[libvirt-users] error: unsupported configuration: block I/O throttling not supported with this QEMU binary

Osier Yang jyang at redhat.com
Mon Dec 3 09:30:28 UTC 2012


On 2012年12月03日 17:19, hzguanqiang wrote:
> On 2012-12-03 16:53, Osier wrote:
>  >On 2012年12月03日 11:39, hzguanqiang wrote:
>  >> On 2012-12-03 10:33, Osier wrote:
>  >> >On 2012年11月28日 16:34, hzguanqiang wrote:
>  >> >> Hi, guys.
>  >> >> I want to change block I/O throttle using 'virsh blkdeviotune'
> with vm
>  >> >> not running, it reported an error:
>  >> >> # virsh blkdeviotune instance-000000dc /dev/loop0 --total-bytes-sec
>  >> >> 20000000 --total-iops-sec 20 --config
>  >> >> error: Unable to change block I/O throttle
>  >> >> error: unsupported configuration: block I/O throttling not supported
>  >> >> with this QEMU binary
>  >> >> I can do this when the vm is running. Is it a bug? How can I fix
> this
>  >> >> problem.
>  >> >
>  >> >It just looks strange, as you can tune when the VM is active. Because
>  >> >the qemu capabilities is collected during the qemu driver starts up,
>  >> >which is independant with the domain states.
>  >> >
>  >> >Probably you will want to debug the qemu capabilities collecting
>  >> >process.
>  >> >
>  >> >Osier
>  >> Osier, Firstly, thanks for your reply.
>  >> The situation is that I want to resize the disk and meanwhile reset the
>  >> block I/O throttle. And during the whole process, The vm is stopped.
>  >> I've found this was directly caused by the patch "qemu: reject
>  >> blockiotune if qemu too old" post by Eric Blake long before. I've sent
>  >> an email to ask him the reason. Maybe, People think you can't set I/O
>  >> throtting if vm not started, I guess.
>  >
>  >Eric's path is good, because it makes no sense to set block IO tune
>  >for inactive guest if the qemu doesn't support it, otherwise the guest
>  >will fails to start.
>  >
>  >The key point is why the IO tuning can be set when the guest is active,
>  >while fails when it's inactive. Because the qemu caps is not related
>  >with the domain states.
> Yes, You are right. And I agree with you on Eric's practice.
> It failed just because the qemu caps is Null when the vm is inactive.
> But what do you mean by saying "the qemu caps is not related with the
> domain states"?
> Do you mean It's just that way and I/O tuning can not be set when vm is
> inactive,

No, it's allowed as long as the qemu supports it.

  Or something is wrong with my libvirt configure that caused
> this problem?

I don't think it relates with configuration.

Regards,
Osier




More information about the libvirt-users mailing list