I think it's debatable whether the original patch should be reverted. There are certain SDN solutions out there that do not require the router for a floating IP to be visible from the actual VM. In OpenContrail, the whole DNAT and SNAT magic happens on the vrouter level and even works if there is no connection at all between the tenant network and the external network. The original patch that has caused this regression, in fact, breaks floating IP assignment with Juno and OpenContrail 2.0.
I think it's debatable whether the original patch should be reverted. There are certain SDN solutions out there that do not require the router for a floating IP to be visible from the actual VM. In OpenContrail, the whole DNAT and SNAT magic happens on the vrouter level and even works if there is no connection at all between the tenant network and the external network. The original patch that has caused this regression, in fact, breaks floating IP assignment with Juno and OpenContrail 2.0.