shelve/unshelve notification may be out of order

[hougangliu]

when shelve a vm and CONF.shelved_offload_time == 0,the vm would be offloaded right now.
referring to nova/compute/manager.py:def shelve_instance(self, context, instance, image_id):
        if CONF.shelved_offload_time == 0:
            instance.task_state = task_states.SHELVING_OFFLOADING
        instance.power_state = current_power_state
        instance.save(expected_task_state=[
                task_states.SHELVING,
                task_states.SHELVING_IMAGE_UPLOADING])

        if CONF.shelved_offload_time == 0:
            self.shelve_offload_instance(context, instance)

        self._notify_about_instance_usage(context, instance, 'shelve.end')

thus,notification may occurs like:
compute.instance.shelve.start
compute.instance.shelve_offload.start
compute.instance.shelve_offload.end
compute.instance.shelve.end

in fact,order should be like:
compute.instance.shelve.start
compute.instance.shelve.end
compute.instance.shelve_offload.start
compute.instance.shelve_offload.end

I suggest that "self._notify_about_instance_usage(context, instance, 'shelve.end')" should be moved before
 " if CONF.shelved_offload_time == 0:
            self.shelve_offload_instance(context, instance)"

[Andrew Laski (alaski)]

I think it makes sense either way, but I can see the reasoning behind ordering them sequentially.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/56811

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/57933

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/56811
Committed: http://github.com/openstack/nova/commit/dd6765f751f9c6f36c027ba7c945c919ecf825e2
Submitter: Jenkins
Branch: master

commit dd6765f751f9c6f36c027ba7c945c919ecf825e2
Author: Jay Lau <email address hidden>
Date: Sat Nov 23 09:05:07 2013 +0800

    Adjust the order of notification for shelve instance

    When shelve instance, if CONF.shelved_offload_time == 0, then the
    notification order for shelve_instance() is as following:
    1) compute.instance.shelve.start
    2) compute.instance.shelve_offload.start
    3) compute.instance.shelve_offload.end
    4) compute.instance.shelve.end

    This order is not correct, as actually before shelve_offload.start,
    the instance was already reached shelve.end, so we should adjust the
    order of notification for shelve instance as following:
    1) compute.instance.shelve.start
    2) compute.instance.shelve.end
    3) compute.instance.shelve_offload.start
    4) compute.instance.shelve_offload.end

    Change-Id: I6e075b3f29e1c3f641ab1e459e1fbe3b05869a9a
    Closes-Bug: #1248802