On OSP18+BGP setups the CRC/OCP worker is not directly connected (at L2-level) to the external provider network that OSP uses for its FIPs and/or provider network VMs.

Without any extra configuration, the test pod cannot connect with the VMs on a BGP setup. The problem is that the spine/leaf infrastructure (the routes) don't have the reverse route to the tempest/tobiko pod IP.

This problem can be resolved configuring the test pods with hostNetwork set to True.

We think this could be the default configuration, not only for BGP, but also to be used with any setup.