hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zhaofei Meng (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-10494) hive metastore server can't release its java heap with no work on it
Date Wed, 08 Mar 2017 06:39:37 GMT

    [ https://issues.apache.org/jira/browse/HIVE-10494?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15900798#comment-15900798
] 

Zhaofei Meng commented on HIVE-10494:
-------------------------------------

Try to adjust jvm parameter and set CMSInitiatingOccupancyFraction smaller.

> hive metastore server can't release its java heap with no work on it
> --------------------------------------------------------------------
>
>                 Key: HIVE-10494
>                 URL: https://issues.apache.org/jira/browse/HIVE-10494
>             Project: Hive
>          Issue Type: Bug
>          Components: Metastore
>    Affects Versions: 0.13.0
>         Environment: cloudera cdh 5.2.0
> 10 nodes
>  128G ram, 10T disk, 32core CPU for each node
> using impala for data analysis
>            Reporter: liqida
>
> I use impala for data analysis
> after a long time runing , impala DDL statements need a long time to complete "Planning
finished" and "DML Metastore update finished" steps.
> Both of them take 50 seconds and more.
> I found that HMS java heap has affected it so much .and I restart the hive metastore
server , the problem fixed .
> The HMS java ops like this :
> -XX:+UseParNewGC 
> -XX:+UseConcMarkSweepGC 
> -XX:-CMSConcurrentMTEnabled
> -XX:CMSInitiatingOccupancyFraction=70 
> -XX:+CMSParallelRemarkEnabled 
> -XX:+UseCMSCompactAtFullCollection 
> -XX:CMSFullGCsBeforeCompaction=0 
> -XX:SurvivorRatio=1
> and the total heap size is 3GB 
> after 3 days or less , I found the old genaration is full , and no matter what kind of
GC I tried , it never works .
> And then , after the whole work is done() , I run " jmap -F -histo PID "
> I found this :
> Object Histogram:
> num       #instances    #bytes  Class description
> --------------------------------------------------------------------------
> 1:              3955457 696160432       com.mysql.jdbc.JDBC4ResultSet
> 2:              3942714 630834240       com.mysql.jdbc.StatementImpl
> 3:              4051520 194472960       java.util.HashMap
> 4:              4714330 150858560       java.util.HashMap$Entry
> 5:              3990264 63844224        java.util.HashSet
> 6:              3978657 63658512        java.util.HashMap$KeySet
> 7:              3955458 63463696        com.mysql.jdbc.Field[]
> 8:              3964025 63424400        java.util.concurrent.atomic.AtomicBoolean
> 9:             3961293 63380688        java.lang.Object
> I think this is the causation
> So, what can I do with this , should I change some configuration or do something to fix
this , or HMS has any CACHE ? THANKS 
> BTW: Hive version 0.13.0 , I only use impala 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message