Comment 9 for bug 1458887

I took a closer look at RabbitMQ logs: it seems that cluster failover finished fairly fast (at 11:31 node-1 goes away, at 11:39 cluster was fully operational with 2 nodes left). However for some reason neutron-server continues to think all agents are down, despite the fact they are up and have successfully reconnected to other RabbitMQ nodes. Looks like neutron-server ignores agent state reports for some reason.