Using Composite Types in GORM

In PostgreSQL, a composite type is structured like a row or record, consisting of field names and their corresponding data types. Setting a GORM field as a composite type enables you to store complex and structured data in a single column.

This guide explains how to define a schema field type as a composite type in your GORM models and configure the schema migration to manage both the composite types and the GORM models as a single migration unit using Atlas.

Atlas Pro Feature

Atlas support for Composite Types is available exclusively to Pro users. To use this feature, run:

atlas login

Getting started with Atlas and GORM

Before we continue to composite types, ensure you have installed the Atlas GORM Provider on your GORM project.

To set up, follow along the getting started guide for GORM and Atlas.

Composite Schema

The GORM package is mostly used for defining tables (our Go types) and interacting with the database. Composite types, or any other database objects do not have representation in GORM models - a composite type can be defined once, and may be used multiple times in different fields and models.

In order to extend our PostgreSQL schema to include both custom composite types and our GORM types, we configure Atlas to read the state of the schema from a Composite Schema data source. Follow the steps below to configure this for your project:

1. Create a schema.sql that defines the necessary composite type. In the same way, you can configure the composite type in Atlas Schema HCL language:

Using SQL

schema.sql

CREATE TYPE address AS (
  street text,
  city   text
);

Using HCL

schema.hcl

schema "public" {}

composite "address" {
  schema = schema.public
  field "street" {
    type = text
  }
  field "city" {
    type = text
  }
}

2. In your GORM models, define a field that uses the composite type only in PostgreSQL dialect:

User Model

models.go

type User struct {
    gorm.Model
    Name string
    Address Address `gorm:"type:address"`
}

Address Type

address.go

package models

import (
    "database/sql/driver"
    "fmt"
)

type Address struct {
    Street, City string
}

// Scan implements the database/sql.Scanner interface.
func (a *Address) Scan(v interface{}) (err error) {
    switch v := v.(type) {
    case nil:
    case string:
        _, err = fmt.Sscanf(v, "(%q,%q)", &a.Street, &a.City)
    case []byte:
        _, err = fmt.Sscanf(string(v), "(%q,%q)", &a.Street, &a.City)
    }
    return
}

// Value implements the driver.Valuer interface.
func (a *Address) Value() (driver.Value, error) {
    return fmt.Sprintf("(%q,%q)", a.Street, a.City), nil
}

3. In your atlas.hcl config file, add a composite_schema that includes both your custom types defined in schema.sql and your GORM models:

atlas.hcl

data "composite_schema" "app" {
  # Load custom types first.
  schema "public" {
    url = "file://schema.hcl"
  }
  // Then, load the GORM models.
  schema "public" {
    url = data.external_schema.gorm.url
  }
}

env "local" {
  src = data.composite_schema.app.url
  dev = "docker://postgres/15/dev?search_path=public"
}

Usage

After setting up our schema, we can get its representation using the atlas schema inspect command, generate migrations for it, apply them to a database, and more. Below are a few commands to get you started with Atlas:

Inspect the Schema

The atlas schema inspect command is commonly used to inspect databases. However, we can also use it to inspect our composite_schema and print the SQL representation of it:

atlas schema inspect \
  --env local \
  --url env://src \
  --format '{{ sql . }}'

The command above prints the following SQL. Note, the address composite type is defined in the schema before its usage in the address field:

-- Create composite type "address"
CREATE TYPE "address" AS ("street" text, "city" text);
-- Create "users" table
CREATE TABLE "users" ("id" bigserial NOT NULL, "created_at" timestamptz NULL, "updated_at" timestamptz NULL, "deleted_at" timestamptz NULL, "name" text NULL, "address" "address" NULL, PRIMARY KEY ("id"));
-- Create index "idx_users_deleted_at" to table: "users"
CREATE INDEX "idx_users_deleted_at" ON "users" ("deleted_at");

Generate Migrations For the Schema

To generate a migration for the schema, run the following command:

atlas migrate diff \
  --env local

Note that a new migration file is created with the following content:

migrations/20240712090543.sql

-- Create composite type "address"
CREATE TYPE "address" AS ("street" text, "city" text);
-- Create "users" table
CREATE TABLE "users" ("id" bigserial NOT NULL, "created_at" timestamptz NULL, "updated_at" timestamptz NULL, "deleted_at" timestamptz NULL, "name" text NULL, "address" "address" NULL, PRIMARY KEY ("id"));
-- Create index "idx_users_deleted_at" to table: "users"
CREATE INDEX "idx_users_deleted_at" ON "users" ("deleted_at");

Apply the Migrations

To apply the migration generated above to a database, run the following command:

atlas migrate apply \
  --env local \
  --url "postgres://postgres:pass@localhost:5432/database?search_path=public&sslmode=disable"

Apply the Schema Directly on the Database

Sometimes, there is a need to apply the schema directly to the database without generating a migration file. For example, when experimenting with schema changes, spinning up a database for testing, etc. In such cases, you can use the command below to apply the schema directly to the database:

atlas schema apply \
  --env local \
  --url "postgres://postgres:pass@localhost:5432/database?search_path=public&sslmode=disable"

Or, using the Atlas Go SDK:

ac, err := atlasexec.NewClient(".", "atlas")
if err != nil {
	log.Fatalf("failed to initialize client: %w", err)
}
// Automatically update the database with the desired schema.
// Another option, is to use 'migrate apply' or 'schema apply' manually.
if _, err := ac.SchemaApply(ctx, &atlasexec.SchemaApplyParams{
	Env: "local",
	URL: "postgres://postgres:pass@localhost:5432/database?search_path=public&sslmode=disable",
}); err != nil {
    log.Fatalf("failed to apply schema changes: %w", err)
}