[one-users] Gratuitous ARP replies: VM live-migration without loss of IP connectivity

[Stefan Kooman]

Quoting Javier Fontan (jfontan at opennebula.org):
> Thanks Stefan!
> 
> Do you think it could be do this in the network scripts?
Yes, actually that would be the best place. The arp reply can be send
right away, before the "mac_spoofing" scripts kick in. That way there's
also no need for an extra temporary rule to circumvent the
"mac_spoofing". The script now creates a  temporary "NORMAL" rule with
priority "50000" to allow the arp reply to be send.  It also means the
arp reply is send earlier in the process, which is even better. Another
benefit of including this in the network driver scripts is that they can
be optimized per driver (openvswitch / legacy bridging, 802.1Q, etc.).
> are executed in the node to prepare the node for a new machine or a
> migrated one.

>Does it make sense for newly created VMs? They take more to get the IP
>but it may also be good to tell the switch beforehand.
Yes, it does make sense to do it for newly created VM's. Actually, the
hook is triggered for every VM in state "RUNNING", thus also newly
created VM's. Normally a gratuitous arp reply is send by the VM itself
as soon as network interface is up and running. Sending a few extra
won't hurt at all :).
> 
> Even if this does not make sense it is a nice candidate for an addon.

I've taken a look at the Ruby script for OpenvSwitch (OpenvSwitch.rb).
I'll see if I can make it work for OpenvSwitch.

Gr. Stefan

-- 
| BIT BV  http://www.bit.nl/        Kamer van Koophandel 09090351
| GPG: 0xD14839C6                   +31 318 648 688 / info at bit.nl