[virt-tools-list] KVM Live Migration

[Phil Meyer]

  On 10/05/2010 05:18 PM, Mike Hall wrote:
> We have the following test environment:
>
> 2 x CentOS 5.5 KVM virtual hosts
> 2 x CentOS 5.5 network storage boxes with NFS shares on LVs on DRBD (primary/secondary, no HA, just replication).
>
> We have installed a Win7 guest which can be live migrated almost instantly between hosts using Virt Manager (or virsh). The problem is that the guest does not retain its running state after migration, but instead arrives on the new host in "reboot after unclean shutdown" mode (Windows Error Recovery page on screen), asking how we would like to reboot the guest (Safe Mode, normal, etc).
>
> The migrated machine also loses IP connectivity (static IP) upon migration to new host, but regains it on migration back to original host.
>
> So, live migration in this case isn't producing the expected results (a running, usable guest).
> Same results using "migrate offline" and pausing the machine before migration.
>
> What could we be doing wrong?
>
> Michael Hall
> IT Communications Officer
> Alice Springs Town Council
> mhall at astc.nt.gov.au
> (08) 8950 0561
>
> _______________________________________________
> virt-tools-list mailing list
> virt-tools-list at redhat.com
> https://www.redhat.com/mailman/listinfo/virt-tools-list


We use the same basic software here.  The only difference being that we 
use iscsi targets for each VM, instead of one big iscsi pool, or NFS.

Our live migrates of all types of Linux hosts seem to work well.

One issue with Windows is that APIC events are not handled at all unless 
a user is logged in.

For instance, Win2k8 based VMSs do not respond to 'virsh shutdown' 
unless someone is logged in.  If it receives a second shutdown (power 
off) event within a minimum time frame, it will process it.

There is no way that the KVM host can tell if a client has processed an 
event.  It can only assume it did.

So for windows, test your live migrate with a user logged in, and see if 
it processes the pause event properly.

Although I have personally tested live migrates hundreds of times on 
RHEL5.5 and Centos5.5 servers with Fedora, Centos, and Ubuntu VMs, I did 
not test windows just because I didn't want to.  Maybe I should. :)

We did test our automated Win2k8 installer hundreds of times, but I was 
the only one who tested our migrate scripts.

I would rather just tell our customers, "Oh well.  I recommend you use 
Linux instead."  But I suppose management may not go for that. :)

I will run some Win2k8 (that is all we are offering customers for 
Windows) tests and respond here.

Good Luck!