In Hadoop 1.0, there is only 1 NameNode in whole cluster, this brings the risk of “single point failure”. In order to solve this problem, Hadoop 2 introduces HA mechanism.
In the picture, there are 4 DataNodes, and 4 NameNodes. 2 NameNodes are active, 2 NameNodes are standby.
Let’s see the left group of NameNode, one is active, another one is standby. In a cluster, there are at least 3 JournalNode. JournalNode share the active NameNode with the standby NameNode. In this way, the group of active NameNode and standby NameNode is a HA1.
In the right, there is another HA2(one active NameNode and one standby NameNode). HA2 manages the 4 DataNodes too. But HA1 and HA2 have different files, directories. For examlpe, if you use ls -l HA1, it may shows:
file1
file2
file3
When you use ls -l HA2, it may shows:
file4
file5
file6
In this case, we call HA1 is a federeation, HA2 is another federation. They operates its own HDFS separately. But they use the same DataNodes.