Proxmox PVE will crash out when Win Server 2025 and MS SQL 2025 is used.

eatonjb

New Member
Jan 20, 2025
1
0
1
I am really new to this, I have a client that I setup a proxmox server with a Windows 2025 server and all the updates, and installed a copy of Microsoft SQL 2025.

After some time, of use. (right now it's only 1 workstation using it) it will crash out the Proxmox server and the user has to reboot the server, and start all over again.

I am at a total lost,

I turned off ballooning memory
turned on NUMA
Running on X86-64-v2-AES
Vertools IO version 0.1.302,
10gig of memory,
SeaBIOS
I440fx 10.0+pve2 (I had it on q35 10.1)
vIOMMU to none.

I am really not sure what else is going on, i have no idea how its crashing the Proxmox VE.

I am really new to this, so if there is anything I can look at or change that is recommended, please let me know. I am clueless on how this crashes the Proxmox host itself.

edit: I am having the user just RDP into the Windows server and try doing his stuff there, it seems stable for some reason, but ill have to wait.
 
Last edited:
Hi @eatonjb, welcome to the forum.

The first thing I would check is the system log:journalctl. There may be some clues there.
If the customer has to manually power-cycle the server, that may indicate that there are messages on the console. If a monitor is not currently connected, it may be worth connecting one and taking a photo of the screen before rebooting the server.

Very often, the symptoms you describe can be hardware-related. Is the server you are using enterprise-grade hardware, or is it more of a home-lab type system?

Cheers


Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
 
Right after the next forced restart, I'd read the last messages from the boot that crashed with journalctl -b -1 -p warning (run journalctl --list-boots first to confirm the previous boot was saved). Also compare the host's total RAM with the 10 GB assigned to the VM plus what the host uses itself, and run arc_summary if the VM disk is on ZFS.

Inside Windows, set SQL Server's maximum server memory a few GB below 10 GB.

Drafted with AI, reviewed by me.
 
@Francois Baillargeon , the VirtIO version is mentioned in the opening post, and it appears to be the latest version.

That said, a VM should not normally be able to hard-crash the hypervisor host. Even in the case of memory exhaustion, I would expect the OOM killer to affect the relevant processes, such as the KVM process, rather than bringing down the entire host.

@eatonjb, what type of hardware is this running on? In particular, is this an enterprise-grade server, or is it something more along the lines of a repurposed desktop/laptop that the client had available? :)


Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
 
I would suggest reviewing the PVE Wiki page for Windows VirtIO drivers:
https://pve.proxmox.com/wiki/Windows_VirtIO_Drivers

I note that you mentioned 0.1.302 as the installed driver version - but currently 0.1.271 is currently recommended for production use.
Hello, where did you sax that 0.1.271 is currently recommended for production use ? they are quite old...
As said @bbgeek17 , latests stable are 0.1.302.
Latests stable can always be downloaded from : https://fedorapeople.org/groups/virt/virtio-win/direct-downloads/stable-virtio/virtio-win.iso
 
Hello, where did you sax that 0.1.271 is currently recommended for production use ? they are quite old...

Hi @ghusson - This is documented in the link provided:
Currently, there are no known issues for version virtio-win 0.1.271. We recommend to test new versions in a testing environment with realistic workloads.

As of 2026-09-10, we are currently testing virtio-win 0.1.302. We are not aware of issues with this version, but since this version is relatively new, we cannot recommend it yet for production environments.
 
  • Like
Reactions: ghusson
it will crash out the Proxmox server and the user has to reboot the server
I just want to reiterate, absent some new severe bug, nothing in the VM (including virtio drivers) should be able to cause the hypervisor host to crash to a point that a user has to manually reboot the server. Except, may be, if a misbehaving or misconfigured device pass-through is involved.

If I remember correctly, the prior virtio driver issues were completely VM contained.


Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
 
Not addressing the specific complaint (others are already) but two items to consider:
1. https://blogs.damiendye.uk/en/proxmox/q35-not-i440fx/ according to this its possible and even likely i440fx will be deprecated in the future.
2. note your vcpu type. Windows 2025 makes use of avx/avx2 which isnt exposed with cpu type X86-64-v2-AES. you really want V3 or better. if your hardware doesnt support it, considering dropping to server 2019/2022.
 
  • Like
Reactions: s.haigh
Windows Server 2025 requires high-end hardware specifications, so:
2 cores/4 threads = 8 vCPUs almost x86-64-v3
from Proxmox documentation:
x86-64-v3: Compatible with Intel CPU >= Haswell, AMD CPU >= EPYC. Added CPU flags compared to x86-64-v2-AES: +avx, +avx2, +bmi1, +bmi2, +f16c, +fma, +movbe, +xsave.

x86-64-v4: Compatible with Intel CPU >= Skylake, AMD CPU >= EPYC v4 Genoa. Added CPU flags compared to x86-64-v3: +avx512f, +avx512bw, +avx512cd, +avx512dq, +avx512vl.
24 GB RAM (32 GB is better; 16 GB minimum + 8 GB for SQL—set a usage limit)
q35 UEFI / OVMF + TPM2
Video card: at least VirtIO.
The Standard version with Desktop Experience running on a server like the HPE Gen9 is very resource-heavy, and performance seems quite sluggish; therefore, you need at least a Gen10 (Xeon E-22xx or Xeon Silver) and SAS drives.
By comparison, a VM running Windows Server 2022 with Desktop Experience is much more responsive; alternatively, there is the Server 2025 Core version, though the command line can be quite challenging.

What host do you have?
 
  • Like
Reactions: sgw