_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
aae5c7aa38