This bug was initially created as a copy of Bug #2291300

I am copying this bug because:

The tempest scenario for the feature belongs to a separate package. We would like to include it and, eventually, enable in CI.

Description of problem:
On RHOSP13 (OVS) it was possible to built an environment with two networks (n0 & n1) and two routers (r0 & r1). Only one Router has a GW to the internet (r0).

The r1 routes everything to r0 and from there it should route to the internet with the SNAT address of r0.
For VMs which are in the n0 which is directed attached to the router with GW IP, there is no problem. All VMs in r1 network, which have to hop via r1 and then to r0, will not be routed with the SNAT address to the internet. Theses systems will routed with their private address to GW. They never will get a response because its private address.

Now with RHOSP16.2 (OVN) this does not work anymore.

CU is now stuck in migrating this active customer workload. The compute part of the workload is already migrated and running on the RHOSP16.2 stack but networking is still running on the RHOSP13 environment.

CU is also testing upgrades to 17.1 that's why I've added it to the selected components.

Version-Release number of selected component (if applicable):
RHOSP 16.2
RHOSP 17.1

How reproducible:
Setting up the environment as described
Additional details as well as graphic about this setup can be found in the attached case

Additional info:
Similar to https://issues.redhat.com/browse/FDP-448