[rhelv6-list] Using specific physical interface to migrate vms

[Gianluca Cecchi]

On Thu, 14 Apr 2011 12:44:39 +0200 carlopmart wrote:
> How can I configure cluster.conf file to assign eth1 interface when vm live migration will be required??

This is how I understood it.
suppose your nodes' hostnames are host1 and host2
suppose your nodes' intracluster names are node1 and node2

when you migrate a vm from host1 to host2, this is the command issued
by default with qemu:
virsh migrate --live guest_vm qemu+ssh://node2/system tcp:node2

so the ssh connection is done through intracluster
(qemu+ssh://node2/system) and memory transfer is done through
intracluster (tcp:node2)

You can also verify this, comparing RX/TX bytes on interfaces before
and after the migration
(for example during a live migration of a RH EL 5 quite inactive guest
with 2Gb of ram, I see about 250-300 MBytes transferred)
This with a config inside cluster.conf such as

 <vm name="guest_vm" use_virsh="1" xmlfile="/etc/libvirt/qemu/guest-vm.xml"/>

Based on vm.sh in /usr/share/cluster I think you can only change both
of them together (so both ssh link and image transfer link)...

suppose your desired interface/bond device for migration is bound to
these names:
hostlive1 and hostlive2

I think you have to use the parameter migration_mapping with syntax like:

memberhost:targethost,memberhost:targethost

so that your cluster.conf line becomes:

 <vm name="guest_vm" use_virsh="1"
migration_mapping="node1:hostlive1,node2:hostlive2"
xmlfile="/etc/libvirt/qemu/guest-vm.xml"/>

Not tried yet but it should work.

You can see the meta-data for your resource agent with the command:

/usr/share/cluster/vm.sh meta-data

or walk through the script itself....

HIH,
Gianluca