[
https://issues.apache.org/jira/browse/HIVE-457?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Paul Yang updated HIVE-457:
---------------------------
Attachment: err.sh
HIVE-457.1.patch
* Modified TextRecordReader to return a line when the size has reached a
threshold, even if the newline hasn't been encoutered.
Test Plan:
err.sh is a bash script that continuously writes a string to stderr. Using
err.sh, the error can be reproduced by running in hive:
{code}
set mapred.child.java.opts=-Xmx8m;
add file err.sh;
SELECT TRANSFORM(*) USING 'err.sh' FROM src;
{code}
The job will fail quickly with a memory error (set
mapred.child.java.opts=-Xmx8m reduces available memory so that the job fails
quickly).
With the patch, the job should run continuously.
> ScriptOperator should NOT cache all data in stderr
> --------------------------------------------------
>
> Key: HIVE-457
> URL: https://issues.apache.org/jira/browse/HIVE-457
> Project: Hadoop Hive
> Issue Type: Bug
> Reporter: Zheng Shao
> Assignee: Paul Yang
> Priority: Blocker
> Fix For: 0.5.0
>
> Attachments: err.sh, HIVE-457.1.patch
>
>
> Sometimes user scripts output a lot of data to stderr without a new line, and
> this causes Hive to go out-of-memory.
> We should directly output the data from stderr without caching it.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
