keepalived can't perform failover if the l3 agent is down

[Alex Katz]

If the l3 agent is down on the new master node of keepalived all the interfaces remain down so no packets can pass through the new router.

The replication scenario is the following:

1. Shut down all the l3 agents
2. Kill the keepalived process on the master node
3. Try to access VM behind the HA router (it will fail)

[LIU Yulong (dragon889)]

L3-agent should be alive in "new-master" node to link up the gateway after this patch:
https://review.opendev.org/#/c/707406/

[Slawek Kaplonski (slaweq)]

@LIU: yes, and I'm not sure how to deal with that since for me that is kind of regression in Neutron. But from the other hand, it's like that "by design" now and we had good reasons to do it that way.
So should we close this LP as "Won't fix" or do we want to maybe try to find some other solution for that?

[Brian Haley (brian-haley)]

This is similar to a bug filed a few years ago, https://bugs.launchpad.net/neutron/+bug/1677279 where we noticed an issue with IPv6 RAs when the l3-agent was down. It *could* be moved into keepalived-state-change if someone wanted to take on that task, just not sure of the priority with shift towards OVN.

[LIU Yulong (dragon889)]

OK, let's mark this as "Won't Fix". And move to state change can be a way to new install envrionment. But for running could, the existing keepalived-state-change processes may need to re-create.