Comment 4 for bug 1808228

It turns out Horizon dashboard didn't correctly pass HARD_REBOOT to Nova.
Here is the code where this issue comes out:

@profiler.trace
def server_reboot(request, instance_id, soft_reboot=False):
    hardness = nova_servers.REBOOT_HARD
    if soft_reboot:
        hardness = nova_servers.REBOOT_SOFT
        novaclient(request).servers.reboot(instance_id, hardness)

If it is not soft_reboot, novaclient won't be called.
BTW: there is no such an issue in upstream Horizon project: https://github.com/openstack/horizon/blob/master/openstack_dashboard/api/nova.py

Will file a patch to fix this issue.