You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ambari.apache.org by "Arpit Agarwal (JIRA)" <ji...@apache.org> on 2016/11/18 18:24:58 UTC

[jira] [Updated] (AMBARI-18936) DataNode JVM heap settings should include CMSInitiatingOccupancy

     [ https://issues.apache.org/jira/browse/AMBARI-18936?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Arpit Agarwal updated AMBARI-18936:
-----------------------------------
    Description: 
This is a followup to AMBARI-18694 to fix remaining stack versions:
___________________________

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
{noformat}
-XX:CMSInitiatingOccupancyFraction=70 -XX:+UseCMSInitiatingOccupancyOnly -XX:ConcGCThreads=8 -XX:+UseConcMarkSweepGC
{noformat}


  was:
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
{noformat}
-XX:CMSInitiatingOccupancyFraction=70 -XX:+UseCMSInitiatingOccupancyOnly -XX:ConcGCThreads=8 -XX:+UseConcMarkSweepGC
{noformat}



> DataNode JVM heap settings should include CMSInitiatingOccupancy
> ----------------------------------------------------------------
>
>                 Key: AMBARI-18936
>                 URL: https://issues.apache.org/jira/browse/AMBARI-18936
>             Project: Ambari
>          Issue Type: Improvement
>    Affects Versions: 2.2.2
>            Reporter: Arpit Agarwal
>            Assignee: Xiaobing Zhou
>             Fix For: 2.5.0
>
>
> This is a followup to AMBARI-18694 to fix remaining stack versions:
> ___________________________
> 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
> {noformat}
> -XX:CMSInitiatingOccupancyFraction=70 -XX:+UseCMSInitiatingOccupancyOnly -XX:ConcGCThreads=8 -XX:+UseConcMarkSweepGC
> {noformat}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)