Hadoop 2

Network Performance Metrics

HADTWO-1625: QDS sends the following two new metrics to the Ganglia server to help detect network performance issues on worker nodes:

  • ping.packet.loss: The percentage packet loss during a ping command from a worker node to the coordinator node.
  • ping.time: The Return Travel Time (RTT) for a ping command sending 1000 packets to the coordinator node.

Enhancements

  • HADTWO-1745: QDS now supports running Hadoop 2 clusters on Java8. Via Support
  • HADTWO-1787: QDS Hadoop now allocates containers to nodes on the basis of real-time CPU utilization. If containers are packed on a node whose CPU becomes too busy, the containers are moved to to a different node.

Bug Fixes

  • HADTWO-1780: Ports the open-source fix HDFS-3384 to QDS to resolve the DFSClient bug that reported java.io.EOFException: Premature EOF: no length prefix available.