利用eclipse编写自定义hive udf函数

在作日志分析的过程当中，用到了hadoop框架中的hive，不过有些日志处理用hive中的函数处理显得力不从心，就须要用udf来进行扩展处理了

1  在eclipse中新建java project   hiveudf   而后新建class  package(com.afan)  name(UDFLower)

2  添加jar library  hadoop-0.20.2-core.jar   hive-exec-0.7.0-cdh3u0.jar两个文件到project

3  编写代码

1. package com.afan;  
2. import org.apache.hadoop.hive.ql.exec.UDF;  
3. import org.apache.hadoop.io.Text;  
4.   
5. public class UDFLower extends UDF{  
6.     public Text evaluate(final Text s){  
7.         if (null == s){  
8.             return null;  
9.         }  
10.         return new Text(s.toString().toLowerCase());  
11.     }  
12. }  

4  编译输出打包文件为 udf_hive.jar

5 将udf_hive.jar放入配置好的linux系统的文件夹中路径为/home/udf/udf_hive.jar

6 打开hive命令行测试

   hive> add jar /home/udf/udf_hive.jar;

Added udf_hive.jar to class path
Added resource: udf_hive.jar

建立udf函数
hive> create temporary function my_lower as 'com.afan.UDFLower';

建立测试数据
hive> create table dual (info string);

导入数据文件data.txt

data.txt文件内容为

WHO

AM

I

HELLO

hive> load data local inpath '/home/data/data.txt' into table dual;

hive> select info from dual;

Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_201105150525_0003, Tracking URL = http://localhost:50030/jobdetails.jsp?jobid=job_201105150525_0003
Kill Command = /usr/local/hadoop/bin/../bin/hadoop job  -Dmapred.job.tracker=localhost:9001 -kill job_201105150525_0003
2011-05-15 06:46:05,459 Stage-1 map = 0%,  reduce = 0%
2011-05-15 06:46:10,905 Stage-1 map = 100%,  reduce = 0%
2011-05-15 06:46:13,963 Stage-1 map = 100%,  reduce = 100%
Ended Job = job_201105150525_0003
OK
WHO
AM
I
HELLO

使用udf函数
hive> select my_lower(info) from dual;
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_201105150525_0002, Tracking URL = http://localhost:50030/jobdetails.jsp?jobid=job_201105150525_0002
Kill Command = /usr/local/hadoop/bin/../bin/hadoop job  -Dmapred.job.tracker=localhost:9001 -kill job_201105150525_0002
2011-05-15 06:43:26,100 Stage-1 map = 0%,  reduce = 0%
2011-05-15 06:43:34,364 Stage-1 map = 100%,  reduce = 0%
2011-05-15 06:43:37,484 Stage-1 map = 100%,  reduce = 100%
Ended Job = job_201105150525_0002
OK
who
am
i
hello

经测试成功经过

参考文章http://landyer.iteye.com/blog/1070377

——————————————————————————————————

一、编写函数

1. package com.example.hive.udf;  
2.   
3. import org.apache.hadoop.hive.ql.exec.UDF;  
4. import org.apache.hadoop.io.Text;  
5.   
6. public final class LowerCase extends UDF {  
7.   public Text evaluate(final Text s) {  
8.     if (s == null) { return null; }  
9.     return new Text(s.toString().toLowerCase());  
10.   } 
11. }  

二、用eclipse下的fatjar插件进行打包

先下载net.sf.fjep.fatjar_0.0.31.jar插件包，cp至eclipse/plugins目录下，重启eclipse，右击项目选Export，选择用fatjar导出(能够删掉没用的包，否则导出的jar包很大)

三、将导出的hiveudf.jar复制到hdfs上

hadoop fs -copyFromLocal hiveudf.jar hiveudf.jar

四、进入hive，添加jar，

add jar hdfs://localhost:9000/user/root/hiveudf.jar

五、建立一个临时函数

create temporary function my_lower as 'com.example.hive.udf.LowerCase';

六、调用

select LowerCase(name) from teacher;

注：这种方法只能添加临时的函数，每次从新进入hive的时候都要再执行4-6，要使得这个函数永久生效，要将其注册到hive的函数列表

添加函数文件$HIVE_HOME/src/ql/src/java/org/apache/hadoop/hive/ql/udf/UDFLowerCase.java

修改$HIVE_HOME/src/ql/src/java/org/apache/hadoop/hive/ql/exec/FunctionRegistry.java文件

import org.apache.hadoop.hive.ql.udf.UDFLowerCase;

registerUDF(“LowerCase”, UDFLowerCase.class,false);

(上面这个方法未测试成功)

为了不每次都有add jar 能够设置hive的'辅助jar路径'
在hive-env.sh中 export HIVE_AUX_JARS_PATH=/home/ckl/workspace/mudf/mudf_fat.jar;

from：http://blog.chinaunix.net/uid-28194925-id-3453844.html