Comment 1 for bug 2018608

Revision history for this message
Luis Tomas Bolivar (ltomasbo) wrote :

Hi Maximilian! Thanks for the RFE. Yes, that looks really nice and aligned with what we wanted to do. Let's discuss this to ensure we are aligned and not doing the work twice :)

After some discussions with upstream folks we realized that maybe the most optimal way is not to have a BGP + EVPN drivers, but just one doing kind of both depending on configuration. For that, I've been adapting the BGP driver so add support for different ways of exposing the routes, see [1].

This is being done for the new NB BGP driver (using the NB instead of the SB DBs, to be more protected from core OVN changes), and the idea was to:
1) Expose the VMs as we are doing today, in the default VRF (available)
2) Add an option to expose using l2vni instead (as in [2]) (not started)
3) Add another option to expose each provider network in a different VRF/VNI (not started). This would be similar to the request on [3], but on the opossite direction. Instead of allowing IPs on the provider on the EVPN driver, exposing them with the BGP driver on different VRFs, using a similar wiring mechanism
4) Add an option to use OVN instead of kernel routing (very first WIP in [4])

What you propose (if I got it right) looks really aligned with 2) and 3), right? Are you targeting the current BGP driver?

[1] https://opendev.org/openstack/ovn-bgp-agent/src/branch/master/doc/source/contributor/bgp_supportability_matrix.rst
[2] https://bugs.launchpad.net/ovn-bgp-agent/+bug/2017890
[3] https://bugs.launchpad.net/ovn-bgp-agent/+bug/2017887
[4] https://review.opendev.org/c/openstack/ovn-bgp-agent/+/881779