Migration & Backup issues with NFS Dell unity storage

[sorsi]

Hello everyone,

We have a Dell Unity storage system that exports an NFS 4.2 volume to our Proxmox 8.4.1 cluster.
The VMs with their disk images on it run normally without any issues.

The problems are :
- When the Proxmox backup job is triggered, the disk images read by the hypervisors only contain zeros.
- When the disk image is moved out of the Dell Unity storage, the destination disk is also full of zeros and the VM ceases to function.

On the joined picture, you can see the backup size before we moved to the unity storage and after.

The strange thing is that accessing the disk image from any Hypervisor Shell works correctly.
- "du -h" shows the correct size
- "hexdump" at a random place shows the data
- "cp" of the disk image to another location (out of the Dell Unity storage) copies the expected data

Also, it seems that some VMs are not affected.

What could cause 'vzdump' and the disk move to fail like that?
Do these commands or features access the storage differently?

Thank you in advance for your help.

 

[MagicFab]

How is the NFS 4.2 volume mounted ? The NFS share mount could be tested with something like mount -o vers=4.2,nosharecache,noac . This has to do with QEMU block-level operation expecting sparse file handling and hole punching. Commands like vzdump and qimg-convert don't act like the file copy commands, that's why you see such difference.

You can check NFS 4.2 docs about the mount options, I lack the time to elaborate here.

Make sure your Unity storage software (UnityOS) is up to date too - NFS 4.2 is supported since v5.5 - you can see some useful info here.

---

Fabián Rodríguez | Le Goût du Libre Inc. | Montreal, Canada | Mastodon
Proxmox Silver Partner, server and desktop enterprise support in French, English and Spanish

 

[bbgeek17]

You have shown in the screenshot a hexdump of the disk that is explicitly skipped by the backup per the other screenshot you are showing.
So the zeroes read are from the different disk, is it possible that it is mostly sparse?

---

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