How did you handle re-partitioning and rebalancing every time you scaled your cluster in or out?