OVS/linuxbridge driver (bagpipe)


The BaGPipe driver for the BGPVPN service plugin is designed to work jointly with the openvswitch and linuxbridge ML2 mechanism drivers.

It relies on the use of the bagpipe-bgp BGP VPN implementation on compute nodes and the MPLS implementation in OpenVSwitch and or linuxbridge.

Architecture overview

The bagpipe driver for the BGPVPN service plugin interacts with the Neutron agent on each compute node, which is extended to support new RPCs to trigger the local configuration on compute nodes of BGP VPN instances and of their MPLS dataplane.

Example with the OpenVSwitch mechanism driver and agent:

blockdiag BGPVPN API Neutron DB bagpipe driver on each compute node ... : OpenVSwitch Agent + BGPVPN extension OVS br-int/br-tun OVS br-mpls bagpipe-bgp BGP Peers MPLS routers admin, tenant RPCs packets MPLS or .. MP-BGP BGPVPN service plugin bagpipe backend


On DHCP ports, Router interface ports, external network ports, etc.

No connectivity will be setup with BGP VPNs for DHCP ports or Router interface ports, or other network specific ports. This improves the load on network nodes by avoiding them to import/export a significant amount of routes, without limiting BGP VPN deployment scenarios because no useful traffic would be exchanged between a router or DHCP interface of a network associated to a BGP VPN.

Similarly, the driver will not bind a port on an external network. This behavior will be revisited once a use case is well identified.

bagpipe_v2 driver

For Queens release, the mechanism used by this driver for RPCs was changed.

The v1 driver networking_bgpvpn.neutron.services.service_drivers.bagpipe.bagpipe.BaGPipeBGPVPNDriver is backwards compatible with pre-Queens neutron agents and can be used during a rolling upgrade, e.g. from Pike to Queens.

The v2 driver networking_bgpvpn.neutron.services.service_drivers.bagpipe.bagpipe_v2.BaGPipeBGPVPNDriver does not produce the old RPCs anymore and can be used:

  • on a greenfield deployment

  • after an upgrade

  • during a non-rolling upgrade (some BGPVPN operations would be disrupted during the time where pre-Queens agent still run)

Future developments may happen only on the v2 driver and the v1 driver will be ultimately abandoned.

How to use ?

The steps to take to use this driver are generally:

  • install the networking-bagpipe package on both control nodes and compute nodes

  • on control node, configure neutron to use bagpipe driver

  • on compute nodes, configure the neutron agent to use bagpipe_bgpvpn extension and configure bagpipe-bgp

Of course, the typical way is to have all this taken care of by an automated Openstack installer.

In devstack

  • follow the instruction in README.rst

  • local.conf:

    • add the following to enable the BaGPipe driver for the BGPVPN service plugin:

    • enable networking-bagpipe, which contains code for agent extensions:

      enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git
      # enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git stable/pike
      # enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git stable/queens
  • on a control node, if you want to run the Fake Route-Reflector there (relevant only for a multinode setup):

    enable_service b-fakerr
  • on compute nodes:

    • the compute node Neutron agent is the Neutron openvswitch or linuxbridge agent, with the bagpipe_bgpvpn agent extension:

      • install networking-bagpipe (the code to interact with bagpipe-bgp comes from there):

        enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git
        # enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git stable/queens
        # enable_plugin networking-bagpipe https://opendev.org/openstack/networking-bagpipe.git stable/pike
      • the bagpipe_bgpvpn agent extension is automatically added to the agent configuration by the devstack plugin

    • bagpipe-bgp will be installed automatically (part of networking-bagpipe since Pike, or as a submodule before)

    • you need to enable and configure bagpipe-bgp, typically with a peering to a BGP Route-Reflector or BGP router(s):

      enable_service b-bgp
      # IP of your route-reflector or BGP router, or fakeRR
      # BAGPIPE_BGP_PEERS defaults to $SERVICE_HOST, which will point to the controller in a
      # multi-node devstack setup