage shouldn’t be a problem because the node still has some reserves)
  • LXC with mounted host zfs pool (eg zamba-lxc-toolbox)
    • Since this fileserver will be accessible by a lot of people, I am not really happy with lxc because of the shared kernel from the pve node. Because of that I prefere the isolated vm.
  • VM with ext4/lvm (and use lvm snapshots)
    • Disk usage isn’t “the best” and the lvm snapshots can get inaccessible if the snapshots exceed a predefined snapshot size limit (that’s why I want to use zfs in the first place)
I want to backup the vm/lxc with all the data using pbs. Does the pbs backup includes the mounted zfs host pool?

Currently my favorite option is to use a vm with zfs, what do you think? Does anyone have experience operating such a file server?
Thanks in advance!

PVE Node info:
  • Dell PowerEdge R7615 with
    • 386 GB DDR5 RAM
    • AMD EPYC 9124 16-Core Processor
    • 8TB ZFS mirror (with 2x Samsung mzql2t6)
 
Last edited:
So you must use ZFS and setup a dedi Samba Fileserver (TrueNAS) on a raw Hardware.

OR

You can run a unprivileged LXC with zamba-lxc-toolbox, witch will work.
 
Currently my favorite option is to use a vm with zfs, what do you think? Does anyone have experience operating such a file server?
Yes, we're running this for many years without any problem. ZFS on ZFS can be a problem performancewise, but will technically work without a problem. With this setup you seperated both adequately and can manage everything directly in the VM, have proper PBS backups and not have to do stuff on your hypervisor, which is always good. Less is more on the hypervisor, unless you only operate with IaC and automation.

Keep in mind that you need regular snapshots and snapshot retention in you ZFS in order to keep the metadata overhead as slow as possible. ZFS will get very slow if the metdata does not fit in the ARC.
 
  • Like
Reactions: Johannes S and PRJ
I assume this Virtiofs-Device is then also included in the pbs backup?

Yes, we're running this for many years without any problem. ZFS on ZFS can be a problem performancewise, but will technically work without a problem. With this setup you seperated both adequately and can manage everything directly in the VM, have proper PBS backups and not have to do stuff on your hypervisor, which is always good. Less is more on the hypervisor, unless you only operate with IaC and automation.

Keep in mind that you need regular snapshots and snapshot retention in you ZFS in order to keep the metadata overhead as slow as possible. ZFS will get very slow if the metdata does not fit in the ARC.

Thanks for sharing, I think both ways will work now I have to just decide which way to use :D