Rebooting not allowed for instances stuck in rebooting_hard state

i checked commit 26227b79e9246a87eeb83766cfcc8e96d294d28b, it remove many scheduler_api.

 git log -p 26227b79

- @exception.novaclient_converter
- @scheduler_api.redirect_handler
     def _action_reboot(self, req, id, body):
         if 'reboot' in body and 'type' in body['reboot']:
             valid_reboot_types = ['HARD', 'SOFT']

and i can't find any schedule api in openstack any more.

Saw this recently in a real deployment - a nova-compute service locked up for 12 hours (looks like that was a kernel bug) and during that time a user tried to hard reboot their instance

The reboot cast message was lost, so the instance stayed in task_state=REBOOTING_HARD. After the compute node came back, the user wasn't able to reboot the instance because:

    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.RESCUED],
                          task_state=[None, task_states.REBOOTING])
    def reboot(self, context, instance, reboot_type):

i.e. reboot is allowed while in task_state=REBOOTING_HARD

Looking at https://review.openstack.org/5090 and https://review.openstack.org/12368 I'm reading into Vish's comments that there can be problems if you do attempt hard reboot while a hard reboot is in progress

ISTM that if that's a concern, the compute manager should just take a lock on the instance while it's rebooting

I posted to the mailing list about this: http://lists.openstack.org/pipermail/openstack-dev/2013-January/004662.html

There are a lot more cases where hard reboot resolves a user's issue. I've proposed this patch to relax the restrictions in the API:

https://review.openstack.org/20009

Reviewed: https://review.openstack.org/20009
Committed: http://github.com/openstack/nova/commit/a69251804e62f871f93ffa420978f9b61a36df9c
Submitter: Jenkins
Branch: master

commit a69251804e62f871f93ffa420978f9b61a36df9c
Author: Rafi Khardalian <email address hidden>
Date: Fri Jan 18 07:01:50 2013 +0000

    Relax API restrictions around the use of reboot

    Fixes bug 1101082 bug 982108

    Reboots are the only sledge hammer users have available to resolve
    issues with their instances without administrative intervention.
    This patch modifies the API policy to allow reboot calls to be made
    in many more power and task states.

    Change-Id: Ia8702448f6b7b863da40e4d498f2e2ee0a12882e