CREATE PROCEDURE¶
Creates a new stored procedure.
A procedure can be written in one of the following languages:
Note
When you want to create and call a procedure that is anonymous (rather than stored), use CALL (with Anonymous Procedure). Creating an anonymous procedure does not require a role with CREATE PROCEDURE schema privileges.
- See also:
ALTER PROCEDURE, DROP PROCEDURE , SHOW PROCEDURES , DESCRIBE PROCEDURE, CALL
Syntax¶
Java and Scala
You can create a stored procedure that either includes its handler code in-line, or refers to its handler code in a JAR file. For more information, see Keeping Handler Code In-line or on a Stage.
For in-line stored procedures, use the following syntax:
CREATE [ OR REPLACE ] [ SECURE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS { <result_data_type> [ [ NOT ] NULL ] | TABLE ( [ <col_name> <col_data_type> [ , ... ] ] ) }
LANGUAGE { SCALA | JAVA }
RUNTIME_VERSION = '<scala_or_java_runtime_version>'
PACKAGES = ( 'com.snowflake:snowpark:<version>' [, '<package_name_and_version>' ...] )
[ IMPORTS = ( '<stage_path_and_file_name_to_read>' [, '<stage_path_and_file_name_to_read>' ...] ) ]
HANDLER = '<fully_qualified_method_name>'
[ TARGET_PATH = '<stage_path_and_file_name_to_write>' ]
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
AS '<procedure_definition>'
For a stored procedure that uses a precompiled handler, use the following syntax.
CREATE [ OR REPLACE ] [ SECURE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS { <result_data_type> [ [ NOT ] NULL ] | TABLE ( [ <col_name> <col_data_type> [ , ... ] ] ) }
LANGUAGE { SCALA | JAVA }
RUNTIME_VERSION = '<scala_or_java_runtime_version>'
PACKAGES = ( 'com.snowflake:snowpark:<version>' [, '<package_name_and_version>' ...] )
[ IMPORTS = ( '<stage_path_and_file_name_to_read>' [, '<stage_path_and_file_name_to_read>' ...] ) ]
HANDLER = '<fully_qualified_method_name>'
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
JavaScript
CREATE [ OR REPLACE ] [ SECURE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS <result_data_type> [ NOT NULL ]
LANGUAGE JAVASCRIPT
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
AS '<procedure_definition>'
Important
JavaScript is case-sensitive, whereas SQL is not. See Case-sensitivity in JavaScript Arguments for important information about using stored procedure argument names in the JavaScript code.
Python
For in-line stored procedures, use the following syntax:
CREATE [ OR REPLACE ] [ SECURE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS { <result_data_type> [ [ NOT ] NULL ] | TABLE ( [ <col_name> <col_data_type> [ , ... ] ] ) }
LANGUAGE PYTHON
RUNTIME_VERSION = '<python_version>'
PACKAGES = ( 'snowflake-snowpark-python[==<version>]'[, '<package_name>[==<version>]` ... ])
[ IMPORTS = ( '<stage_path_and_file_name_to_read>' [, '<stage_path_and_file_name_to_read>' ...] ) ]
HANDLER = '<function_name>'
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
AS '<procedure_definition>'
For a stored procedure in which the code is in a file on a stage, use the following syntax:
CREATE [ OR REPLACE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS { <result_data_type> [ [ NOT ] NULL ] | TABLE ( [ <col_name> <col_data_type> [ , ... ] ] ) }
LANGUAGE PYTHON
RUNTIME_VERSION = '<python_version>'
PACKAGES = ( 'snowflake-snowpark-python[==<version>]'[, '<package_name>[==<version>]` ... ])
[ IMPORTS = ( '<stage_path_and_file_name_to_read>' [, '<stage_path_and_file_name_to_read>' ...] ) ]
HANDLER = '<module_file_name>.<function_name>'
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
Snowflake Scripting
CREATE [ OR REPLACE ] PROCEDURE <name> ( [ <arg_name> <arg_data_type> ] [ , ... ] )
[ COPY GRANTS ]
RETURNS { <result_data_type> | TABLE ( [ <col_name> <col_data_type> [ , ... ] ] ) }
[ NOT NULL ]
LANGUAGE SQL
[ { CALLED ON NULL INPUT | { RETURNS NULL ON NULL INPUT | STRICT } } ]
[ VOLATILE | IMMUTABLE ] -- Note: VOLATILE and IMMUTABLE are deprecated.
[ COMMENT = '<string_literal>' ]
[ EXECUTE AS { CALLER | OWNER } ]
AS <procedure_definition>
Note
If you are creating a Snowflake Scripting procedure in SnowSQL or the Classic Console, you must
use string literal delimiters ('
or $$
) around
procedure definition
. See Using Snowflake Scripting in SnowSQL and the Classic Console.
Required Parameters¶
All Languages
name ( [ arg_name arg_data_type ] [ , ... ] )
Specifies the identifier (
name
) and any input arguments for the stored procedure.For the identifier:
The identifier does not need to be unique for the schema in which the procedure is created because stored procedures are identified and resolved by the combination of the name and argument types.
The identifier must start with an alphabetic character and cannot contain spaces or special characters unless the entire identifier string is enclosed in double quotes (e.g. “My object”). Identifiers enclosed in double quotes are also case-sensitive. See Identifier Requirements.
For the input arguments:
For
arg_name
, specify the name of the input argument.For
arg_data_type
, use the Snowflake data type that corresponds to the language that you are using.For Java stored procedures, see SQL-Java Data Type Mappings.
For JavaScript stored procedures, see SQL and JavaScript Data Type Mapping.
For Python stored procedures, see SQL-Python Data Type Mappings.
For Scala stored procedures, see SQL-Scala Data Type Mappings.
For Snowflake Scripting, a SQL data type.
Note
For stored procedures you write in Java, Python, or Scala (which use Snowpark APIs), omit the argument for the Snowpark
Session
object.The
Session
argument is not a formal parameter that you specify in CREATE PROCEDURE or CALL. When you call your stored procedure, Snowflake automatically creates aSession
object and passes it to the handler function for your stored procedure.
RETURNS result_data_type [ NOT NULL ]
Specifies the type of the result returned by the stored procedure.
For
result_data_type
, use the Snowflake data type that corresponds to the type of the language that you are using.For Java stored procedures, see SQL-Java Data Type Mappings.
For JavaScript stored procedures, see SQL and JavaScript Data Type Mapping.
For Python stored procedures, see SQL-Python Data Type Mappings.
For Scala stored procedures, see SQL-Scala Data Type Mappings.
For Snowflake Scripting, a SQL data type.
Note
Stored procedures you write in Snowpark (Java or Scala) must have a return value. In Snowpark (Python), when a stored procedure returns no value, it is considered to be returning
None
. Note that every CREATE PROCEDURE statement must include a RETURNS clause that defines a return type, even if the procedure does not explicitly return anything.For
RETURNS TABLE ( [ col_name col_data_type [ , ... ] ] )
, if you know the Snowflake data types of the columns in the returned table, specify the column names and types:CREATE OR REPLACE PROCEDURE get_top_sales() RETURNS TABLE (sales_date DATE, quantity NUMBER) ...
Otherwise (e.g. if you are determining the column types during run time), you can omit the column names and types:
CREATE OR REPLACE PROCEDURE get_top_sales() RETURNS TABLE ()
Note
Currently, in the
RETURNS TABLE(...)
clause inCREATE PROCEDURE
, you cannot specify GEOGRAPHY as a column type.CREATE OR REPLACE PROCEDURE test_return_geography_table_1() RETURNS TABLE(g GEOGRAPHY) ...
If you do so, calling the stored procedure results in the error:
CALL test_return_geography_table_1();
Stored procedure execution error: data type of returned table does not match expected returned table type
To work around this, you can omit the column arguments and types in
RETURNS TABLE()
.CREATE OR REPLACE PROCEDURE test_return_geography_table_1() RETURNS TABLE() ...
RETURNS TABLE(...)
is supported only in when the handler is written in the following languages:
As a practical matter, the returned value cannot be used because the call cannot be part of an expression.
LANGUAGE language
Specifies the language of the stored procedure code. Note that this is optional for stored procedures written with Snowflake Scripting.
Currently, the supported values for
language
include:JAVA
(for Java)JAVASCRIPT
(for JavaScript)PYTHON
(for Python)SCALA
(for Scala)SQL
(for Snowflake Scripting)
Default:
SQL
.AS procedure_definition
Defines the code executed by the stored procedure. The definition can consist of any valid code.
Note the following:
For stored procedures for which the code is not in-line, omit the AS clause. This includes stored procedures with staged handlers.
Instead, use the IMPORTS clause to specify the location of the file containing the code for the stored procedure. For details, see:
For more information on in-line and staged handlers, see Keeping Handler Code In-line or on a Stage.
You must use string literal delimiters (
'
or$$
) aroundprocedure definition
if:You are using a language other than Snowflake Scripting.
You are creating a Snowflake Scripting procedure in SnowSQL or the Classic Console. See Using Snowflake Scripting in SnowSQL and the Classic Console.
For stored procedures in JavaScript, if you are writing a string that contains newlines, you can use backquotes (also called “backticks”) around the string.
The following example of a JavaScript stored procedure uses
$$
and backquotes because the body of the stored procedure contains single quotes and double quotes:CREATE OR REPLACE TABLE table1 ("column 1" VARCHAR);
CREATE or replace PROCEDURE proc3() RETURNS VARCHAR LANGUAGE javascript AS $$ var rs = snowflake.execute( { sqlText: `INSERT INTO table1 ("column 1") SELECT 'value 1' AS "column 1" ;` } ); return 'Done.'; $$;
For languages other than Snowflake Scripting, Snowflake does not completely validate the code when you execute the CREATE PROCEDURE command.
For example, for Snowpark (Scala) stored procedures, the number and types of input arguments are validated, but the body of the function is not validated. If the number or types do not match (e.g. if the Snowflake data type NUMBER is used when the argument is a non-numeric type), executing the CREATE PROCEDURE command causes an error.
If the code is not valid, the CREATE PROCEDURE command will succeed, and errors will be returned when the stored procedure is called.
For more details about stored procedures, see Working with Stored Procedures.
Java, Python, or Scala
RUNTIME_VERSION = 'language_runtime_version'
The language runtime version to use. Currently, the supported versions are:
Java:
11
Python:
3.8
Scala:
2.12
PACKAGES = ( 'snowpark_package_name' [, 'package_name' ...] )
A comma-separated list of the names of packages deployed in Snowflake that should be included in the handler code’s execution environment. The Snowpark package is required for stored procedures, so it must always be referenced in the PACKAGES clause. For more information about Snowpark, see Snowpark API.
By default, the environment in which Snowflake runs stored procedures includes a selected set of packages for supported languages. When you reference these packages in the PACKAGES clause, it is not necessary to reference a file containing the package in the IMPORTS clause because the package is already available in Snowflake. You can also specify the package version.
For the list of supported packages and versions for a given language, query the INFORMATION_SCHEMA.PACKAGES view for rows, specifying the language. For example:
select * from information_schema.packages where language = '<language>';where
language
isjava
,python
, orscala
.The syntax for referring to a package in PACKAGES varies by the package’s language, as described below.
Java
Specify the package name and version number using the following form:
domain:package_name:versionTo specify the latest version, specify
latest
forversion
.For example, to include a package from the latest Snowpark library in Snowflake, use the following:
PACKAGES = ('com.snowflake:snowpark:latest')When specifying a package from the Snowpark library, you must specify version 1.3.0 or later.
Python
Snowflake includes a large number of packages available through Anaconda; for more information, see Using Third-Party Packages.
Specify the package name and version number using the following form:
package_name[==version]To specify the latest version, omit the version number.
For example, to include the spacy package version 2.3.5 (along with the latest version of the required Snowpark package), use the following:
PACKAGES = ('snowflake-snowpark-python', 'spacy==2.3.5')When specifying a package from the Snowpark library, you must specify version 0.4.0 or later. Omit the version number to use the latest version available in Snowflake.
Scala
Specify the package name and version number using the following form:
domain:package_name:versionTo specify the latest version, specify
latest
forversion
.For example, to include a package from the latest Snowpark library in Snowflake, use the following:
PACKAGES = ('com.snowflake:snowpark:latest')Snowflake supports using Snowpark version 0.9.0 or later in a Scala stored procedure. Note, however, that these versions have limitations. For example, versions prior to 1.1.0 do not support the use of transactions in a stored procedure.
HANDLER = 'fully_qualified_method_name'
Python
Use the name of the stored procedure’s function or method. This can differ depending on whether the code is in-line or referenced at a stage.
When the code is in-line, you can specify just the function name, as in the following example:
CREATE OR REPLACE PROCEDURE MYPROC(from_table STRING, to_table STRING, count INT) ... HANDLER = 'run' AS $$ def run(session, from_table, to_table, count): ... $$;When the code is imported from a stage, specify the fully-qualified handler function name as
<module_name>.<function_name>
.CREATE OR REPLACE PROCEDURE MYPROC(from_table STRING, to_table STRING, count INT) ... IMPORTS = ('@mystage/my_py_file.py') HANDLER = 'my_py_file.run';Java and Scala
Use the fully qualified name of the method or function for the stored procedure. This is typically in the following form:
com.my_company.my_package.MyClass.myMethodwhere:
com.my_company.my_packagecorresponds to the package containing the object or class:
package com.my_company.my_package;
Optional Parameters¶
All Languages
SECURE
Specifies that the procedure is secure. For more information about secure procedures, see Protecting Sensitive Information with Secure UDFs and Stored Procedures.
[ [ NOT ] NULL ]
Specifies whether the stored procedure can return NULL values or must return only NON-NULL values.
The default is NULL (i.e. the stored procedure can return NULL).
CALLED ON NULL INPUT
or .RETURNS NULL ON NULL INPUT | STRICT
Specifies the behavior of the stored procedure when called with null inputs. In contrast to system-defined functions, which always return null when any input is null, stored procedures can handle null inputs, returning non-null values even when an input is null:
CALLED ON NULL INPUT
will always call the stored procedure with null inputs. It is up to the procedure to handle such values appropriately.RETURNS NULL ON NULL INPUT
(or its synonymSTRICT
) will not call the stored procedure if any input is null, so the statements inside the stored procedure will not be executed. Instead, a null value will always be returned. Note that the procedure might still return null for non-null inputs.
Default:
CALLED ON NULL INPUT
VOLATILE | IMMUTABLE
Deprecated
Attention
These keywords are deprecated for stored procedures. These keywords are not intended to apply to stored procedures. In a future release, these keywords will be removed from the documentation.
COMMENT = 'string_literal'
Specifies a comment for the stored procedure, which is displayed in the DESCRIPTION column in the SHOW PROCEDURES output.
Default:
stored procedure
EXECUTE AS CALLER
or .EXECUTE AS OWNER
Specifies whether the stored procedure executes with the privileges of the owner (an “owner’s rights” stored procedure) or with the privileges of the caller (a “caller’s rights” stored procedure):
If you execute the statement
CREATE PROCEDURE ... EXECUTE AS CALLER
, then in the future the procedure will execute as a caller’s rights procedure.If you execute
CREATE PROCEDURE ... EXECUTE AS OWNER
, then the procedure will execute as an owner’s rights procedure.
By default (if neither OWNER nor CALLER is specified explicitly at the time the procedure is created), the procedure runs as an owner’s rights stored procedure.
Owner’s rights stored procedures have less access to the caller’s environment (for example the caller’s session variables), and Snowflake defaults to this higher level of privacy and security.
For more information, see Understanding Caller’s Rights and Owner’s Rights Stored Procedures.
Default:
OWNER
COPY GRANTS
Specifies to retain the access privileges from the original procedure when a new procedure is created using CREATE OR REPLACE PROCEDURE.
The parameter copies all privileges, except OWNERSHIP, from the existing procedure to the new procedure. The new procedure will inherit any future grants defined for the object type in the schema. By default, the role that executes the CREATE PROCEDURE statement owns the new procedure.
Note:
The SHOW GRANTS output for the replacement procedure lists the grantee for the copied privileges as the role that executed the CREATE PROCEDURE statement, with the current timestamp when the statement was executed.
The operation to copy grants occurs atomically in the CREATE PROCEDURE command (i.e. within the same transaction).
Java, Python, or Scala
IMPORTS = ( 'stage_path_and_file_name_to_read' [, 'stage_path_and_file_name_to_read' ...] )
The location (stage), path, and name of the file(s) to import. You must set the
IMPORTS
clause to include any files that your stored procedure depends on:If you are writing an in-line stored procedure, you can omit this clause, unless your code depends on classes defined outside the stored procedure or resource files.
Java or Scala: If you are writing a stored procedure with a staged handler, you must also include a path to the JAR file containing the stored procedure’s handler code.
Python: If your stored procedure’s code will be on a stage, you must also include a path to the module file your code is in.
The IMPORTS definition cannot reference variables from arguments that are passed into the stored procedure.
Each file in the
IMPORTS
clause must have a unique name, even if the files are in different subdirectories or different stages.TARGET_PATH = 'stage_path_and_file_name_to_write'
For stored procedure code in Java or Scala, the
TARGET_PATH
clause specifies the location to which Snowflake should write the compiled code (JAR file) after compiling the source code specified in theprocedure_definition
. If this clause is omitted, Snowflake re-compiles the source code each time the code is needed.Note
This clause is not needed for stored procedures whose code is in Python.
If you specify this clause:
You cannot set this to an existing file. Snowflake returns an error if the
TARGET_PATH
points to an existing file.If you specify both the
IMPORTS
andTARGET_PATH
clauses, the file name in theTARGET_PATH
clause must be different from each file name in theIMPORTS
clause, even if the files are in different subdirectories or different stages.If you no longer need to use the stored procedure (e.g. if you drop the stored procedure), you must manually remove this JAR file.
Usage Notes¶
For all stored procedures:
Stored procedures support overloading. Two procedures can have the same name if they have a different number of parameters or different data types for their parameters.
Stored procedures are not atomic; if one statement in a stored procedure fails, the other statements in the stored procedure are not necessarily rolled back. For information about stored procedures and transactions, see Transaction Management.
CREATE OR REPLACE <object> statements are atomic. That is, when the object is replaced, the old object deletion and the new object creation are processed in a single transaction.
Regarding metadata:
Attention
Customers should ensure that no personal data (other than for a User object), sensitive data, export-controlled data, or other regulated data is entered as metadata when using the Snowflake service. For more information, see Metadata Fields in Snowflake.
Tip
If your organization uses a mix of caller’s rights and owner’s rights stored procedures, you might want to use a naming convention for your stored procedures to indicate whether an individual stored procedure is a caller’s rights stored procedure or an owner’s rights stored procedure.
For JavaScript stored procedures:
A JavaScript stored procedure can return only a single value, such as a string (for example, a success/failure indicator) or a number (for example, an error code). If you need to return more extensive information, you can return a VARCHAR that contains values separated by a delimiter (such as a comma), or a semi-structured data type, such as VARIANT.
For Java stored procedures, see the known limitations.
For Python stored procedures, see the known limitations.
For Scala stored procedures, see the known limitations.
Examples¶
This creates a trivial stored procedure that returns a hard-coded value. This is unrealistic, but shows the basic SQL syntax with minimal JavaScript code:
create or replace procedure sp_pi() returns float not null language javascript as $$ return 3.1415926; $$ ;
This shows a more realistic example that includes a call to the JavaScript API. A more extensive version of this
procedure could allow a user to insert data into a table that the user didn’t have privileges to insert into directly.
JavaScript statements could check the input parameters and execute the SQL INSERT
only if certain requirements
were met.
create or replace procedure stproc1(FLOAT_PARAM1 FLOAT) returns string language javascript strict execute as owner as $$ var sql_command = "INSERT INTO stproc_test_table1 (num_col1) VALUES (" + FLOAT_PARAM1 + ")"; try { snowflake.execute ( {sqlText: sql_command} ); return "Succeeded."; // Return a success/error indicator. } catch (err) { return "Failed: " + err; // Return a success/error indicator. } $$ ;
For examples of Java stored procedures, see Writing Stored Procedures in Java.
For examples of Python stored procedures, see Writing Stored Procedures in Python.
For examples of Scala stored procedures, see Writing Stored Procedures in Scala.
For more examples, see Working with Stored Procedures.