[libvirt] kvm migration

[Chris Lalancette]

Łukasz Mierzwa wrote:
> Dnia wtorek 19 maj 2009 o 22:20:06 Daniel P. Berrange napisał(a):
>> On Tue, May 19, 2009 at 09:29:15PM +0200, ??ukasz Mierzwa wrote:
>>> Hi,
>>>
>>> can someone explain me how migration works in libvirt? I got 2 machines
>>> with ubuntu 9.04 (libvirt 0.6.1 and kvm 84), disk images are stored on
>>> nfs share so both machines can access them. When I run live migration
>>> (virsh migrate --live domain uri) my domain is migrated to second host
>>> but:
>>> 1. domain is defined on both hosts after migration, it does not disappear
>>> from original host
>>> 2. when I shutdown this domain on second host it gets undefined
>>> so I guess that migration is only temporary but this is not what I want.
>>> When I migrate domain to other host I want it to stay there, how can I do
>>> that? I can't find anything about it in documentation on libvirt website,
>>> google does not seem to know anything either.
>> You need to distinguish between a persistent and transient guest.
>> A persistent guest has a config file, a transient guest does not.
>>
>> If the guest on the source host is persistent, then after migration
>> you should still see it on the source host as inactive. If it is
>> transient, then all trace should have gone after migration.
>>
>> If the destination does not already have  a config file for the incoming
>> guest, then it will become a transient guest. Once you shut it down on
>> the destination, all trace will go away., If the destination has a config
>> for the guest it will become persistent, and the guest should still
>> exist.
>>
>> Based on your description I'd say your source host had a persistent guest,
>> and the destination host did not have a config, so after migration the
>> guest was transient.
>>
>>
>> Daniel
> 
> So if I want my guest to be undefined from the source host and stay on target 
> host I need to define him on target host before migration? Some info in 'virsh 
> help migrate' would be nice, and maybe '--persistent' option to auto-define it 
> on target before migration.
> Thanks for help.

Yeah, the persistent flag idea is a good one.  I think people would find that
useful, and it's fairly easy to implement.  I'll give it a shot.

-- 
Chris Lalancette