Comment 16 for bug 1605089

@Kevin, that duplicated DB read may solve the problem in the example here, but in some cases it is not enough by simply overwriting the old value. For example, an update from A to B triggers removal of A from a network device (by some mech-driver) and then add B. In the race scenario A does not exist on the device yet, so it could trigger some unnecessary validation error. I agree that adding more logic may solve that problem again, but I think it is more clean to have the task-flow mechanism.