[Openstack] Multiple nova-compute hosts, single-host nova-network, and a guest unreachable via its floating IP

Vishvananda Ishaya vishvananda at gmail.com
Tue Jun 19 19:09:40 UTC 2012


On Jun 19, 2012, at 10:52 AM, Florian Haas wrote:

> Hi everyone,
> 
> perhaps someone can shed some light on a floating IP issue.
> 
> I have 2 nova-compute nodes (call them alice and bob), one of them
> (alice) is also running nova-network. bob uses alice as its
> --metadata_host and --network_host.
> 
> I assign a floating IP to a guest running on bob. Expectedly, that IP
> is bound to the NIC specified as the --public_interface on alice (my
> nova-network host).
> 
> However, since alice has a route into the --fixed_range network over
> its local bridge, the incoming traffic for the floating IP is routed
> there, where there's no guest to answer it -- because the guest is,
> after all, running on bob.

the fixed range should be bridged into an actual ethernet device, which means bobs guest should be able to respond just fine.

I would track down where the packet is getting lost.  If the floating ip is coming in on eth2 on alice, and then it is forwarded to br100 on eht1, the packets should be going out eth1 where bob will pick them up and forward them to its own br100.  Then the vm should be able to respond properly.  If this is all working, it maybe that the response packets from the guest are going back out the wrong interface.  They should be going back to alice's nova-network ip where they will be conntracked back to the floating ip.

Vish






More information about the Openstack mailing list