[libvirt-users] About the change of a VM's vcpu count in runtime?

[Dennis Chen]

On 06/01/2013 06:49 PM, Dennis Chen wrote:
> On 06/01/2013 03:16 PM, Dennis Chen wrote:
>> On 05/29/2013 04:28 AM, Eric Blake wrote:
>>> On 05/28/2013 02:24 PM, Eric Blake wrote:
>>>
>>>>> So the question is: is there any method I can't find yet to change 
>>>>> the
>>>>> vcpu count of a running vm dynamically?
>>>> Wait for newer qemu and libvirt, or else pitch in and help write 
>>>> patches.
>>> You might not have to wait that long - if you could help test these
>>> patches, it would be appreciated:
>>>
>>> https://www.redhat.com/archives/libvir-list/2013-May/msg01857.html
>>>
>> Hi,
>>
>> After apply the patch, I've made a test. The testing environment as:
>>
>> virsh # version
>> Compiled against library: libvirt 1.0.5
>> Using library: libvirt 1.0.5
>> Using API: QEMU 1.0.5
>> Running hypervisor: QEMU 1.5.0
>>
>> The xml piece of the vm domain:
>>     <channel type='unix'>
>>         <source mode='bind' 
>> path='/var/lib/libvirt/qemu/f16x86_64.agent'/>
>>         <target type='virtio' name='org.qemu.guest_agent.0'/>
>>     </channel>
>>
>> the network connection between the vm and host is ok.
>>
>> Also I installed the qemu-guest-agent in the guest os:
>> [root at localhost ~]#rpm -qa | grep qemu-guest
>> qemu-guest-agent-0.12.1.2-2.355.el6.i686
>>
>> An error happened when I try to offline a guest vcpu in virsh:
>> virsh # setguestvcpu 8  1 --offline
>> error: Guest agent is not responding: Guest agent not available for now
>>
>> Is there some step I missed before I invoke the 'setguestvcpu' 
>> command in virsh or I need to write a program to hot-plug/unplug the 
>> vcpu with the help of the new API introduced by this patch? Thanks!
>>
>> BRs,
>> Dennis
>
> update after investigation, I need to start up the 'qemu-guest-agent' 
> daemon in the guest 
> (https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Virtualization_Host_Configuration_and_Guest_Installation_Guide/qemu-ga.html):
> [root at localhost ~]#qemu-ga
>
> After that, I tried the command again:
> virsh # setguestvcpu 8 1 --offline
> error: internal error unable to execute QEMU agent command 
> 'guest-set-vcpus': The command guest-set-vcpus has not been found
>
> I think the communication between monitor and guest is ok, because 
> below command is OK:
> virsh # shutdown 8 --mode agent
> Domain 8 is being shutdown
>
> VM shutdown after this command. Also, I double checked the 
> src/qemu/qemu_agent.c, and I can find the 2 functions:
> qemuAgentGetVCPUs() and qemuAgentSetVCPUs()
>
> Another,
> virsh # qemu-agent-command 9 '{"execute":"cpu-add","arguments":{"id":2}}'
> {"error":{"class":"CommandNotFound","desc":"The command cpu-add has 
> not been found","data":{"name":"cpu-add"}}}
>
> Seems the guest agent can't find vcpu hot-plug/unplug command, the 
> qemu I am using is 1.5.0, does it need some patches to handle this or 
> something I still missed?

hmm, I forget to upgrade the qemu-ga in guest os from 1.0 to 1.5.0. So I 
did that, the latest update --
The internal error doesn't show for "setguestvcpu 10 1 --offline" cmd, 
it's ok to offline the logic cpu id 1 when I checked it within the 
guest. But seems the output of "vcpuinfo/vcpucount" is not changed 
accordingly, e.g:
virsh # vcpuinfo 10
VCPU:           0
CPU:            1
State:          running
..

VCPU:           1
CPU:            3
State:          running
...
both are running (actually the vcpu 1 is offline). plus,
virsh # vcpumap 10 --offline --agent
CPU map:        -y

but
virsh # vcpumap 10 --offline
CPU map:        --

Last, in the qmp-commands.txt of qemu-1.5.0 and also mentioned in the 
patch [8/11]:

"This patch implements support for the "cpu-add" QMP command that plugs
CPUs into a live guest. The "cpu-add" command was introduced in QEMU
1.5"

but in my environment,
virsh # qemu-agent-command 10 '{"execute":"cpu-add","arguments":{"id":2}}'
{"error":{"class":"CommandNotFound","desc":"The command cpu-add has not 
been found"}}

is "cpu-add" supported by qemu 1.5.0 now? or something I still missed? 
IMO, the "cpu-add" is used to hot-plug a vcpu into the guest, so if the 
vcpu count specified in a vm xml is, e.g., 4, I can hotplug a new vcpu 
(totally 5 then) into the guest during runtime, is that correct?

BRs,
Dennis