Spark SQL 指南

CREATE FUNCTION

描述

CREATE FUNCTION 语句用于在 Spark 中创建临时或永久函数。临时函数的作用域为会话级别，而永久函数则在持久化目录中创建，并且对所有会话可用。首次执行时，USING 子句中指定的资源对所有执行器都可用。除了 SQL 接口之外，Spark 还允许用户使用 Scala、Python 和 Java API 创建自定义用户定义的标量函数和聚合函数。有关更多信息，请参阅标量 UDF 和 UDAF。

语法

CREATE [ OR REPLACE ] [ TEMPORARY ] FUNCTION [ IF NOT EXISTS ]
    function_name AS class_name [ resource_locations ]

参数

示例

-- 1. Create a simple UDF `SimpleUdf` that increments the supplied integral value by 10.
--    import org.apache.hadoop.hive.ql.exec.UDF;
--    public class SimpleUdf extends UDF {
--      public int evaluate(int value) {
--        return value + 10;
--      }
--    }
-- 2. Compile and place it in a JAR file called `SimpleUdf.jar` in /tmp.

-- Create a table called `test` and insert two rows.
CREATE TABLE test(c1 INT);
INSERT INTO test VALUES (1), (2);

-- Create a permanent function called `simple_udf`. 
CREATE FUNCTION simple_udf AS 'SimpleUdf'
    USING JAR '/tmp/SimpleUdf.jar';

-- Verify that the function is in the registry.
SHOW USER FUNCTIONS;
+------------------+
|          function|
+------------------+
|default.simple_udf|
+------------------+

-- Invoke the function. Every selected value should be incremented by 10.
SELECT simple_udf(c1) AS function_return_value FROM test;
+---------------------+
|function_return_value|
+---------------------+
|                   11|
|                   12|
+---------------------+

-- Created a temporary function.
CREATE TEMPORARY FUNCTION simple_temp_udf AS 'SimpleUdf' 
    USING JAR '/tmp/SimpleUdf.jar';

-- Verify that the newly created temporary function is in the registry.
-- Please note that the temporary function does not have a qualified
-- database associated with it.
SHOW USER FUNCTIONS;
+------------------+
|          function|
+------------------+
|default.simple_udf|
|   simple_temp_udf|
+------------------+

-- 1. Modify `SimpleUdf`'s implementation to add supplied integral value by 20.
--    import org.apache.hadoop.hive.ql.exec.UDF;
  
--    public class SimpleUdfR extends UDF {
--      public int evaluate(int value) {
--        return value + 20;
--      }
--    }
-- 2. Compile and place it in a jar file called `SimpleUdfR.jar` in /tmp.

-- Replace the implementation of `simple_udf`
CREATE OR REPLACE FUNCTION simple_udf AS 'SimpleUdfR'
    USING JAR '/tmp/SimpleUdfR.jar';

-- Invoke the function. Every selected value should be incremented by 20.
SELECT simple_udf(c1) AS function_return_value FROM test;
+---------------------+
|function_return_value|
+---------------------+
|                   21|
|                   22|
+---------------------+