Comment 8 for bug 1821373

Reviewed: https://review.opendev.org/658845
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=aae5c7aa3819ad9161fd2effed3872d540099230
Submitter: Zuul
Branch: master

commit aae5c7aa3819ad9161fd2effed3872d540099230
Author: Matthew Booth <email address hidden>
Date: Mon May 13 16:04:39 2019 +0100

    Fix retry of instance_update_and_get_original

    _instance_update modifies its 'values' argument. Consequently if it is
    retried due to an update conflict, the second invocation has the wrong
    arguments.

    A specific issue this causes is that if we called it with
    expected_task_state a concurrent modification to task_state will cause
    us to fail and retry. However, expected_task_state will have been popped
    from values on the first invocation and will not be present for the
    second. Consequently the second invocation will fail to perform the
    task_state check and therefore succeed, resulting in a race.

    We rewrite the old race unit test which wasn't testing the correct
    thing for 2 reasons:

    1. Due to the bug fixed in this patch, although we were calling
       update_on_match() twice, the second call didn't check the task state.
    2. side_effect=iterable returns function items without executing them,
       but we weren't hitting this due to the bug fixed in this patch.

    Closes-Bug: #1821373
    Change-Id: I01c63e685113bf30e687ccb14a4d18e344b306f6