Windows Server 2022 problem

[Jesus Blanco]

Since a few months we are having problems with Windows Server 2022 virtual machines. When applying Microsoft updates the machines hang with a BSOD.
The updates that we believe are causing the problems are these:
KB5022291
KB5022842

All our servers have 2 sockets and in the CPU configuration of the virtual machines we have the number of sockets and cores that the processors of the hosts have and we limit the machines with the vcpu and the cpulimit. By having NUMA and the CPU and RAM hotplug enabled, we can resize the machines hot.

We have found that after applying the updates to get the machines booting again there are only three ways to do this:

1. Change the configuration of the disks from scsi to sata with the consequent loss of performance.
2. Change the cpu configuration matching the number of cpu cores to the number of vcpu and cpulimint, with this we lose the ability to increase the cores on the client.
3. Change the cpu configuration leaving only one socket even if the host processor has two.

We are currently using this third option but it is a misconfiguration that we don't like and we are not clear about the performance problem it entails as the documentation does not recommend having NUMA enabled with only one socket.

Is anyone else having this problem?
Has anyone found a better and/or more direct solution?

Thank you very much in advance.

 

[Robert Smit]

I think you have NUMA enabled on your Processor settings, perheps with CPU and MEMORY enabled in you Hotplug Settings.
Try to disable NUMA on your Processor and make sure that CPU and Memory are not enabled on the hotplug settings (otherwise you VM will not start) This solved the problems for us!

 

[Jesus Blanco]

Thanks por your reply Robert,
Unfortunately we cannot do without the hotplug, for us it is an indispensable function.
For us the ideal would be to get some configuration or change in the Windows 2022 to keep the Proxmox configuration with the 2 sockets and NUMA.
Searching about this problem I found that for VMware they released a patch to solve the problem but I'm not a VMware user, since Proxmox exists it doesn't make sense to use it, so I couldn't verify if what this information said was really true.