LBaaS - No load balancing when client and VIP on same subnet

[Michael Micucci]

I have various topologies in my Zephyr (https://launchpad.net/zephyr-neutron) LBaaS test (tests.neutron.features.lbaas.test_lbaas_vip) which place the client VM (which sends TCP packets and expects an echo reply from the LB) onto the same subnet as the VIP. These all fail to receive any echo replies.

The root cause is apparently due to none of the traffic being sent to the router (as they are on the same subnet, it's purely L2), so the LB never has a chance to run.

One possible solution might be to set the Bridge device to force all traffic destined for a VIP to the router which has an interface on the same subnet as the associated pool for the VIP.

The Zephyr tests that fail are:

* test_lbaas_vip_on_pinger_subnet_pinger_on_pinger_subnet
* test_lbaas_vip_on_member_subnet_pinger_on_member_subnet
* test_lbaas_vip_on_pool_subnet_pinger_on_pool_subnet

[JF Joly (jfj)]

This is tracked in https://midonet.atlassian.net/browse/MNA-903