Comment 5 for bug 1327124

I also hit this issue, but it happened in the case of evacuating an instance. The phenomenon I saw is that the vm couldn't get ip address after evacuated. It's caused by that neutron refused to activate the port the new host since the port was still owned by the original host before evacuating. Without the process of activating, the bridge forwarding entries can't be populated on new host. Therefore the dhcp request packet from the instantce can't reach dhcp server. So in all the cases which will launch an instance on a different host, like unshelve, evacuate and so on, nova has to call neutron API to update the port binding with new host because neutron has no idea of the change.

I have verified the approach in https://review.openstack.org/#/c/116578/4 by calling the new added update api in the code path of evacuating. In my opinion, this is a very severe issue and it should be fixed as soon as possible.