Hello, 

While testing failover for an active-backup bonding with ovs 1.4.0, I noticed 
that idle virtual machines (KVM) became unreachable for up to 1 minute.

Bond setup on KVM host:
# ovs-appctl bond/show bond0
bond_mode: active-backup
bond-hash-basis: 0
updelay: 200 ms
downdelay: 200 ms
lacp_negotiated: false

slave eth1: enabled
active slave
may_enable: true

slave eth0: enabled
may_enable: true



Doing a "ovs-appctl bond/set-active-slave bond0 eth1", the failover goes 
smoothly on the host and the standby interface becomes the active one (which is 
on a different switch).
For the kvm guests however, it seems like the uplink switches have to learn the 
new situation first and the guests are unreachable meanwhile.

Doing the same failover test, while a ping is running in a "screen" on the 
guests results in a smooth transition: no interruption in network communication 
to the guest.

Because the guest doesn't know about the underlying network change, it doesn't 
send a gratuitous ARP like it does with live migrations.  Is this a 
misconfiguration in openvswitch, since openvswitch is the only one who knows 
about the failover?  Is there a way to have it send gratuitous ARP requests for 
the virtual mac addresses? 

-- 
Frido Roose


_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to