[ 
https://issues.apache.org/jira/browse/HBASE-20361?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16429493#comment-16429493
 ] 

Yuki Tawara commented on HBASE-20361:
-------------------------------------

I attached a patch.
In this patch, I wrote a test to reproduce the bug, but I don't know the right 
way and the right place to write this test.
If my approach is not good, please give me an advice.

Thanks

> Non-succesisve TableInputSplits may wrongly be merged by auto balancing 
> feature
> -------------------------------------------------------------------------------
>
>                 Key: HBASE-20361
>                 URL: https://issues.apache.org/jira/browse/HBASE-20361
>             Project: HBase
>          Issue Type: Bug
>          Components: mapreduce
>            Reporter: Yuki Tawara
>            Priority: Major
>         Attachments: HBASE-20361.master.001.patch
>
>
> TableInputFormatBase class offers users a mechanism to exclude specific 
> splits from returned list of TableInputFormatBase#getSplits through 
> TableInputFormatBase#includeRegionInSplit.
> It also offers users a feature called "auto balancing" to mitigate data skew 
> by splitting large splits and merging small splits.
> If a user overrides TableInputFormatBase#includeRegionInSplit, i th split and 
> i+1 th split may not be successive(i th split's end key is smaller than i+1 
> th split's start key).
> If he or she further enable auto balancing feature, non-successive splits can 
> be merged, which means excluded splits between merged non-successive splits 
> "revive".
> To avoid such cases, we should not merge non-successive splits.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to