how to use create function to register a user-defined function in a maxcompute project - MaxCompute

Prerequisites

Before you begin, ensure that you have:

Added the required resources to your MaxCompute project using the ADD JAR command. For more information, see ADD JAR.

Limitations

Function names must be unique within a project. You cannot create a function with the same name as an existing function.
UDFs cannot overwrite built-in functions of MaxCompute. Only the project owner can use UDFs to overwrite built-in functions. If a UDF overwrites a built-in function, a warning appears in Summary of the Logview for the job after the SQL statement runs.

Syntax

CREATE FUNCTION <function_name> AS <'package_to_class'> USING <'resource_list'>;

Parameters

Parameter	Required	Description
`function_name`	Yes	The name of the UDF to create. Must be unique within the project.
`package_to_class`	Yes	The class of the UDF. Case-sensitive. Must be enclosed in single quotation marks (`'`). For a Java UDF, specify the fully qualified class name from the top-level package to the UDF class. For a Python UDF, use the `<Python_script_name>.<Class_name>` format.
`resource_list`	Yes	The resources used by the UDF. Must be enclosed in single quotation marks (`'`). Separate multiple resources with commas (`,`). To reference a resource from another project, use the `<project_name>/resources/<resource_name>` format.

Notes on `package_to_class` for Python UDFs

The Python script name refers to the underlying resource name that uniquely identifies the resource. For example, if you upload a resource as pyudf_test.py and then rename it to PYUDF_TEST.py in DataStudio or overwrite it using the MaxCompute client, the underlying resource name remains pyudf_test.py. When you register the UDF, the class name must therefore be pyudf_test.SampleUDF. Run LIST RESOURCES; to view the underlying names of all resources.

Notes on `resource_list`

The list must include the resources that contain the UDF code. Make sure those resources are already uploaded to MaxCompute.
If the UDF calls the Distributed Cache API to read resource files, include those resource files in the list as well.
If schema is enabled and you need to use resources from other projects, see Work with objects in a schema.

Examples

Create a Java UDF

The following example creates the my_lower function. The Java UDF class org.alidata.odps.udf.examples.Lower is in my_lower.jar.

CREATE FUNCTION my_lower AS 'org.alidata.odps.udf.examples.Lower' USING 'my_lower.jar';

Create a Python UDF from another project

The following example creates the my_lower function. The Python UDF class MyLower is defined in pyudf_test.py within the test_project project.

CREATE FUNCTION my_lower as 'pyudf_test.MyLower' using 'test_project/resources/pyudf_test.py';

Create a UDF with multiple resource dependencies

The following example creates the test_udtf function. The Java UDF class com.aliyun.odps.examples.udf.UDTFResource is in udtfexample1.jar. The function also depends on a FILE resource (file_resource.txt), a Table resource (table_resource1), and an Archive resource (test_archive.zip).

CREATE FUNCTION test_udtf AS 'com.aliyun.odps.examples.udf.UDTFResource' USING 'udtfexample1.jar, file_resource.txt, table_resource1, test_archive.zip';

What's next

FUNCTION: Create a temporary SQL function that applies only to the current SQL script and is not stored in the MaxCompute metadata system.
DROP FUNCTION: Delete an existing UDF from a MaxCompute project.
DELETE FUNCTION: Delete a function using the delete_function() method of a MaxCompute entry object.
DESC FUNCTION: View the details of a UDF, including its name, owner, creation time, class name, and resource list.
LIST FUNCTIONS: View all UDFs in a MaxCompute project.
UPDATE FUNCTION: Update a UDF using the update method of MaxCompute.