shelved_image_id is deleted before completing unshelving instance on compute node

[Pranali Deore]

Steps to reproduce:

1. Boot an instance from image.
2. Call Shelve instance, instance becomes SHELVED_OFFLOADED or SHELVED state depending on 'shelved_offload_time' configured in nova.conf.
3. Call unshelve instance.

   For shelved_offload_time >= 0:
   3.1 nova-conductor calls RPC.Cast to nova-compute
   If some failure happens in nova-compute. e.g. "Instance failed to spawn" error from libvirt
   3.2 nova-conductor deletes instance_system_metadata.shelved_image_id after RPC.cast for unshelving the instance.
   3.3 Instance becomes SHELVED_OFFLOADED again by revert_task_state, but instance_system_metadata.shelved_image_id is already deleted for this instance

For shelved_offload_time = -1:
   3.1 nova-conductor calls RPC.Cast to nova-compute
   If some failure happens in nova-compute. e.g. "InstanceNotFound" error while starting the instance.
   3.2 nova-conductor deletes snapshot and instance_system_metadata.shelved_image_id after RPC.cast to start the instance.
   3.3 Instance becomes SHELVED again by revert_task_state, but snapshot and instance_system_metadata.shelved_image_id is already deleted for this instance

Problems:
1. As there is no shelved_image_id, during unshelving the instance again, it gives error while getting image-meta in
   libvirt driver and instance remains in SHELVED_OFFLOADED state.

2. As there is no shelved_image_id, deleting the instance will try to delete "image_id=None" image from glance, but 404 error will be returned from glance, instance will be successfully deleted, and shelved image remains.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/160658

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/160658
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=b1fbbd375ccbecf8d70db633270fd7063728f05c
Submitter: Jenkins
Branch: master

commit b1fbbd375ccbecf8d70db633270fd7063728f05c
Author: PranaliDeore <email address hidden>
Date: Mon Mar 2 03:48:04 2015 -0800

    Delete shelved_* keys in n-cpu unshelve call

    During unshelve process, if some failure happens in compute manager,
    instance_system_metadata.shelved_image_id gets deleted in
    unshelve_insatnce call of conductor manager after RPC.cast to
    compute manager and instance remains in SHELVED_OFFLOADED or SHELVED
    state.
    It leads to below problems,
    1. Unshelving the instance again, it gives error while retrieving
    image-meta in libvirt driver.
    2. As there is no shelved_image_id, deleting the instance will try to
    delete "image_id=None" image from glance, but 404 error will be returned
    from glance.

    Removed shelved_* keys deletion code from unshelve_instance call of
    conductor manager and added it to unshelve_insatnce call of compute
    manager after spawning the instance.

    Also removed snapshot deletion code from unshelve_instance call of
    conductor manager and added it to start_instance call of compute
    manager.

    Closes-Bug: #1427056
    Change-Id: Iead979b5f6d1519cae58cc494f49e2fd8f323dd5

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/kilo
Review: https://review.openstack.org/183404

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/183404
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=aaaf7cebcad2c88876e0455c6569a09d04309a19
Submitter: Jenkins
Branch: stable/kilo

commit aaaf7cebcad2c88876e0455c6569a09d04309a19
Author: PranaliDeore <email address hidden>
Date: Mon Mar 2 03:48:04 2015 -0800

    Delete shelved_* keys in n-cpu unshelve call

    During unshelve process, if some failure happens in compute manager,
    instance_system_metadata.shelved_image_id gets deleted in
    unshelve_instance call of conductor manager after RPC.cast to
    compute manager and instance remains in SHELVED_OFFLOADED or SHELVED
    state.
    It leads to below problems,
    1. Unshelving the instance again, it gives error while retrieving
    image-meta in libvirt driver.
    2. As there is no shelved_image_id, deleting the instance will try to
    delete "image_id=None" image from glance, but 404 error will be returned
    from glance.

    Removed shelved_* keys deletion code from unshelve_instance call of
    conductor manager and added it to unshelve_instance call of compute
    manager after spawning the instance.

    Also removed snapshot deletion code from unshelve_instance call of
    conductor manager and added it to start_instance call of compute
    manager.

    Closes-Bug: #1427056
    Change-Id: Iead979b5f6d1519cae58cc494f49e2fd8f323dd5
    (cherry picked from commit b1fbbd375ccbecf8d70db633270fd7063728f05c)