Comment 10 for bug 1590490

I think it's ok to work on this in 9.2 time frame, as the impact is limited here: this is only reproduced with Neutron when instances are evacuated *right after* a compute host failure, so that neutron-openvswitch-agent does not recognize that a port should have gone into DOWN state - in this case nova-api won't deliver network-vif-plugged event to the destination compute host. Hard rebooting of instance should help. FWIW, the original compute node already failed and these VMs are down anyway, so hard-rebooting is a workaround for now.