[lvs-users] IPVS, linux containers, virtual bridge

[Tomas Carnecky]

I have a single physical machine, the real servers are running inside linux
containers, connected to a virtual bridge (br0), and are assigned 10.0.0.x IP
addresses. I use IPVS to seamlessly deploy new versions of my software: create
new container, start the application inside it. add the IP address as a real
real server, remove the old real server and destroy the container.

This works well. Except when I try to connect from one container to another
through the VIP. Let's assume the following setup:

    VIP:        5.9.x.y
    br0:        10.0.0.1
    container1: 10.0.0.2
    container2: 10.0.0.3

    VIP port 9999 has assigned one real server, container1 (10.0.0.2), using
    masquerading.

When I try to connect from container2 (10.0.0.3) to 5.9.x.y:9999 (which should
forward the packets to container1) it does not work. The connection times out.
When I use tcpdump on the br0 device I see two packets:

   10.0.0.3 -> 5.9.x.y
   10.0.0.3 -> 10.0.0.2

The second one I assume is after IPVS has rewritten the destination. However,
no reply is ever sent back to 10.0.0.3. What am I missing?

tom