[SOLVED] Who should run ZFS volume: Proxmox or virtual file server ?

[Wumpus]

Hi all,
I am currently planing to virtualize my private/personal OpenMediaVault server (OMV) as I plan to run more (virtual) servers in the future.
Currently, OMV is installed on a single SDD (with regular backups, I might add redundancy in the future) and runs a ZFS pool on "traditional" hard disks as "data tank" (RAID-Z1 plus regular backups). So far, I planned to give the virtualized OMV full access to the hard disks and have OMV to manage/drive the ZFS pool, just the same way as before. I did some promising tests with KVM/Qemu on regular Linux distros.
Currently I am evaluating Proxmox, and I discovered that Proxmox supports ZFS natively. So now the question is: should I let Proxmox run the ZFS pool and have OMV accessing it via emulated hard disk as usual, or should I have OMV running the ZFS-pool by configuring a passthrough of the hard disks as explained here in Proxmox?

The downside of having Proxmox running the ZFS pool is that I have to migrate the content into the disk image(s). Are there any advantages? Live migration is not an issue.

Current server configuration:
Pentium Gold + 32 GiB ECC on MB Fujitsu D3641-S12
50 GiB SSD (to be extended...)
3 WDC WD40EFRX-68N as ZFS Raid-Z1

Kind Regards,
Wumpus

 

[LnxBil]

You could also mount your zpool in PVE and use a LX(C) container for your OpenMediaVault, couldn't you? With that in mind, you only have von ZFS pool and can use it for both.

 

[Wumpus]

You could also mount your zpool in PVE and use a LX(C) container for your OpenMediaVault, couldn't you?

Is it really possible to run OMV as (LXC-)container ? Judging from this thread on the OMV forums it is not possible and it seems that it will never be, however several users do run it under PVE using KVM.

 

[Wumpus]

It seems that I was somewhat deceived by the post in the OMV forum concerning LXC containers, I found a couple of information about this (thanx @LnxBil for pointing me to that direction). Therefore, I am considering this thread as solved for now.