I worried about the cost of this approach, spin up new cluster for every streaming job running non stop required a lot of resources.