User-Defined Functions

On this page

A user-defined function (UDF) is a named function defined at the database level that can be called in queries and other contexts. CockroachDB supports invoking UDFs in SELECT, FROM, and WHERE clauses of DML statements.

Both stored procedures and user-defined functions are types of routines. However, they differ in the following ways:

Functions return a value, and procedures do not return a value.
Procedures must be invoked using a CALL statement. Functions can be invoked in nearly any context, such as SELECT, FROM, and WHERE clauses, DEFAULT expressions, and computed column expressions.
Functions have volatility settings, and procedures do not.

Overview

The basic components of a user-defined function are a name, list of arguments, return type, volatility, language, and function body.

An argument has a mode and a type.
- CockroachDB supports the IN (default), OUT, and INOUT argument modes. For an example, see Create a function that uses OUT and INOUT parameters.
- The type can be a built-in type, user-defined ENUM or composite type, or implicit record type. A type can have a DEFAULT value.
The return type can be a built-in SQL type, user-defined ENUM or composite type, RECORD, PL/pgSQL REFCURSOR type, implicit record type, TRIGGER, or VOID.
- Preceding a type with SETOF indicates that a set, or multiple rows, may be returned. For an example, see Create a function that returns a set of results.
- VOID indicates that there is no return type and NULL will always be returned.
The volatility indicates whether the function has side effects. VOLATILE and NOT LEAKPROOF are the default.
- Annotate a function with side effects with VOLATILE. This also prevents the cost-based optimizer from pre-evaluating the function.
- A STABLE or IMMUTABLE function does not mutate data. You cannot create a STABLE or IMMUTABLE function that executes a mutation (INSERT, UPSERT, UPDATE, DELETE) statement.
- LEAKPROOF indicates that a function has no side effects and that it communicates nothing that depends on its arguments besides the return value (i.e., it cannot throw an error that depends on the value of its arguments). You must precede LEAKPROOF with IMMUTABLE, and only IMMUTABLE can be set to LEAKPROOF. NOT LEAKPROOF is allowed with any other volatility.
- Non-VOLATILE functions can be optimized through inlining. For more information, see Create an inlined UDF.
LANGUAGE specifies the language of the function body. CockroachDB supports the languages SQL and PLpgSQL (PL/pgSQL).
The function body:
- Can reference arguments by name or by their ordinal in the function definition with the syntax $1.
- Can be enclosed in a single line with single quotes '' or multiple lines with $$.
- Can reference tables.
- Can reference only the SELECT statement.

Examples

Create a UDF

The following is a UDF that returns the sum of two integers:

CREATE FUNCTION add(a INT, b INT) RETURNS INT IMMUTABLE LEAKPROOF LANGUAGE SQL AS 'SELECT a + b';

Where:

name: add
arguments: a of type INT, b of type INT
return type: INT
volatility: IMMUTABLE LEAKPROOF
language: SQL
function body: 'SELECT a + b'

Alternatively, you could define this function as:

CREATE FUNCTION add(a INT, b INT) RETURNS INT IMMUTABLE LEAKPROOF LANGUAGE SQL AS 'SELECT $1 + $2';

Or as:

CREATE FUNCTION add(a INT, b INT) RETURNS INT LANGUAGE SQL AS $$
  SELECT a + b;
$$;

For more examples of UDF creation, see CREATE FUNCTION.

View a UDF definition

To view the definition for the add() function:

SHOW CREATE FUNCTION add;

If you do not specify a schema for the function add when you create it, the default schema is public:

  function_name |                 create_statement
----------------+---------------------------------------------------
 add            | CREATE FUNCTION public.add(IN a INT8, IN b INT8)
                |     RETURNS INT8
                |     IMMUTABLE
                |     LEAKPROOF
                |     CALLED ON NULL INPUT
                |     LANGUAGE SQL
                |     AS $$
                |     SELECT a + b;
                | $$
(1 row)

Invoke a UDF

You invoke a UDF like a built-in function.

To invoke the add() function:

SELECT add(3,5) as sum;

  sum
-------
    8
(1 row)

Create a UDF using PL/pgSQL

The following user-defined function returns the nth integer in the Fibonacci sequence.

It uses the PL/pgSQL LOOP syntax to iterate through a simple calculation, and RAISE EXCEPTION to return an error message if the specified n is negative.

CREATE FUNCTION fib(n int) RETURNS INT AS $$
    DECLARE
        tmp INT;
        a INT := 0;
        b INT := 1;
        i INT := 2;
    BEGIN
        IF n < 0 THEN
            RAISE EXCEPTION 'n must be non-negative';
        END IF;
        IF n = 0 THEN RETURN 0; END IF;
        IF n = 1 THEN RETURN 1; END IF;
        LOOP
            IF i > n THEN EXIT; END IF;
            tmp := a + b;
            a := b;
            b := tmp;
            i := i + 1;
        END LOOP;
        RETURN b;
    END
  $$ LANGUAGE PLpgSQL;

SELECT fib(8);

  fib
-------
   21

Create an inlined UDF

When possible, the cost-based optimizer will improve a function's performance by inlining the UDF within the query plan. The UDF must have the following attributes:

It is labeled as IMMUTABLE, STABLE, or LEAKPROOF (i.e., non-VOLATILE).
It has a single statement.
It is not a set-returning function.
Its arguments are only variable or constant expressions.
It is not a record-returning function.

The following example demonstrates how inlining improves a UDF's performance.

Create tables a and b:

CREATE TABLE a (
  a INT
);

CREATE TABLE b (
  b INT PRIMARY KEY
);

Insert a value (10) into 1000 rows in a and 1 row in b:
```
INSERT INTO a SELECT 10 FROM generate_series(1, 1000);
INSERT INTO b VALUES (10);
```

Create a VOLATILE function foo_v() and a STABLE function foo_s():

CREATE FUNCTION foo_v(x INT) RETURNS INT VOLATILE LANGUAGE SQL AS $$
  SELECT b FROM b WHERE b = x
$$;

CREATE FUNCTION foo_s(x INT) RETURNS INT STABLE LANGUAGE SQL AS $$
  SELECT b FROM b WHERE b = x
$$;

Each function returns a specified value from table b.

View the query plan when foo_v() (the VOLATILE function) is used in a selection query to retrieve equal values from table a:

EXPLAIN ANALYZE SELECT foo_v(a) FROM a WHERE a = 10;

                                            info
--------------------------------------------------------------------------------------------
  planning time: 2ms
  execution time: 77ms
  distribution: local
  vectorized: true
  rows read from KV: 1,000 (39 KiB, 1 gRPC calls)
  cumulative time spent in KV: 330µs
  maximum memory usage: 80 KiB
  network usage: 0 B (0 messages)
  sql cpu time: 75ms
  estimated RUs consumed: 0

  • render
  │
  └── • filter
      │ nodes: n1
      │ actual row count: 1,000
      │ sql cpu time: 75ms
      │ estimated row count: 1,000
      │ filter: a = 10
      │
      └── • scan
            nodes: n1
            actual row count: 1,000
            KV time: 330µs
            KV contention time: 0µs
            KV rows read: 1,000
            KV bytes read: 39 KiB
            KV gRPC calls: 1
            estimated max memory allocated: 60 KiB
            sql cpu time: 87µs
            estimated row count: 1,000 (100% of the table; stats collected 19 seconds ago)
            table: a@a_pkey
            spans: FULL SCAN
(33 rows)

The query takes 77ms to execute because the function is invoked for each row scanned in table a.

View the query plan when using foo_s() (the STABLE function) instead:

EXPLAIN ANALYZE SELECT foo_s(a) FROM a WHERE a = 10;

                                              info
------------------------------------------------------------------------------------------------
  planning time: 5ms
  execution time: 4ms
  distribution: local
  vectorized: true
  rows read from KV: 1,001 (39 KiB, 2 gRPC calls)
  cumulative time spent in KV: 832µs
  maximum memory usage: 420 KiB
  network usage: 0 B (0 messages)
  sql cpu time: 3ms
  estimated RUs consumed: 0

  • render
  │
  └── • merge join (left outer)
      │ nodes: n1
      │ actual row count: 1,000
      │ estimated max memory allocated: 340 KiB
      │ estimated max sql temp disk usage: 0 B
      │ sql cpu time: 3ms
      │ estimated row count: 1,000
      │ equality: (a) = (b)
      │ right cols are key
      │
      ├── • filter
      │   │ nodes: n1
      │   │ actual row count: 1,000
      │   │ sql cpu time: 5µs
      │   │ estimated row count: 1,000
      │   │ filter: a = 10
      │   │
      │   └── • scan
      │         nodes: n1
      │         actual row count: 1,000
      │         KV time: 722µs
      │         KV contention time: 0µs
      │         KV rows read: 1,000
      │         KV bytes read: 39 KiB
      │         KV gRPC calls: 1
      │         estimated max memory allocated: 60 KiB
      │         sql cpu time: 202µs
      │         estimated row count: 1,000 (100% of the table; stats collected 42 seconds ago)
      │         table: a@a_pkey
      │         spans: FULL SCAN
      │
      └── • scan
            nodes: n1
            actual row count: 1
            KV time: 110µs
            KV contention time: 0µs
            KV rows read: 1
            KV bytes read: 30 B
            KV gRPC calls: 1
            estimated max memory allocated: 20 KiB
            sql cpu time: 11µs
            estimated row count: 1 (100% of the table; stats collected 42 seconds ago)
            table: b@b_pkey
            spans: FULL SCAN
(57 rows)

The query takes only 4ms to execute because the function is inlined and transformed to a join with an equality comparison (a) = (b), which has much less overhead than invoking a function for each row scanned in table a.

Video Demo

For a deep-dive demo on UDFs, watch the following video:

Known limitations

User-defined functions have the following limitations:

A RECORD-returning UDF cannot be created without a RETURN statement in the root block, which would restrict the wildcard type to a concrete one. #122945
User-defined functions are not currently supported in:
- Expressions (column, index, constraint) in tables. #87699
- Views. #87699
User-defined functions cannot call themselves recursively. #93049
Common table expressions (CTE), recursive or non-recursive, are not supported in user-defined functions (UDF). That is, you cannot use a WITH clause in the body of a UDF. #92961
The setval function cannot be resolved when used inside UDF bodies. #110860
Casting subqueries to user-defined types in UDFs is not supported. #108184
Routines cannot be invoked with named arguments, e.g., SELECT foo(a => 1, b => 2); or SELECT foo(b := 1, a := 2);. #122264
Routines cannot be created if they reference temporary tables. #121375
Routines cannot be created with unnamed INOUT parameters. For example, CREATE PROCEDURE p(INOUT INT) AS $$ BEGIN NULL; END; $$ LANGUAGE PLpgSQL;. #121251
Routines cannot be created if they return fewer columns than declared. For example, CREATE FUNCTION f(OUT sum INT, INOUT a INT, INOUT b INT) LANGUAGE SQL AS $$ SELECT (a + b, b); $$;. #121247
Routines cannot be created with an OUT parameter of type RECORD. #123448
DDL statements (e.g., CREATE TABLE, CREATE INDEX) are not allowed within UDFs or stored procedures. #110080
Polymorphic types cannot be cast to other types (e.g., TEXT) within routine parameters. #123536
Routine parameters and return types cannot be declared using the ANYENUM polymorphic type, which is able to match any ENUM type. 123048