[jira] [Commented] (SPARK-14524) In SparkSQL, it can't be select column of String type because of UTF8String when setting more than 32G for executors.

["Sean Owen (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-14524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15234651#comment-15234651 ] 

Sean Owen commented on SPARK-14524:
-----------------------------------

Can you try this in 1.6 or master? I think this was resolved already. It does sound like a CompressedOops issue. I think you could verify by setting -XX:-UseCompressedOops on executors to be sure. Can you give a short reproduction? CC [~davies]

> In SparkSQL, it can't be select column of String type because of UTF8String when setting more than 32G for executors.
> ---------------------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-14524
>                 URL: https://issues.apache.org/jira/browse/SPARK-14524
>             Project: Spark
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 1.5.2
>         Environment: Centos
>            Reporter: Deng Changchun
>            Priority: Critical
>              Labels: UTF8String
>
> (related Issue:https://github.com/apache/spark/pull/8210/files)
> When we set 32G(or more) for executor, select the column of String type, it shows the Wrong result, such as:
> 'abcde'               (less than 8 chars)   => ''   (it will show nothing)
> 'abcdefghijklmn' (more than 8 chars) =>'ijklmn' ( it will cut the the front of 8 chars)
> However, when we set 31G( or less)  for executor, all is good.
> We also have debugged this problem, we found that SparkSQL uses UTF8String internally, it depends on some properties of locally JVM Memmory allocation ( see class 'org.apache.spark.unsafe.Platform'). 



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org