Table of Contents

Custom Mutations

PostGraphile automatically generates CRUD Mutations for you; but it's rare that these will cover all your needs - and many people just disable them outright. Custom mutations enable you to write exactly the business logic you need with access to all of your data all wrapped up in a PostgreSQL function. You can even bypass the RLS and GRANT checks, should you so choose, by tagging your function as SECURITY DEFINER - but be very careful when you do so!

To create a function that PostGraphile will recognise as a custom mutation, it must obey the following rules:

Functions matching these requirements will be represented in GraphQL in a way that is compatible with the Relay Input Object Mutations Specification. For example the function

CREATE FUNCTION my_function(a int, b int) RETURNS text AS $$ … $$ LANGUAGE sql VOLATILE;

could be called from GraphQL like this:

mutation {
  myFunction(input: { a: 1, b: 2 }) {

Look at the documentation in GraphiQL to find the parameters you may use!


Here's an example of a custom mutation, which will generate the graphql acceptTeamInvite mutation:

CREATE FUNCTION app_public.accept_team_invite(team_id integer)
RETURNS app_public.team_members
AS $$
  UPDATE app_public.team_members
    SET accepted_at = now()
    WHERE accepted_at IS NULL
    AND team_members.team_id = accept_team_invite.team_id
    AND member_id = app_public.current_user_id()

Notes on the above function:

  • STRICT is optional, it means that if any of the arguments are null then the mutation will not be called (and will thus return null with no error) - this allows us to mark teamId as a required argument.
  • SECURITY INVOKER is the default, it means the function will run with the security of the person who invoked the function
  • SECURITY DEFINER means that the function will run with the security of the person who defined the function, typically the database owner - this means that the function may bypass RLS, RBAC and other security concerns. Be careful when using SECURITY DEFINER - think of it like sudo!
  • we use LANGUAGE sql here, but you can use LANGUAGE plpgsql if you need variables or looping or if blocks or similar concerns; or if you want to write in a more familiar language you can use LANGUAGE plv8 (JavaScript, requires extension), or one of the built in LANGUAGE options such as Python, Perl or Tcl

If you'd like PostGraphile to treat all function arguments as required (non-null) unless they have a default then you can use the graphileBuildOptions.pgStrictFunctions = true setting. This is similar to marking the function as STRICT but with the subtle difference that arguments with defaults may be specified as NULL without necessitating that the function returns null. With this setting enabled, arguments without default value will be set mandatory while arguments with default value will be optional. For example: CREATE FUNCTION foo(a int, b int, c int = 0, d int = null)... would give a mutation foo(a: Int!, b: Int!, c: Int, d: Int). To set this in the library version, pass it as part of graphileBuildOptions:

  postgraphile(connectionString, schemaName, {
    graphileBuildOptions: {
      pgStrictFunctions: true,

To use it with the CLI you need to do similar using the .postgraphilerc.js file.

Bulk Insert Example

Here's an example of a custom mutation that performs a "bulk insert" - inserting and returning a set of records:

CREATE FUNCTION app_public.create_documents(num integer, type text, location text)
RETURNS SETOF app_public.document
AS $$
  INSERT INTO app_public.document (type, location)
    SELECT create_documents.type, create_documents.location
    FROM generate_series(1, num) i

A note on anonymous types: if you have a function that RETURNS SETOF TABLE(a int, b text) (an anonymous record type) then PostGraphile will not currently pick it up due to the common PostGraphile function restrictions. It's easy to work around - define a named type:

CREATE TYPE my_function_return_type AS (
  a int,
  b text

and then change your function to RETURNS SETOF my_function_return_type.