Earlier ,till Hadoop 1.0 namenode was single point of failure.but 2.0 onwards HDFS federation, introduced , cluster to scale by adding namenodes, each of which manages a portion of the filesystem namespace/meta data.They are independent of each other.