[AMBARI-18694] DataNode JVM heap settings should include CMSInitiatingOccupancy

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.2.2
• Fix Version/s: 2.5.0
• Assignee: Xiaobing Zhou Assign to me
• Reporter: Xiaobing Zhou
• Created: 25/Oct/16 22:41
• Updated: 18/Nov/16 18:23
• Resolved: 28/Oct/16 21:39

Description

As HDFS-11047 reported, DirectoryScanner does scan by deep copying FinalizedReplica. In a deployment with 500,000+ blocks, we've seen the DN heap usage being accumulated to high peaks very quickly. Deep copies of FinalizedReplica will make DN heap usage even worse if directory scans are scheduled more frequently.

Another factor is that huge number of ScanInfo instances corresponding to HDFS blocks are lingering in garbage to eat many heap memories until a full GC takes place.

This proposes adding JVM settings to force GC more frequently to release DataNode heap consumed as a result of two aforementioned reasons, i.e. add the options to HADOOP_DATANODE_OPTS

-XX:CMSInitiatingOccupancyFraction=70 -XX:+UseCMSInitiatingOccupancyOnly -XX:ConcGCThreads=8 -XX:+UseConcMarkSweepGC

Issue Links

• is cloned by: AMBARI-18936 DataNode JVM heap settings should include CMSInitiatingOccupancy
• links to: ReviewBoard