kvm: Desc next is 2

[fti-fred]

I have a Windows Server 2022 VM that is getting event 129: Source: vioscsi, Message: Reset to device, \Device\RaidPort4, was issued.

This is happening about every 8 days in the evening during high io load. In proxmox there is a corresponding error kvm: Desc next is 2
Proxmox v9.2.3, VM is q35 9.2, UEFI, virtio scsi single, ceph, nvme storage, cache=writeback, discard=on, iothread=1, ssd=1 , five disks with the same settings , Windows and SQL Server 2022 Standard.
When this occurs the VM must be gracefully restarted if early on after the error, or hard reset if it is left on for too long.

I've tried virtio scsi driver versions .262, .285, .271....and after a reboot tonight downgrading to .208 I understand some people have better experience with v204.

I've reviewed

Y

Thread 'Redhat VirtIO developers would like to coordinate with Proxmox devs re: "[vioscsi] Reset to device ... system unresponsive"'

Jan 3, 2024

vrozenfe_at_redhat_dot_com (Vadim Rozenfeld) with Redhat would like to connect with Proxmox developers regarding a particular point of pain that Proxmox users seem to be having with their IO drivers: https://github.com/virtio-win/kvm-guest-drivers-windows/issues/623#issuecomment-1875076293

• ylluminate
• virtio virtio disk virtio scsi
• Replies: 299
• Forum: Proxmox VE: Installation and configuration

Does anyone have advice?

I've also created a test VM and tried to reproduce the error using Microsoft Diskspd and CrystalDiskMark without success.

 

[j.theisen]

Hi @fti-fred

thanks for posting on the forum!
Do you by chance have multiple SCSI disks connected to your VM?
How did you change the drivers between tests?

Background why i'm asking:
Windows is very unwilling to change drivers to lower versions without forcing it to.
You can check the currently used version with the following command:

Get-CimInstance Win32_PnPSignedDriver | Where-Object {$_.DeviceName -match 'VirtIO|SCSI'} | Select-Object DeviceName,DriverVersion,DriverDate,InfName

So the issue you linked as far as i remember was the one that was fixed with v0.1.271 which was mainly related to the "Controller reset" messages.
v0.1.285 has introduced new issues especially under heavy load for Windows Server 2025 but there ware also reports for 2022, so i'd recommend against using that one.

Currently still 271 is the recommended version and we have not seen any reset issues with that except if the backing storage was really blocked for more than 60 seconds.

Best regards
Jonas