[jira] [Comment Edited] (HIVE-18206) Merge of RC/ORC file should follow other fileformate which use merge configuration parameter

["Wang Haihua (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-18206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16276257#comment-16276257 ] 

Wang Haihua edited comment on HIVE-18206 at 12/4/17 3:46 AM:
-------------------------------------------------------------

[~prasanth_j] For MergeFileTask, {{mapred.max.split.size}} is having effect, but the source of {{mapred.max.split.size}}, which comes from {{hive.merge.size.per.task}}, is not taking effect.

I saw {{hive.merge.size.per.task}} has been setting to MaxSplitSize in  {{MergeFileWork}}  from ConditionalResolverMergeFiles.java:
{code}

  private void setupMapRedWork(HiveConf conf, MapWork mWork, long targetSize, long totalSize) {
    mWork.setMaxSplitSize(targetSize);
    mWork.setMinSplitSize(targetSize);
    mWork.setMinSplitSizePerNode(targetSize);
    mWork.setMinSplitSizePerRack(targetSize);
  }

{code}

But value of {{MaxSplitSize}} not take effect when executing, it just comes from hadoopconf. 


was (Author: wanghaihua):
[~prasanth_j] For MergeFileTask, {{mapred.max.split.size}} is having effect, but the source of {{mapred.max.split.size}}, which comes from {{hive.merge.size.per.task}}, is not taking effect.

I saw {{hive.merge.size.per.task}} has been setting to MaxSplitSize in  {{MergeFileWork}}  from ConditionalResolverMergeFiles.java:
{{

  private void setupMapRedWork(HiveConf conf, MapWork mWork, long targetSize, long totalSize) {
    mWork.setMaxSplitSize(targetSize);
    mWork.setMinSplitSize(targetSize);
    mWork.setMinSplitSizePerNode(targetSize);
    mWork.setMinSplitSizePerRack(targetSize);
  }

}}

But value of {{MaxSplitSize}} not take effect when executing, it just comes from hadoopconf. 

> Merge of RC/ORC file should follow other fileformate which use merge configuration parameter
> --------------------------------------------------------------------------------------------
>
>                 Key: HIVE-18206
>                 URL: https://issues.apache.org/jira/browse/HIVE-18206
>             Project: Hive
>          Issue Type: New Feature
>    Affects Versions: 1.2.1, 2.1.1, 2.2.0, 3.0.0
>            Reporter: Wang Haihua
>            Assignee: Wang Haihua
>         Attachments: HIVE-18206.1.patch, HIVE-18206.2.patch
>
>
> Merge configuration parameter, like {{hive.merge.size.per.task}} , decide the average file after merge stage.
> But we found it only work for file format like {{Textfile/SequenceFile}}. With {{RC/ORC}} file format, it {{does not work}}.
> For {{RC/ORC}} file format we found the file size after merge stage, depends on parameter like {{mapreduce.input.fileinputformat.split.maxsize}.
> it is better to use {{hive.merge.size.per.task}} to decide the the average file size for RC/ORC fileformat, which results in unifying.
> Root Cause is for RC/ORC file format, merge class is {{MergeFileTask}} instead of {{MapRedTask}} for Textfile/SequenceFile. And {{MergeFileTask}}  just has not accept the configuration value in MergeFileWork, so the solution is passing it into  {{MergeFileTask}}



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)