7. Key takeaways
• NameNode is critical to cluster
• NameNode doesn’t equal to SecondaryNameNode, no back
up etc.
• Client access cluster nodes
• NameNode doesn’t take part in Data transfer
10. Key takeaways
• YARN – promotes hadoop
cluster to “universal
computational cluster”
• Map-Reduce is just one
application running on
cluster
• Hadoop is not just a Map-
Reduce since Hadoop 2.0
11. High Availability
• Issue for Hadoop 1.x
– NameNode SPOF
– Problems with cluster maintenance
– “Split the brain scenario”
– “Shoot me in the HEAD”
• Solutions:
– NFS
– Facebook’s “Avatar Node”
– Hadoop 2.0
• Things to consider
– Cold, Warm or Hot stand by
– Manual, Semi-automated, Automated failover
12. Hadoop 2.0 HA – Key points
• Hadoop HA doesn’t influence just HDFS
• Provides semi-automated or automatic failover
• Simplifies cluster maintenance
• Complicates node installations
• Cluster operations more complicated