hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "liuzongquan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-11663) Auto load/unload custom udf function for hive cli and hiveserver2
Date Thu, 27 Aug 2015 04:31:45 GMT

    [ https://issues.apache.org/jira/browse/HIVE-11663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14716076#comment-14716076
] 

liuzongquan commented on HIVE-11663:
------------------------------------

when add a custom udf that want to be used in hiveserver2 and hive cli, you just upload the
UDF jar file on to ${hive.server2.customized.udf.jar.path}, which is a hdfs:// path. And then,
append the udf properties to hive-udf.properties as following.
 # mls udf
 multi_concat={com.meilishuo.hive.udf.jar,com.meilishuo.hive.udf.UDFMultiConcat}
 unfold={com.meilishuo.hive.udf.jar,com.meilishuo.hive.udf.UDFUnfoldURL}
 ip2long={com.meilishuo.hive.udf.jar,com.meilishuo.hive.udf.UDFIp2Long}

That's all you do for new custom UDF! And the hiveserver2 can use them immediately´╝ü

> Auto load/unload custom udf function for hive cli and hiveserver2
> -----------------------------------------------------------------
>
>                 Key: HIVE-11663
>                 URL: https://issues.apache.org/jira/browse/HIVE-11663
>             Project: Hive
>          Issue Type: Improvement
>          Components: CLI, Configuration
>    Affects Versions: 0.14.0, 1.0.0, 1.0.1, 1.1.1, 1.2.1
>            Reporter: liuzongquan
>            Assignee: liuzongquan
>              Labels: features, patch
>         Attachments: HIVE-11663.patch
>
>   Original Estimate: 96h
>  Remaining Estimate: 96h
>
> when adding custom functions used in hiveserver2, the most method is re-build the hive
source code, re-dist and restart hiveserver2. This way will produce big cost for service user
and cluster manager. The best way, in my opinion, the custom udf should be like a plugin to
the hiveserver2 and hive cli, and users should be add and remove at run-time, especially for
hiveserver2.



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

Mime
View raw message