Proxmox 9.2.2 and Rhel 9 (kernel 5.14.0) - potential CPU issues

[esh]

I have read a few places that earlier versions of proxmox running RH might have issues where the cores are under utilized. But then I also read that this is addressed in 9.2.2+
I have a scenario where this is exactly the case.
We appear to be spiking our memory utilization and then based on the vCPU percentages it looks like this is only using 1 of 16 CPU.

Has anyone else seen this in 9.2.2 specifically and if so.. any recommendations on a resolution?

Does the changing the type from default to "host" address this issue (at least this is one of the options that i have read which might address this).
We had not seen any similar issues in another unnamed hypervisor that we have left far behind now

 

[j.theisen]

Hi @esh

thank you for posting on the forum!

Does the VM recover by itself from this state or do you have to manually reset it?
Are there any journal messages in either the VM or the host indicating what is happening?
Are there any tasks running on the host during that time?

In any case could you try setting the CPU type to highest generalized one (i.e. x86_64_v#) assuming your VM doesn't need nested virtualization.
If you do need vmx please try the closest matching generalized CPU type to your hosts hardware (i.e. SapphireRapids-v#)

Best regards
Jonas

 

[esh]

So - the issue seems to occur when i have a report running on the guest system.
It does not recover on its' own. And over time this seems to grow in the time it takes to genereate said report.
After a restart, 2 hours .. then a few days later 9 hours .. etc.
It sounds like a potential memory leak on the client side - but trying to rule things out.
So the primary task and the items that is consuming the memory is tied to java generating the reports.