[ https://issues.apache.org/jira/browse/HDFS-14702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16900585#comment-16900585 ]
He Xiaoqiao commented on HDFS-14702: ------------------------------------ [~crh], Thanks for your feedback, just notice that HDFS-8859 try to improve memory footprint but not backport. `ReplicaMap` information as following, ||Type||Name||Value|| |ref|entrySet|null| |int|hashSeed|0| |int|modCount|1261173816| |float|loadFactor|0.75| |int|threshold|6291456| |int|size|4093978| |ref|table|java.util.HashMap$Entry[8388608] @ 0x78cb249f0| |ref|values|java.util.HashMap$Values @ 0x778155350| |ref|keySet|null| !datanode.dump.png! > Datanode.ReplicaMap memory leak > ------------------------------- > > Key: HDFS-14702 > URL: https://issues.apache.org/jira/browse/HDFS-14702 > Project: Hadoop HDFS > Issue Type: Bug > Components: datanode > Affects Versions: 2.7.1 > Reporter: He Xiaoqiao > Priority: Major > Attachments: datanode.dump.png > > > DataNode memory is occupied by ReplicaMaps and cause GC high frequency then > write performance degrade. > It is about 600K block replicas located at DataNode, but when dump heap, > there are over 8M items of ReplicaMaps and footprint over 500MB. It seems > that memory leak. One more situation, the block w/r ops is very high. > Do not test HDFS-8859 and no idea if it can solve this issue. -- This message was sent by Atlassian JIRA (v7.6.14#76016) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org