Creating a user-defined function¶

Create a UDF¶

1. Write function logic as a handler using one of several supported languages, including Python, Java, and Scala.

2. Choose a tool or API to create the function with the handler you wrote.

   For more information about each of these, see Tools for creating UDFs.

   SQL	Use SQL and write logic in one of several languages.
   Snowpark	Use the Snowpark API for Java, Python, or Scala.
   Command line	Execute CLI commands to create the function.
   Python API	Execute client-side Python commands to create the function.
   REST	Make requests to a RESTful API to create the function.

3. Execute the function using one of several tools, depending on your needs.

Tools for creating UDFs¶

You can create a UDF using any of several methods available with Snowflake, depending on the language and skill set you have available. Choose the tool that’s right for your needs from the following table.

Key properties¶

The following describes some of the properties required or typically used when creating a function.

Function name:

The function name does not need to match the name of the handler. For more about name constraints and conventions, see Naming and overloading procedures and UDFs.

Arguments:

For more information on requirements, see Defining arguments for UDFs and stored procedures.

Return type:

For information about how Snowflake maps SQL data types to handler data types, see Data Type Mappings Between SQL and Handler Languages.

Handler name:

When required, this is the name of the class or method containing code that executes when the function is executed. You need specify a handler name only for handlers written in Java, Python, and Scala. For JavaScript and SQL handlers, all code specified in-line will be executed as the handler.

Dependencies:

For a handler written in Java, Python, or Scala, you might also need to specify the Snowpark library, such as when creating the function.

For more about making dependencies available to your handler, see Making dependencies available to your code.

Handler language runtime:

When the handler language is Java, Python, or Scala, specify the runtime version to indicate which supported runtime version to use. Keep in mind that if you use the default version, that default will change over time.

Access control requirements¶

A role used to execute this operation must have the following privileges at a minimum:

The USAGE privilege on the parent database and schema are required to perform operations on any object in a schema.

For instructions on creating a custom role with a specified set of privileges, see Creating custom roles.

For general information about roles and privilege grants for performing SQL actions on securable objects, see Overview of Access Control.

Usage notes¶

The following notes describe usage notes relevant to the languages supported for writing handlers. Although notes in the following sections refer to clauses of the SQL CREATE FUNCTION command, these clauses are typically represented in other ways in other tools you can use to create functions.

Create a UDF with SQL¶

You can create a UDF with SQL using the following steps.

You create a UDF with the following steps:

1. Write handler code that executes when the UDF is called.

   You can use one of the supported handler languages. For more information, see Supported languages and tools.

2. Choose whether you’ll keep the handler code in-line with the CREATE FUNCTION SQL statement or refer to it on a stage.

   Each has its advantages. For more information, see Keeping handler code in-line or on a stage.

3. Execute a CREATE FUNCTION statement in SQL, specifying properties of the function.

   Code in the following example creates a UDF called function_name with the in-line handler HandlerClass.handlerMethod.

   create function function_name(x integer, y integer)
     returns integer
     language java
     handler='HandlerClass.handlerMethod'
     target_path='@~/HandlerCode.jar'
     as
     $$
         class HandlerClass {
             public static int handlerMethod(int x, int y) {
               return x + y;
             }
         }
     $$;
   

   Copy

   The following describes some of the properties required or typically used when creating a function.

   • Function name.

     The UDF name does not need to match the name of the handler. The CREATE FUNCTION statement associates the UDF name with the handler.

     For more about name constraints and conventions, see Naming and overloading procedures and UDFs.

   • Function arguments, if any.

     See Defining arguments for UDFs and stored procedures.

   • Return type with the RETURNS clause.

     For a scalar return value, the RETURNS clause will specify a single return type; for a tabular return value, RETURNS will specify the TABLE keyword specifying column type in the tabular return value.

     For information about how Snowflake maps SQL data types to handler data types, see Naming and overloading procedures and UDFs.

   • Handler name with the HANDLER clause.

     When required, this is the name of the class or method containing code that executes when the UDF is called. You need specify a handler name only for handlers written in Java and Python. For JavaScript and SQL handlers, all code specified in-line will be executed as the handler.

     The following table describes the form of the HANDLER clause’s value based on the handler language and function type.

     Handler Language	UDF	UDTF
     Java	Class and method name. For example: MyClass.myMethod	Class name only. Handler method name is predetermined by the required interface.
     JavaScript	None.	None.
     Python	Class and method name if a class is used; otherwise, function name. For example: module.my_function or my_function	Class name only. Handler method name is predetermined by the required interface.
     SQL	None.	None.

   • Dependencies required by the handler, if any, using the IMPORTS or PACKAGES clauses.

     For more about making dependencies available to your handler, see Making dependencies available to your code.

   • Handler language runtime with RUNTIME_VERSION clause.

     When the handler language is Java or Python, use the RUNTIME_VERSION clause to specify which supported runtime version to use. Omitting the clause will prompt Snowflake to use the default, which may change in the future.