HA router schedule for another active agent

[Ann Taraday]

Neutron Liberty, Ubuntu 14.04

3 controllers(3 l3-agents are available and active.), 1 compute

I have max_l3_agents_per_router=2, router is scheduled to 2 agents. If you of one these agents is stopped I will have something like:

root@node-1:~# neutron l3-agent-list-hosting-router r3
+--------------------------------------+-------------------+----------------+-------+----------+
| id | host | admin_state_up | alive | ha_state |
+--------------------------------------+-------------------+----------------+-------+----------+
| 1fa93569-fc36-4fb4-b467-85b11ba5f20c | node-3.domain.tld | True | xxx | active |
| 709998f7-cc64-4b6a-9777-6c815dfebb5c | node-2.domain.tld | True | :-) | active |
+--------------------------------------+-------------------+----------------+-------+----------+

The point is: router won't be rescheduled from dead agent and scheduled for another available active one. This situation can be avoided by simply setting max_l3_agents_per_router=0, but for current case is it possible to have a solution?

[Assaf Muller (amuller)]

With allow_automatic_l3agent_failover turned on, I made sure HA routers are purposely ignored. The rationale was that we expect the issue to be fixed and for the agent to come back. I don't have strong feelings about this. What do you think?

[Miguel Angel Ajo (mangelajo)]

It could be another configuration knob. Whether you want ha routers automatically rescheduled or not.

I'd tend to think that most of the deployments would set this to false, because you don't want to let temporary messaging issues mess with the scheduling of your routers, and if you want higher availability you can just spawn more agents and put a higher number of max_l3_agents_per_router.

[Ann Taraday (akamyshnikova)]

I'm also not sure whether this a bug or not. The main thing here is that with multiple dhcp agents when one failed it is replaced with new one, so here we don't have a consistency.

[Assaf Muller (amuller)]

-2 on a configuration knob. Pick an approach, have consistent deployments.

[Liping Mao (limao)]

Hi , I think HA router reschedule is useful for opertation purpose. for example, if admin want to maintain some of the network nodes, admin need a way to migrate HA router out of the network nodes.

[Armando Migliaccio (armando-migliaccio)]

This bug is > 180 days without activity. We are unsetting assignee and milestone and setting status to Incomplete in order to allow its expiry in 60 days.

If the bug is still valid, then update the bug status.

[Launchpad Janitor (janitor)]

[Expired for neutron because there has been no activity for 60 days.]