Use CassandraKeyValueServiceRuntimeConfig.servers().
These are only the initial 'contact points' that will be used in connecting with the cluster. AtlasDB will
subsequently discover additional hosts in the cluster. (This is true for both Thrift and CQL endpoints.)
This value, or values derived from it (e.g. the number of Thrift hosts) must ONLY be used on KVS initialization
to generate the initial connection(s) to the cluster, or as part of startup checks.