[library] l3-agent active-passive failover takes 30+ seconds when a controller fails

[Chris Clason]

l3-agent takes 30+ seconds to failover to a standby controller when a controller node fails. Customers are asking for this time to be minimized as much as possible.

[Andrew Woodward (xarses)]

There are hold down sleeps inside both the ocf script (+33 Sec) https://github.com/stackforge/fuel-library/blob/master/deployment/puppet/neutron/files/ocf/neutron-agent-l3#L411 and some in reconnect sleeps in https://github.com/stackforge/fuel-library/blob/master/deployment/puppet/neutron/files/q-agent-cleanup.py#L420. Since we have cleaned up service recovery we should be able to remove most of these and have a fast re-connect attempt to fix this for us if there are still gaps.

[Matthew Mosesohn (raytrac3r)]

Sergey, can you provide an update? Will this be fixed in 6.0 or moved to 6.1?

[Sergey Vasilenko (xenolog)]

This will be automatically fixed after "multiple-l3-agents" feature implemented.
But DOS timeout dapends also of rabbitmq stability. If failover leads to the rabbitmq cluster rebuild process, router can't be rescheduling to another L3 agent while rabbitmq cluster done rebuild.

[Matthew Mosesohn (raytrac3r)]

This big may be fixed by multiple l3 agents feature which is now merged.

[Vladimir Kuklin (vkuklin)]

+100500

[Vladimir Kuklin (vkuklin)]

this bug was fixed by multiple-l3-agents blueprint