[jira] Commented: (HIVE-708) for tranform,let the user specify the serde

["Zheng Shao (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-708?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12736835#action_12736835 ] 

Zheng Shao commented on HIVE-708:
---------------------------------

@hive.708.1.patch:

For the grammar change, can we accept SerDe at both input and output of the user script?

{code}
SELECT TRANSFORM(a, b, c)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '\001'
USING 'mypython.py'
AS (d STRING, e INT, f DOUBLE)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
{code}


> for tranform,let the user specify the serde
> -------------------------------------------
>
>                 Key: HIVE-708
>                 URL: https://issues.apache.org/jira/browse/HIVE-708
>             Project: Hadoop Hive
>          Issue Type: New Feature
>          Components: Serializers/Deserializers
>            Reporter: Namit Jain
>            Assignee: Namit Jain
>         Attachments: hive.708.1.patch
>
>
> Currently, LazySimpleSerDe is used to send data to the user transformation functions - it would be useful to let the user specify the format of the data.
> Specifically, it would be very easy and useful to accommodate: 
> (cut and paste from Venky's mail)
> Here's the typedbytes stuff that Dumbo uses.
> http://issues.apache.org/jira/browse/HADOOP-1722
> From:
> http://static.last.fm/johan/huguk-20090414/klaas-hadoop-1722.pdf
> Timings for IP count program on 300gigs of weblogs:
> Java: 8 minutes
> Dumbo with typed bytes: 10 minutes
> Hive: 13 minutes
> Dumbo without typed bytes: 16 minutes
> They also have a fast python decoder for this, which is apparently 25% faster than the python version. 
> http://github.com/klbostee/ctypedbytes/tree/master
> http://dumbotics.com/2009/05/31/dumbo-on-clouderas-distribution/

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.