Functions

A catalog function stores a reusable function definition and its metadata. The compute engine loads and executes the definition, so supported languages and function operations depend on the engine and catalog implementation.

Catalog and Engine Support

The REST Catalog implements persistent function operations. The built-in Hive, JDBC, and Filesystem catalogs do not implement persistent function creation, alteration, or deletion.

The engine adapter determines which stored definitions it can load:

Integration	Definitions loaded by the adapter
Flink catalog	File functions with implementation class and resource metadata.
Spark V1 function interface	Java file functions.
Spark V2 function interface	Lambda functions with a single return value.

Types of Functions Supported

Paimon's function metadata can represent three definition types:

Type	Definition
File function	References implementation resources, such as JAR files, with language and entry-point metadata.
Lambda function	Stores a lambda definition and its language.
SQL function	Stores a SQL definition.

The metadata model does not imply that every engine can execute every definition type. The following examples show Java file functions in Flink; see Functions in Spark for Spark usage.

File Function Usage in Flink

Select a Paimon REST Catalog and make the implementation JAR accessible to the Flink job, then register the function in an existing database.

Create Function

CREATE FUNCTION mydb.parse_str
    AS 'com.streaming.flink.udf.StrUdf'
    LANGUAGE JAVA
    USING JAR 'oss://my_bucket/my_location/udf.jar';

Add further JAR resources to the USING clause when the function requires additional dependencies.

Alter Function

Change the registered implementation class:

ALTER FUNCTION mydb.parse_str
    AS 'com.streaming.flink.udf.StrUdf2'
    LANGUAGE JAVA;

Drop Function

DROP FUNCTION mydb.parse_str;

Functions in Spark

See Spark SQL Functions for supported definitions and examples.