[pve-devel] qemu kvm_steal_time bug

Alexandre DERUMIER aderumier at odiso.com
Mon Sep 21 09:21:57 CEST 2015


>>is this a guest kernel bug, or a qemu bug?

qemu bug, definitively.

Before kernel 3.16, the kvm_steal_time counter was freezing and since kernel 3.16, it's like it's at 100% usage.
in both case, the value is wrong.

That's occur with live migration, qemu save old value but restoring value is wrong is some case.
(Details are in bug report)


This counter is really just a monitoring counter, to see if host is overloaded and "steal" cpu time to guest.
Until it's fixed, it's 100% safe to disable it.




----- Mail original -----
De: "dietmar" <dietmar at proxmox.com>
À: "aderumier" <aderumier at odiso.com>, "Stefan Priebe" <s.priebe at profihost.ag>
Cc: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Lundi 21 Septembre 2015 06:33:34
Objet: Re: [pve-devel] qemu kvm_steal_time bug

is this a guest kernel bug, or a qemu bug? 

On 09/17/2015 04:41 PM, Alexandre DERUMIER wrote: 
> Some news info about this bug: 
> 
> https://bugs.launchpad.net/qemu/+bug/1494350?comments=all 
> 
> 
> I have hited the bug again. 
> (No production impact, but my monitoring probes become crazy because of cpu steal at 100%) 
> 
> I don't know if we could setup -kvm_steal_time by default for now ? 
> (it'll not break migration) 
> 
> or maybe add cpuflags options to vmid.conf ? 
> 
> 



More information about the pve-devel mailing list