hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Min Zhou (JIRA)" <j...@apache.org>
Subject [jira] Issue Comment Edited: (HIVE-322) cannot create temporary udf dynamically, with a ClassNotFoundException
Date Sat, 07 Mar 2009 15:28:56 GMT

    [ https://issues.apache.org/jira/browse/HIVE-322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12679206#action_12679206
] 

coderplay edited comment on HIVE-322 at 3/7/09 7:28 AM:
-------------------------------------------------------

The create function statement loads a function class staticly from $HIVE_AUX_JARS_PATH and
hive.aux.jars.path
 http://issues.apache.org/jira/browse/HADOOP-4070

We ofen start a Hive thrift server, and it will up for quite a time. We cannot dynamically
load another UDF in another jar unless we restart the server .
It's not practical for us.
This patch introduces a register statement for appending a jar path to hive on-the-fly. It
works well both in CLI and Thrift server mode.
usage:
hive>register '/home/zhoumin/zhoumin.jar';
hive>CREATE TEMPORARY FUNCTION strlen AS 'hadoop.hive.udf.UdfStringLength';
hive>select strlen(time) from log_data;

Nevertheless, this patch cannot pass current unit test of hive, because the MapTask run during
unit test will serialize  MapRedWork w/o having done a register command. It will cause the
ClassLoader cannot find my UDF.  see line 61 and line 63 in MapTask.java.

      was (Author: coderplay):
    The create function statement loads a function class staticly from $HIVE_AUX_JARS_PATH
and hive.aux.jars.path
 http://issues.apache.org/jira/browse/HADOOP-4070

We ofen start a Hive thrift server, and it will up for quite a time. We cannot dynamically
load another UDF in another jar unless we restart the server .
It's not practical for us.
This patch introduces a register statement for appending a jar path to hive on-the-fly. It
works well both in CLI and Thrift server mode.
usage:
hive>register '/home/zhoumin/zhoumin.jar';
hive>CREATE TEMPORARY FUNCTION strlen AS 'taobao.hadoop.hive.udf.UdfStringLength';
hive>select strlen(time) from log_data;

Nevertheless, this patch cannot pass current unit test of hive, because the MapTask run during
unit test will serialize  MapRedWork w/o having done a register command. It will cause the
ClassLoader cannot find my UDF.  see line 61 and line 63 in MapTask.java.
  
> cannot create temporary udf dynamically, with a ClassNotFoundException 
> -----------------------------------------------------------------------
>
>                 Key: HIVE-322
>                 URL: https://issues.apache.org/jira/browse/HIVE-322
>             Project: Hadoop Hive
>          Issue Type: Bug
>          Components: Query Processor
>    Affects Versions: 0.3.0
>            Reporter: Min Zhou
>            Priority: Blocker
>         Attachments: registerjars-v3.patch, registerjars1.patch, registerjars2.patch
>
>
> I found the ClassLoader cannot load my UDF when doing FunctionTask, because the ClassLoader
hasnot append its classpaths on-the-fly yet.
> The ExecDriver' s addToClassPath(String[] newPaths) method is the only entry for ClassLoader
dynamically append its classhpaths (besides hadoop's GenericOptionsParser).
> But that function wasnot called before FunctionTask getting my UDF class by class name.
I think this is the reason why I came across that failure.
> scenario description:
> I set a peroperty in hive-site.xml to configure the classpath of my udf. 
> <property>
>   <name>hive.aux.jars.path</name>
>   <value>/home/hadoop/hdpsoft/hive-auxs/zhoumin.jar</value>
> </property>
> but failed to register it with a ClassNotFoundException when creating udf through the
sql command.
> CREATE TEMPORARY FUNCTION strlen AS 'hadoop.hive.udf.UdfStringLength'
> I'll make a patch soon.

-- 
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