Today replicas are placed according to policy rules but there is no way to drive the number of replicas based on policy e.g.
Example 1: Put a replica of each shard of collection A on each node in the cluster
Example 2: Put a replica of each shard of collection A on every nodes with sysprop.region=east
The above two rules can be used to compute the replication factor of collection A but today the replicationFactor parameter passed in collection/shard creation neither takes the rule into account nor does it prevent conflicts. In fact, the replicationFactor should be omitted while creating collection A if the above rules are present.