Skip to main content
Version: v2.x

Postgres: Extend Schema with Custom SQL Functions

What are Custom functions?

Postgres user-defined SQL functions can be used to either encapsulate some custom business logic or extend the built-in SQL functions and operators.

Hasura GraphQL Engine lets you expose certain types of user-defined functions as top level fields in the GraphQL API to allow querying them with either queries or subscriptions, or for VOLATILE functions as mutations. These are called Custom functions.

Functions as computed fields

User-defined SQL functions can also be added as computed fields of tables.

Supported SQL functions

Currently, only functions which satisfy the following constraints can be tracked as Custom functions to be exposed as top level fields in the GraphQL API (terminology from Postgres docs):

  • Function behavior: STABLE or IMMUTABLE functions may only be exposed as queries. VOLATILE functions may be exposed as mutations or queries.
  • Return type: MUST be SETOF <table-name> OR <table-name> where <table-name> is already tracked
  • Argument modes: ONLY IN
Return type workaround

If the required SETOF table doesn't already exist or your function needs to return a custom type i.e. row set, you can create and track an empty table with the required schema to support the function.

Creating SQL functions

SQL functions can be created using SQL statements which can be executed as follows:

Track SQL functions

Functions can be present in the underlying Postgres database without being exposed over the GraphQL API. In order to expose a function as a top level field in the GraphQL API, it needs to be tracked as custom functions.

While creating functions from the Data -> SQL page, selecting the Track this checkbox will expose the new function over the GraphQL API right after creation if it is supported.

You can track any existing supported functions in your database from the Data -> Schema page:

Track functions

Use cases

Custom functions are ideal solutions for retrieving some derived data based on some custom business logic that requires user input to be calculated. If your custom logic does not require any user input, you can use views instead.

Let's see a few example use cases for custom functions:

Example: Text-search functions

Let's take a look at an example where the SETOF table is already part of the existing schema:

articles(id integer, title text, content text)

Let's say we've created and tracked a custom function, search_articles, with the following definition:

CREATE FUNCTION search_articles(search text)
RETURNS SETOF articles AS $$
SELECT *
FROM articles
WHERE
title ilike ('%' || search || '%')
OR content ilike ('%' || search || '%')
$$ LANGUAGE sql STABLE;

This function filters rows from the articles table based on the input text argument, search i.e. it returns SETOF articles. Assuming the articles table is being tracked, you can use the custom function as follows:

GraphiQL
Query Variables
Request Headers
Documentation Explorer
No Schema Available

Example: Fuzzy match search functions

Let's look at an example of a street address text search with support for misspelled queries.

First install the pg_trgm PostgreSQL extension:

CREATE EXTENSION pg_trgm;

Next create a GIN (or GIST) index in your database for the columns you'll be querying:

CREATE INDEX address_gin_idx ON properties
USING GIN ((unit || ' ' || num || ' ' || street || ' ' || city || ' ' || region || ' ' || postcode) gin_trgm_ops);

And finally create the custom SQL function in the Hasura Console:

CREATE FUNCTION search_properties(search text)
RETURNS SETOF properties AS $$
SELECT *
FROM properties
WHERE
search <% (unit || ' ' || num || ' ' || street || ' ' || city || ' ' || region || ' ' || postcode)
ORDER BY
similarity(search, (unit || ' ' || num || ' ' || street || ' ' || city || ' ' || region || ' ' || postcode)) DESC
LIMIT 5;
$$ LANGUAGE sql STABLE;

Assuming the properties table is being tracked, you can use the custom function as follows:

GraphiQL
Query Variables
Request Headers
Documentation Explorer
No Schema Available

Example: PostGIS functions

Let's take a look at an example where the SETOF table is not part of the existing schema.

Say you have 2 tables, for user and landmark location data, with the following definitions (this example uses the popular spatial database extension, PostGIS):

-- User location data
CREATE TABLE user_location (
user_id INTEGER PRIMARY KEY,
location GEOGRAPHY(Point)
);

-- Landmark location data
CREATE TABLE landmark (
id SERIAL PRIMARY KEY,
name TEXT,
type TEXT,
location GEOGRAPHY(Point)
);

In this example, we want to fetch a list of landmarks that are near a given user, along with the user's details in the same query. PostGIS' built-in function ST_Distance can be used to implement this use case.

Since our use case requires an output that isn't a "subset" of any of the existing tables i.e. the SETOF table doesn't exist, let's first create this table and then create our location search function.

  • create and track the following table:

    -- SETOF table
    CREATE TABLE user_landmarks (
    user_id INTEGER,
    location GEOGRAPHY(Point),
    nearby_landmarks JSON
    );
  • create and track the following function:

    -- function returns a list of landmarks near a user based on the
    -- input arguments distance_kms and userid
    CREATE FUNCTION search_landmarks_near_user(userid integer, distance_kms integer)
    RETURNS SETOF user_landmarks AS $$
    SELECT A.user_id, A.location,
    (SELECT json_agg(row_to_json(B)) FROM landmark B
    WHERE (
    ST_Distance(
    ST_Transform(B.location::Geometry, 3857),
    ST_Transform(A.location::Geometry, 3857)
    ) /1000) < distance_kms
    ) AS nearby_landmarks
    FROM user_location A where A.user_id = userid
    $$ LANGUAGE sql STABLE;

This function fetches user information (for the given input userid) and a list of landmarks which are less than distance_kms kilometers away from the user's location as a JSON field. We can now refer to this function in our GraphQL API as follows:

GraphiQL
Query Variables
Request Headers
Documentation Explorer
No Schema Available

Querying custom functions using GraphQL queries

Aggregations on custom functions

You can query aggregations on a function result using the <function-name>_aggregate field.

For example, count the number of articles returned by the function defined in the text-search example above:

query {
search_articles_aggregate(args: { search: "hasura" }) {
aggregate {
count
}
}
}

Using arguments with custom functions

As with tables, arguments like where, limit, order_by, offset, etc. are also available for use with function-based queries.

For example, limit the number of articles returned by the function defined in the text-search example above:

query {
search_articles(args: { search: "hasura" }, limit: 5) {
id
title
content
}
}

Using argument default values for custom functions

If you omit an argument in the args input field then the GraphQL Engine executes the SQL function without the argument. Hence, the function will use the default value of that argument set in its definition.

For example: In the above PostGIS functions example, the function definition can be updated as follows:

-- input arguments distance_kms (default: 2) and userid
CREATE FUNCTION search_landmarks_near_user(userid integer, distance_kms integer default 2)

Search nearby landmarks with distance_kms default value which is 2 kms:

GraphiQL
Query Variables
Request Headers
Documentation Explorer
No Schema Available

Accessing Hasura session variables in custom functions

Create a function with an argument for session variables and track it with the pg_track_function Metadata API with the session_argument config set. The session argument will be a JSON object where keys are session variable names (in lower case) and values are strings. Use the ->> JSON operator to fetch the value of a session variable as shown in the following example.

-- single text column table
CREATE TABLE text_result(
result text
);

-- simple function which returns the hasura role
-- where 'hasura_session' will be session argument
CREATE FUNCTION get_session_role(hasura_session json)
RETURNS SETOF text_result AS $$
SELECT q.* FROM (VALUES (hasura_session ->> 'x-hasura-role')) q
$$ LANGUAGE sql STABLE;
GraphiQL
Query Variables
Request Headers
Documentation Explorer
No Schema Available
Note

The specified session argument will not be included in the <function-name>_args input object in the GraphQL schema.

Tracking functions with side effects

You can also use the pg_track_function Metadata API to track VOLATILE functions as mutations.

Aside from showing up under the mutation root (and presumably having side-effects), these tracked functions behave the same as described above for queries.

We also permit tracking VOLATILE functions under the query root, in which case the user needs to guarantee that the field is idempotent and side-effect free, in the context of the resulting GraphQL API. One such use case might be a function that wraps a simple query and performs some logging visible only to administrators.

Note

It's easy to accidentally give an SQL function the wrong volatility (or for a function to end up with VOLATILE mistakenly, since it's the default).

Permissions for custom functions

A custom function f is only accessible to a role r if there is a function permission (see Create function permission) defined on the function f for the role r. Additionally, role r must have SELECT permissions on the returning table of the function f.

Access control permissions configured for the SETOF table of a function are also applicable to the function itself.

For example, in our text-search example above, if the role user has access only to certain columns of the table article, a validation error will be thrown if the search_articles query is run selecting a column to which the user role doesn't have access to.

Note

In case of functions exposed as queries, if the Hasura GraphQL Engine is started with inferring of function permissions set to true (by default: true) then a function exposed as a query will be accessible to a role even if the role doesn't have a function permission for the function - provided the role has select permission defined on the returning table of the function.