Currently the ReplicaSetDiscovery is run in a monitor thread during the entire lifecycle of connector. However since we are now using connection strings (where the host list is always used as a seed) this is no longer needed and only introduces ambiguous logs and database requests.

The proposal is to check the server topology once and only start the monitor thread when we are connecting to sharded cluster with replica_set mode (which is going to be removed in 2.6)