bgpaas requires predictable selection of control node at vrouter to manage HA deployments

[richard roberts]

Following Orange VEPC, all vendors are highly relying on bgpaas for routing from/to VNF.

VNF are made up of several nodes spread on different computes:
 - Forwarders (line cards)
 - Control/Routing VMs (a la routing Engine)

This leads to a decoupling between data plane and control plane.

VNF will usually rely on two BGP peers per plane for HA (managed at differnt forwarder and/or Routing VMs).

PROBLEM:
The problem is that for a given plane, all BGP peers from a VNF can be actually terminated on a same control node. If this control node fails, both primary and backup peer get lost at a same time.

Contrail will break the HA design.

ASK: be able to have some control on how vrouter agent estblish peers toward Control Nodes - static definition in vrouter-agent.conf (one compute => all bgpaas to a given control node) is good too -

CAUTION: it is better that on of the peer do not establish at all than having two primary/backup peers established on a same Control Node

[OpenContrail Admin (ci-admin-f)]

Review in progress for https://review.opencontrail.org/40524
Submitter: sangarshan p (<email address hidden>)