Unexpected Task State Error on Snapshot

[Rick Harris]

The following error can be observed when running Nova in a multi-cell environment:

http://paste.openstack.org/show/21595/

The key problem is that the task_state is being updated as 'image_snapshot' in the API db but that task-state isn't getting propagated down to the cell DB.

The reason it's not being propagated down is because instance_test_and_set fetches a *new* instance object and sets the task_state on *that*; however we pass a different instance, with the original task-state to the compute.rpcai.snapshot call.

The solution is to mirror what we do with instance_update and have instance_test_and_set return an instance_ref which we use from then on, e.g.

instance = foo
instance = instance_test_and_set(instance.uuid, 'something')
print instance # this is the new instance that has the task-state set appropriately

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/14313

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/14313
Committed: http://github.com/openstack/nova/commit/3e6829a2012dd557a9e7c64281d4e6c73d1e3632
Submitter: Jenkins
Branch: master

commit 3e6829a2012dd557a9e7c64281d4e6c73d1e3632
Author: Rick Harris <email address hidden>
Date: Wed Oct 10 21:36:24 2012 +0000

    Pass correct task_state on snapshot

    This fixes an issue where we were not passing the correct task_state to
    the compute manager thereby triggering an 'unexpected task-state error'
    when the snapshot finished.

    This error only manifested itself in multi-cell setups where the
    compute-manager and compute-api are using separate DB's and therefore
    must rely on the RPC-layer to enforce consistency.

    Fixes bug 1065250

    Change-Id: Ie07454f0d2075f5c3f2dd53b4e5ff9983ea3241f