VM boot issue after full clone between SAS storages (Dell ME4024) – Proxmox VE 8.4.14

[La1k3]

Hello,

I have a 6-node Proxmox VE cluster (version 8.4.14) connected to two Dell ME4024 storage arrays.
Each array has two controllers and is connected to all nodes.

Storage configuration in Proxmox:

• SAS1A – Array 1, controller A
• SAS1B – Array 1, controller B
• SAS2A – Array 2, controller A
• SAS2B – Array 2, controller B

Each node sees all four storages. Multipath is working correctly, but each path is configured as a separate storage in PVE.

Issue description​

I create templates from:

• Windows Server 2019 Datacenter (GUI) or
• Windows 10

If the template is created on SAS1A or SAS1B:

• full clone works only to SAS1A/SAS1B
• full clone to SAS2A/SAS2B fails to boot

Boot errors:

• with SeaBIOS: boot.ini error
• with OVMF (UEFI): missing BOOTMGFW.EFI

The situation is symmetrical:

• if the template is created on SAS2A/SAS2B, full clone works only there
• cloning to SAS1A/SAS1B results in the same boot problem

The cloning process itself completes without errors. The disk exists, the VM starts, but the firmware cannot find a valid bootloader.

Question​

Is there a way to:

• use one common Windows template
• and successfully full-clone it to all four storages (SAS1A/B and SAS2A/B)

without:

• maintaining two separate templates (one per array)?
  
  
  Thank you in advance.​

 

[AndreasS]

Hi, do the ME4024 have enough SAS Ports to connect to ALL 6 Hosts? Afaik there are only 4 Ports on each controller, not 6 (keep in mind 2 controllers per ME4024 are redundant, so you cannot count as 8 ports u can use).

 

[LnxBil]

Have you checked if the data is actually the same?

Can you create a small VM which yields this error. Shut it down on the source side, do the clone and don't start it. Now try to compare the actual data on disk. Easiest method would be just a simple checksumming, otherwise write a little problem that reads e.g. 1 MB from each disk and compare it.

IIRC, there was a bug in a SAN-backed setup a few years back in which the zeros were not transfered correctly (omitted, so optimized), so that zeros were not written on the destination storage, because it assued, the data would be already fully-zeroed on the target. This may not be the case. Check if there are any zero-related settings you can set.

 

[La1k3]

@AndreasS
ME4024 have 4 SAS port per controller - A0-A3 and B0-B3.
For this thread, we will use servers SRV1-SRV6.
SRV1-3 are connected to A0, A1, A2.
SRV4-6 are connected to B0, B1, B2.
The same applies to the second DELL array.

@LnxBil
So I created Win server 2019 Datacenter Desktop Exp on SAS1A (SeaBIOS - MBR). On this vm i have only installed virtio-win-guest tools - 0.1.285 (reboot after install it) and do Sysprep (OOBE + Generalize and shutdown). Then I converted this vm to template and do Full Clone on SAS1B and SAS2A and SAS2B

SHA256 checksums:

SAS1A (template)	10fe405c485da7f4e3b98378f6a2139203273d93574536aa38098ce773f7acef
SAS1B	55275d6e60a54284fb0e0db9671b1ea8d21b4ece085302764856cca63ca5a6dc
SAS2A	e1360c9b9d408a267049898c261d0b9df63b983fa09b45aa631c49593335c4f0
SAS2B	48c0e8e9bae8b231c83e0c4fa079e4c2d7b85d4644fe46d12714e14d7d86ce3c