Proxmox 9.1 VM Clone Issue (bug?/hw issue?)

[cmatthews]

The Issue:
When cloning typical Ubuntu guests, that have GPT / Logical Volume installations (The default installer setup), the cloned VM has a missing GPT partition table and is unusable. This issue is repeatable after reinstalling 9.1, I do not recall this being an issue in 8.4 on the same hardware. It seems like basic Windows 11 Guests have no issues being cloned.

There are no errors, the thin-lvm is ~50% full, the system is under very little load during this process. I have similar issues with backups as well, if I import backups into the system they have broken GPT partitions that are not recoverable. So this almost seems to involve the same mechanism?

I see this with the VirtIO and SCSI based storage configurations on VM guests. I'm currently running this in a production environment and very concerned about the restorability of backups etc. as a result.

See attached image for the fdisk info on both "disks" on the host.

Hardware:
Running Proxmox 9.1 on multiple Dell R740 Servers with PERC Raid controllers running SSD drives in RAID 1 array. So Proxmox see's just 1 unified local storage drive on each server. I do not recall this issue existing in Proxmox 8.4 with the same setup.

 

[cmatthews]

So after manually unpacking a backup file, that I confirmed had 3 raw disk parts in it, matching the expected partitions, I restored it on two different Dell R740 servers with nearly identical hardware.

fdisk above showed one restored as expected and the VM booted, on the "problematic" server, fdisk showed corrupted GPT table like above, and the VM obviously did not start.

Thus it had to be some sort of hardware issue, but since the OS level was having 0 problems I focused on the RAID / ThinPool... Comparing the layouts, I noticed the problematic server had a partition that may have been formatted with the LVM on top of that... The functional servers did not.

Upon reformatting of the drive, and re-creation of the GPT & LVM-Thin from the GUI, VMs now clone and import without issue...
So apparently someone setup this system manually and somehow introduced an issue in the layer between the LVM and the disk (Dell PERC Raid)


I hope this helps someone else.