hadoop-hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zheng Shao (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HIVE-372) Nested UDFs cause _very_ high memory usage when processing query
Date Fri, 10 Apr 2009 10:49:15 GMT

     [ https://issues.apache.org/jira/browse/HIVE-372?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Zheng Shao updated HIVE-372:
----------------------------

    Attachment: HIVE-372.2.patch

This attached patch contains changed output files.

There is one regression - "CREATE TABL test ..." is reporting error in "CREATE" now.  I tried
to fix that but I didn't succeed. It probably needs much more effort.  Basically, we may see
this kind of problems for any consecutive keywords.

However this patch does detect a lot of the syntax errors much better than before (See other
new/changed test cases), so I still think we should still include this in 0.3.


> Nested UDFs cause _very_ high memory usage when processing query
> ----------------------------------------------------------------
>
>                 Key: HIVE-372
>                 URL: https://issues.apache.org/jira/browse/HIVE-372
>             Project: Hadoop Hive
>          Issue Type: Bug
>          Components: Query Processor
>         Environment: Fedora Linux, 10x Amazon EC2 (Large Instance w/ 8GB Ram)
>            Reporter: Steve Corona
>         Attachments: HIVE-372.1.patch, HIVE-372.2.patch
>
>
> When nesting UDFs, the Hive Query processor takes a large amount of time+memory to process
the query. For example, I ran something along the lines of:
> select trim( trim( trim(trim( trim( trim( trim( trim( trim(column))))))))) from test_table;
> This query needs 10GB+ of memory to process before it'll launch the job. The amount of
memory increases exponentially with each nested UDF.
> Obviously, I am using trim() in this case as a simple example that causes the same problem
to occur. In my actual use-case I had a bunch of nested regexp_replaces.

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


Mime
View raw message