Comment 7 for bug 1583504

I tried to reproduce this bug and i think this is a valid bug. I followed below steps:

Multinode environment with controller and compute node,

!. boot an instance vm1 on compute node.
2. made the compute node to fault state.
3. disabled compute service on controller node.
4. evacaute vm1.
5. recover the fault host and restart the compute service.

As you said vm1 is deleted on compute node. I saw this message from n-cpu log
"[instance: 36663d00-fad8-468e-9a13-462b2bbfbd37] Deleting instance as it has been evacuated from this host"
"INFO nova.virt.libvirt.driver [-] [instance: 36663d00-fad8-468e-9a13-462b2bbfbd37] Instance destroyed successfully."

When i did "nova show vm1", it shows wrongly that it is still on compute node and task_state is 'active' and power_state is 'NOSTATE'.
So the vm is not running and it was not evacuated to controller node even after I enabled the compute service on controller node.

I read that nova evacuate is useful with shared storage. But in my environment i don't have shared storage. I see from your logs you have "onSharedStorage": true.