Shared storage over Fibre-Channel / iSCSI with thin provisioning and snapshots

[Yannick S]

Is there a supported solution for this at the moment?

We are currently running our 2 Hypervisors on OVirt, which is obviously dead and not supported anymore. Thats why we want to migrate to Proxmox.

For shared storage, we use SanSymphony DataCore (Software-SAN on 2x Bare-Metal Windows Server). They are directly connected via Fibre-Channel to each OVirt node. For testing purposes, I have connected a PVE Node aswell. Access can be achieved trough iSCSI or directly via FC. I dont know how, but OVirt enables thin-provisioned, snapshottable storage for the VMs. Proxmox cant do this with standard LVM. LVM-Thin is not shareable (due to a lack of integrated cluster manager on LVMs side, from what I understand).

Going trough the Storage-Type-Table, It seems there is no officially supported FS that covers these criteria. With the removal of GlusterFS in PVE 9, that is also obviously no option.

Edit: ZFS over iSCSI seemed like a good option but obviously works with ssh and some sort of Linux-Backend, not this proprietary software)

Im wondering if anyone else managed to solve this and if so, how? Switching away from DataCore is sadly not an option.

 

[bbgeek17]

Hi @Yannick S , the table you referenced is the official supportability matrix. Note that it does not reflect outside vendors who developed for PVE.

No, there is no officially supported Cluster Aware Filesystem. There are 3rd party guides on how to integrate OCFS2 out there.

PVE9 now provides support for snapshots on block storage as a technology preview.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[LnxBil]

PVE9 now provides support for snapshots on block storage as a technology preview.

Correct, but just to add for completness: still without thin-provisoning on the PVE end.

If your storage is able to thin-provision the volume itself, this may not be that big a deal, because each snapshot layer only has the changed data and everything else is not set. On the PVE end, you would still need the same size as the original disk and it would also be counted like this, yet on the storage side, you would not see this. This could be a situtation that would work for your, yet you may also need to overprovision the volume for PVE.