Re: [Qemu-devel] Live migration bug, possible missing ram in migration?

[Paolo Bonzini <pbonzini@redhat.com>]

On 19/12/2014 14:15, Frerot, Jean-Sébastien wrote:
> Hi, I've been using kvm for some time now using live migration as well
> with ceph backend. Recently I started running into an issue with only
> one of my VM, which happens to be a windows server (2012). When I
> migrate this particular VM it seems that not all the ram is transferred.
> So when the migration completes, the vm that has been migrated simply
> hangs and I have to force a shutdown. Notice that not long ago it was
> working fine, however I didn't notice when it started to have the issue. 
> 
> Notice that my diagnostic of the problem may be wrong, I mean the
> behavior i'm having is that my VM hangs after live migration, and I
> don't see the problem on other VMs. But I noticed the difference in RAM
> usage by the qemu process which leads me to believe that the ram is not
> fully transferred.
> 
> before migration on node2:
> 15225 libvirt+  20   0 7128048 4.200g  13592 S  15.0 27.3   1:14.18
> qemu-system-x86
> libvirt+ 15225 52.7 27.3 7128048 4404244 ?     Sl   12:57   1:14
> qemu-system-x86_64
> 
> after migration on node1:
> 16507 libvirt+  20   0 6571864 1.610g  13152 R   7.6 10.5   0:07.63
> qemu-system-x86
> libvirt+ 16507 15.7 10.4 6571864 1688392 ?     Sl   13:00   0:08
> qemu-system-x86_64
> 

Windows VMs often have a lot of zero pages, because Windows zeroes pages
in the background.  These are migrated, but they are left unallocated on
the destination machine.  This is the reason why you are seeing smaller
memory usage on the destination.

Paolo