There is only one NameNode in a HDFS cluster. This node maintains metadata about DataNodes. Since there is only one NameNode, it is the single point of failure in a HDFS cluster. When NameNode crashes, system may become unavailable.
We can specify a secondary NameNode in HDFS cluster. The secondary NameNode takes the periodic checkpoints of the file system in HDFS. But it is not the backup of NameNode. We can use it to recreate the NameNode and restart it in case of a crash.Read the full book at www.amazon.com