dhcp scheduler should stop redundant agent

[Xurong Yang]

we initiate the counter of dhcp agents between active host and cfg.CONF.dhcp_agents_per_network, suppose that we start dhcp agents correctly, then some dhcp agents are down(host down or kill the dhcp-agent), during this period, we will reschedule and recover the normal dhcp agents. but when down dhcp agents restart, some dhcp agents are redundant.

            if len(dhcp_agents) >= agents_per_network:
                LOG.debug(_('Network %s is hosted already'),
                          network['id'])
                return

IMO, we need stop the redundant agents In above case.

[Xav Paice (xavpaice)]

I would be very happy to have redundant agents so that when a dhcp agent goes down we have some place to re-schedule.

[Xurong Yang (idopra)]

but if this issue happens frequently, there will be more and more dhcp agents. like deploying 10 dhcp agents and 5 down, then re-schedule other 5 agents, when down agents recover, 5 dhcp agents are redundant. what's worse, some resource will be in use, like ip.
,

[Eugene Nikanorov (enikanorov)]

I think the bug description should be updated, we don't need to stop agents, we need to unschedule networks from agents which were down and now are up.

[Armando Migliaccio (armando-migliaccio)]

I don't understand the problem statement.

[Launchpad Janitor (janitor)]

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