pg_graphql: Postgres functions now supported

12 Dec 2023

7 minute read

Supabase GraphQL (pg_graphql) 1.4+ supports the most requested feature: Postgres functions a.k.a. User Defined Functions (UDFs). This addition marks a significant improvement in GraphQL flexibility at Supabase, both as a novel approach to defining entry points into the Graph and as an escape hatch for users to implement custom/complex operations.

As with all entities in Supabase GraphQL, UDFs support is based on automatically reflecting parts of the SQL schema. The feature allow for the execution of custom SQL logic within GraphQL queries to help support complex, user defined, server-side operations with a simple GraphQL interface.

Minimal Example

Consider a function addNums for a basic arithmetic operation:


_10
create function "addNums"(a int, b int default 1)
_10
returns int
_10
immutable
_10
language sql
_10
as $$
_10
select a + b;
_10
$$;

when reflected in the GraphQL schema, the function is exposed as:


_10
type Query {
_10
addNums(a: Int!, b: Int): Int
_10
}

To use this entry point, you could run:


_10
query {
_10
addNums(a: 2, b: 3)
_10
}

which returns the JSON payload:


_10
{
_10
"data": {
_10
"addNums": 5
_10
}
_10
}

Supabase GraphQL does its best to reflect a coherent GraphQL API from all the information known to the SQL layer. For example, the argument a is non-null because it doesn't have a default value while b can be omitted since it does have a default. We also detected that this UDF can be displayed in the Query type rather than the Mutation type because the function was declared as immutable, which means it can not edit the database. Of the other function volatility categories, stable similarly translates into a Query field while volatile (the default) becomes a Mutation field.

Returning Records

In a more realistic example, we might want to return a set of an existing object type like Account. For example, lets say we want to search for accounts based on their email address domains matching a string:


_23
create table "Account"(
_23
id serial primary key,
_23
email varchar(255) not null
_23
);
_23
_23
insert into "Account"(email)
_23
values
_23
('a@foo.com'),
_23
('b@bar.com'),
_23
('c@foo.com');
_23
_23
create function "accountsByEmailDomain"("domainToSearch" text)
_23
returns setof "Account"
_23
stable
_23
language sql
_23
as $$
_23
select
_23
id, email
_23
from
_23
"Account"
_23
where
_23
email ilike ('%@' || "domainToSearch");
_23
$$;

Since our function is stable, it continues to be a field on the Query type. Notice that since we're returning a collection of Account we automatically get support for Relay style pagination on the response including first, last, before, after as well as filtering and sorting.


_35
type Query {
_35
accountsByEmailDomain(
_35
domainToSearch: String!
_35
_35
"""
_35
Query the first `n` records in the collection
_35
"""
_35
first: Int
_35
_35
"""
_35
Query the last `n` records in the collection
_35
"""
_35
last: Int
_35
_35
"""
_35
Query values in the collection before the provided cursor
_35
"""
_35
before: Cursor
_35
_35
"""
_35
Query values in the collection after the provided cursor
_35
"""
_35
after: Cursor
_35
_35
"""
_35
Filters to apply to the results set when querying from the collection
_35
"""
_35
filter: AccountFilter
_35
_35
"""
_35
Sort order to apply to the collection
_35
"""
_35
orderBy: [AccountOrderBy!]
_35
): AccountConnection
_35
}

To complete the example, here's a call to our user defined function:


_10
query {
_10
accountsByEmailDomain(domainToSearch: "foo.com", first: 2) {
_10
edges {
_10
node {
_10
id
_10
email
_10
}
_10
}
_10
}
_10
}

and the response:


_18
{
_18
"data": {
_18
"accountsByEmail": {
_18
"edges": [
_18
{
_18
"node": {
_18
"id": 1,
_18
"email": "a@foo.com"
_18
}
_18
},
_18
"node": {
_18
"id": 3,
_18
"email": "c@foo.com"
_18
}
_18
]
_18
}
_18
}
_18
}

While not shown here, any relationships defined by foreign keys on the response type Account are fully functional so our UDF result is completely connected to the existing Graph.

It’s worth mentioning that we could have supported this query using the default accountCollection field that pg_graphql exposes on the Query type using an ilike filter so the example is only for illustrative purposes.

i.e.:


_10
query {
_10
accountCollection(filter: { email: { ilike: "%foo.com" } }, first: 2) {
_10
edges {
_10
node {
_10
id
_10
email
_10
}
_10
}
_10
}
_10
}

would give the same result as our UDF.

Limitations

The API surface area of SQL functions is surprisingly large. In an effort to bring this feature out sooner, some lesser-used parts have not been implemented yet. Currently functions using the following features are excluded from the GraphQL API:

  • Overloaded functions
  • Functions with a nameless argument
  • Functions returning void
  • Variadic functions
  • Functions that accept a table/views's tuple type as an argument
  • Functions that accept an array type

We look forward to implementing support for many of these features in coming releases.

Takeaways

If you're an existing Supabase user, but new to GraphQL, head over to GraphiQL built right into Supabase Studio for your project to interactively explore your projects through the GraphQL API. User defined function support is new in pg_graphql 1.4+. You can check your project's GraphQL version with:


_10
select *
_10
from pg_available_extensions
_10
where name = 'pg_graphql';

To upgrade, check out our upgrade guide.

For new Supabase users, creating a new project will get you the latest version of Supabase GraphQL with UDF support.

If you're not ready to start a new project but want to learn more about pg_graphql/Supabase GraphQL, our API docs are a great place to learn about how your SQL schema is transformed into a GraphQL API.

Share this article

Build in a weekend, scale to millions