Nova assisted volume snapshot race condition

[Lucian Petrut]

While performing assisted volume snapshots, we're setting the instance task state to 'image_snapshot_pending' in order to prevent instance actions that would impact this operation.

The issue is that we're clearing the task state only after calling back the Cinder API, telling it that the snapshot was created or deleted. For this reason, a subsequent assisted snapshot request
will be rejected by the Nova API if the task state is not cleared in time.

This race condition affects quite a few tempest tests that cover this scenario.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/529322
Committed: https://git.openstack.org/cgit/openstack/compute-hyperv/commit/?id=b6a23479d92a8fd770fc6ad6bf20cab4618e399e
Submitter: Zuul
Branch: master

commit b6a23479d92a8fd770fc6ad6bf20cab4618e399e
Author: Lucian Petrut <email address hidden>
Date: Wed Dec 20 15:09:15 2017 +0200

    Fix assisted volume snapshots race condition

    While performing assisted volume snapshots, we're setting the
    instance task state to 'image_snapshot_pending' in order to prevent
    instance actions that would impact this operation.

    The issue is that we're clearing the task state only after calling
    back the Cinder API, telling it that the snapshot was created or
    deleted. For this reason, a subsequent assisted snapshot request
    will be rejected by the Nova API if the task state is not cleared
    in time.

    This race condition affects quite a few tempest tests that cover
    this scenario.

    This change ensures that the instance task state is cleared before
    calling back the Cinder API.

    Closes-Bug: #1739423

    Change-Id: I1ae57c109ed551ba03d49b2ac7c6318b3939526d