Bug 1294771

Description of problem:
I scaled down a cloud without migrating all instances first, and some nova instances became unreachable. Nova doesn't indicate what's wrong, and still displays the nodes as running.

Trying to reboot or stop the instances will result in those instances getting stuck in rebooting or powering-off state forever. The only possible thing to do is to delete the instances.

Nova should indicate that the nodes are defunct, and not continue and display the nodes as running.


Version-Release number of selected component (if applicable):
openstack-nova-common-12.0.0-2.el7ost.noarch
openstack-nova-compute-12.0.0-2.el7ost.noarch
openstack-nova-api-12.0.0-2.el7ost.noarch


How reproducible:
100%


Steps to Reproduce:
1. Deploy an overcloud with more than 1 compute node
2. Start a few instances and make sure that there are instances hosted on the last compute node(s)
3. Scale down the cloud so the last compute node(s) will be deleted
4. Try to reboot or power down the instances that lived on the deleted hosts


Actual results:
The operation is stuck


Expected results:
Nova should detect that the host doesn't exist any more, and not let instances get stuck like that in powering-off or rebooting state.