使用java开发一个helloworld级别UDF,打包成udf.jar,存放在/home/hadoop/lib下,代码如下:
package com.luogankun.udf;
import org.apache.hadoop.hive.ql.exec.UDF;
public class HelloUDF extends UDF {
public String evaluate(String str) {
try {
return "HelloWorld " + str;
} catch (Exception e) {
return null;
}
}
}
Hive中使用UDF
cd $HIVE_HOME/bin
add jar /home/hadoop/lib/udf.jar
CREATE TEMPORARY FUNCTION hello AS 'com.luogankun.udf.HelloUDF';
select hello(url) from page_views limit ;
SparkSQL中使用UDF
方式一:在启动spark-sql时通过--jars指定
cd $SPARK_HOME/bin
spark-sql --jars /home/hadoop/lib/udf.jar
CREATE TEMPORARY FUNCTION hello AS 'com.luogankun.udf.HelloUDF';
select hello(url) from page_views limit ;
方式二:先启动spark-sql后add jar
cd $SPARK_HOME/bin
spark-sql
add jar /home/hadoop/lib/udf.jar;
CREATE TEMPORARY FUNCTION hello AS 'com.luogankun.udf.HelloUDF';
select hello(url) from page_views limit ;
在测试过程中发现并不支持该种方式,会报java.lang.ClassNotFoundException: com.luogankun.udf.HelloUDF
如何解决?
1)需要先将udf.jar的路径配置到spark-env.sh的SPARK_CLASSPATH中,形如:
export SPARK_CLASSPATH=$SPARK_CLASSPATH:/home/hadoop/software/mysql-connector-java-5.1.-bin.jar:/home/hadoop/lib/udf.jar
2)再启动spark-sql,直接CREATE TEMPORARY FUNCTION即可;
cd $SPARK_HOME/bin
spark-sql
CREATE TEMPORARY FUNCTION hello AS 'com.luogankun.udf.HelloUDF';
select hello(url) from page_views limit ;
方式三:Thrift JDBC Server中使用UDF
在beeline命令行中执行:
add jar /home/hadoop/lib/udf.jar;
CREATE TEMPORARY FUNCTION hello AS 'com.luogankun.udf.HelloUDF';
select hello(url) from page_views limit ;