Please note: All times listed below are in Central Time Zone
Neutron's Router presents a L3 interface that connects VMs to an external network. In its legacy mode, a router can be scheduled only to one agent at a time, presenting a single point of failure - if that node fails, all communications are halted. L3 HA mitigates this issue by presenting an Active/Passive configuration where a router can be scheduled to multiple agents which recover the datapath automatically during failures using a dedicated network that hosts keepalived processes.
Using L3 HA presents an orchestration problem: since a router now appears in more than one agent and another dedicated network is required for the keepalived processes, there are a lot more resources involved. We performed L3 HA scale testing for Liberty and Mitaka, which showed benefits of L3 HA and uncovered a number of race conditions in Neutron code both on L3 agent and on server side. We'll present results of this scale testing and showcase some of these race conditions.
Attendees could expect to learn: