[jira] [Commented] (HIVE-6867) Bucketized Table feature fails in some cases

["Pengcheng Xiong (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-6867?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14569981#comment-14569981 ] 

Pengcheng Xiong commented on HIVE-6867:
---------------------------------------

reassign to [~hsubramaniyan]. Here is what we are expecting in a new patch after discussing: (1) Remove the enforce bucketing and the enforce sorting flags. (2) For insert into a bucketized table, get the table property if it is a bucketized one (sort by some key) and do the correct insert  (3) For load into a bucketized table, always throw an exception. (4) update all the related q files. (5) add a backward compatibility flag to allow to go back to the previous usage.


> Bucketized Table feature fails in some cases
> --------------------------------------------
>
>                 Key: HIVE-6867
>                 URL: https://issues.apache.org/jira/browse/HIVE-6867
>             Project: Hive
>          Issue Type: Bug
>          Components: HiveServer2
>    Affects Versions: 0.12.0
>            Reporter: Laljo John Pullokkaran
>            Assignee: Hari Sankar Sivarama Subramaniyan
>         Attachments: HIVE-6867.01.patch, HIVE-6867.02.patch, HIVE-6867.03.patch, HIVE-6867.04.patch, HIVE-6867.05.patch
>
>
> Bucketized Table feature fails in some cases. if src & destination is bucketed on same key, and if actual data in the src is not bucketed (because data got loaded using LOAD DATA LOCAL INPATH ) then the data won't be bucketed while writing to destination.
> Example
> ----------------------------------------------------------------------
> CREATE TABLE P1(key STRING, val STRING)
> CLUSTERED BY (key) SORTED BY (key) INTO 2 BUCKETS STORED AS TEXTFILE;
> LOAD DATA LOCAL INPATH '/Users/jp/apache-hive1/data/files/P1.txt' INTO TABLE P1;
> – perform an insert to make sure there are 2 files
> INSERT OVERWRITE TABLE P1 select key, val from P1;
> --------------------------------------------------
> This is not a regression. This has never worked.
> This got only discovered due to Hadoop2 changes.
> In Hadoop1, in local mode, number of reducers will always be 1, regardless of what is requested by app. Hadoop2 now honors the number of reducer setting in local mode (by spawning threads).
> Long term solution seems to be to prevent load data for bucketed table.



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