Clustering : Implement a mechanism to balance datastore load in a cluster

Consider a three node cluster on which we have 8 shards with different data contained within it. If you start this 3 node cluster one node at a time you may end up with a situation where all the shards have their leaders on node 1. In the ideal scenario Shard leaders should be distributed across the cluster so that the load is evenly distributed.

We need to introduce some mechanism to make it possible to automatically switch leaders so that the load is more evenly balanced.