Skip to main content

Versioned migrations for sqlc

In our previous sqlc guide, we saw how we can use Atlas to handle the schema management process using the declarative workflow. This works great in many situations, but some teams prefer the imperative approach where schema changes are explicitly checked-in to source control and verified during code review (see this video to learn why).

To accommodate such cases, Atlas supports another kind of workflow for handling migrations, called versioned migrations, where the changes are stored as SQL files and replayed on the database as needed. This ensures that the code to be executed on the database is known and approved beforehand and gives engineers ultimate control on exactly what code is going to be run in production.

In this guide we will see how we can use sqlc with Atlas, using the versioned migrations workflow.

note

If you haven't already done so, it will be helpful to have read the previous sqlc guide where we have explained some concepts that we will expand on in this guide. You should also check the sqlc getting started with PostgreSQL guide since this guide continues where it left off.

Versioned migrations​

Versioned migrations is a migration workflow where whenever we want to change the database schema, we create a migration file, which is an SQL file containing SQL statements to upgrade the schema from the current version to the next. Many schema management tools, such as Flyway and Liquibase, support this strategy and Atlas supports it as well.

Atlas's versioned migrations workflow works very well with sqlc. In fact, in most cases, you can simply point sqlc schema configuration to the migration directory, and things should just work.

Usually, teams using versioned migrations author the migration files manually. While this is a common approach, there are obvious downsides to it. Writing migrations by hand can be both error-prone and time-consuming for developers.

Migration Authoring​

Atlas has support for a strategy that combines the simplicity of the declarative workflow with the control provided by the versioned one. With this approach, which is called "versioned migration authoring", we still define our desired state in HCL or SQL but use Atlas to compute the required changes, storing them in the migration directory as SQL files.

This strategy brings the best of both worlds, where we can have a single source of truth (our desired state) and have all the authored SQL files that will be used to reach the desired state. To learn more about this approach and see it in action, watch this video.

Now that we have a bit of an understanding of all the concepts, let's see how we can use Atlas with sqlc, using the schema.sql as our desired state and combine versioned migrations with migration authoring.

Creating the initial migration directory​

The first thing we should do is create the initial migration directory. If you are following the sqlc getting started with PostgreSQL your schema.sql and query.sql should look like this:

schema.sql
CREATE TABLE authors
(
    id   BIGSERIAL PRIMARY KEY,
    name text NOT NULL,
    bio  text
);
query.sql
-- name: GetAuthor :one
SELECT *
FROM authors
WHERE id = $1
LIMIT 1;

-- name: ListAuthors :many
SELECT *
FROM authors
ORDER BY name;

-- name: CreateAuthor :one
INSERT INTO authors (name, bio)
VALUES ($1, $2)
RETURNING *;

-- name: DeleteAuthor :exec
DELETE
FROM authors
WHERE id = $1;

We have to generate a diff of our desired state (schema.sql) and our current state (the migration directory, that right now doesn't even exist), we can generate the diff with a simple command:

atlas migrate diff initial_schema \
  --dir "file://migrations" \
  --to "file://schema.sql" \
  --dev-url "docker://postgres?search_path=public"

A new migrations directory should be created with two files in it.

.
├── 20230102211759_initial_schema.sql
└── atlas.sum

As expected, the 20230102211759_initial_schema.sql file holds the commands to create our initial schema.

20230102211759_initial_schema.sql
-- create "authors" table
CREATE TABLE "authors" ("id" bigserial NOT NULL, "name" text NOT NULL, "bio" text NULL, PRIMARY KEY ("id"));
info

Like in the last guide, we need a temporary database to check, simulate and validate the generated queries. This database is provided using the dev-url flag. Atlas has support for running database containers, which is what we are using in this example. Fore more information about dev databases, read about them here.

From the sqlc side, we don't have to do anything since we have not changed our schema.sql or query.sql file.

Migrating the database​

In this guide, we are using a Postgres database running in a local Docker container. You can start such a container by running:

docker run --rm -d --name atlas-sqlc -p 5432:5432 -e POSTGRES_PASSWORD=pass -e POSTGRES_DB=sqlc postgres

Now that we have our base migration directory and target database, we can migrate our schema based on it. To do so, we have to use a command very similar to the one used on the previous guide.

atlas migrate apply --url "postgres://postgres:pass@localhost:5432/sqlc?sslmode=disable"

The command migrate apply connects to the given database URL provided in the url flag, checks the current state of the database and applies any pending migrations.

After running the migrate apply command, you should be able to see an output similar to this:

Migrating to version 20230102211759 (1 migrations in total):

  -- migrating version 20230102211759
	-> CREATE TABLE "authors" ("id" bigserial NOT NULL, "name" text NOT NULL, "bio" text NULL, PRIMARY KEY ("id"));
  -- ok (3.282921ms)

  -------------------------
  -- 4.570005ms
  -- 1 migrations
  -- 1 sql statements

Updating the schema​

The process of changing our schema is quite easy and similar to the declarative way. First we update our schema.sql file, let's add a new age column.

schema.sql
CREATE TABLE authors
(
    id   BIGSERIAL PRIMARY KEY,
    name text NOT NULL,
    bio  text,
    age  integer
);

Let's add a new GetAuthorsByAge query and update our CreateAuthor query as well.

query.sql
-- name: GetAuthor :one
SELECT *
FROM authors
WHERE id = $1
LIMIT 1;

-- name: GetAuthorsByAge :many
SELECT *
FROM authors
WHERE age = $1
ORDER BY age;

-- name: ListAuthors :many
SELECT *
FROM authors
ORDER BY name;

-- name: CreateAuthor :one
INSERT INTO authors (name, bio, age)
VALUES ($1, $2, $3)
RETURNING *;

-- name: DeleteAuthor :exec
DELETE
FROM authors
WHERE id = $1;

When working with sqlc, after every change to the schema.sql or query.sql file, we should execute the sqlc generate command again. Even in the cases where the changes seem to not cause problems, like adding a new column, this is a requirement from sqlc, since it replaces the * with explicit column names.

sqlc generate
note

sqlc only shows messages when encountering an error, if the generator process was successful, no output should be shown.

The models.go and query.sql.go should have been updated accordingly. Now we only have to update our migrations directory, we can use the same command from before, using "add_age_field" as our migration name this time:

atlas migrate diff add_age_field \
  --dir "file://migrations" \
  --to "file://schema.sql" \
  --dev-url "docker://postgres?search_path=public"

If you look at the migration directory, a new SQL file should be created with the statements required for adding a new age column.

.
├── 20230102211759_initial_schema.sql
├── 20230102212813_add_age_field.sql
└── atlas.sum

The new 20230102212813_add_age_field.sql should have all the changes required to reach the desired state from our previous state.

20230102212813_add_age_field.sql
-- modify "authors" table
ALTER TABLE "authors" ADD COLUMN "age" integer NULL;

For applying the migrations, we only have to execute atlas migrate apply again:

atlas migrate apply --url "postgres://postgres:pass@localhost:5432/sqlc?sslmode=disable"

If everything went correctly, you should see the output similar to the one below:

Migrating to version 20230102212813 from 20230102211759 (1 migrations in total):

  -- migrating version 20230102212813
	-> ALTER TABLE "authors" ADD COLUMN "age" integer NULL;
  -- ok (1.565535ms)

  -------------------------
  -- 2.839369ms
  -- 1 migrations
  -- 1 sql statements

Complete workflow​

With the versioned strategy, we can visualize the complete workflow of using Atlas with sqlc as follows:

  • Update the schema (schema.sql)
    • Optionally update the queries (query.sql)
  • Run sqlc to generate the updated code
  • Run atlas migrate diff to compute and store the required changes
  • Run Atlas during the development and migration process, referencing the migration directory.

Wrapping up​

In this guide we saw how Atlas can be used with sqlc in a versioned way. We used Atlas to create migration files, while still keeping a single source of truth for our database schema. The versioned strategy requires more steps, but allows for more control as well.