Upgrading to Hasura Migrations Config v3

What has changed?

From v2.0.0 onwards, Hasura allows us to add multiple databases to the same Hasura instance. If we want to take advantage of multiple databases we'll have to use a config v3 project which brings some changes to the project directory when compared to config v2.

Config v3 also brings updates to the Metadata directory structure to improve version control and collaboration workflows.To gain these benefits we recommend upgrading to config v3 even if you do not intend to use multiple databases.

Let's try to understand what these changes are.

Metadata directory

The following is a sample Metadata directory when using config v3.

metadata
├── actions.graphql
├── actions.yaml
├── allow_list.yaml
├── cron_triggers.yaml
├── databases
│   ├── databases.yaml
│   └── s1
│       ├── functions
│       │   ├── functions.yaml
│       │   └── public_search_artists.yaml
│       └── tables
│           ├── public_albums.yaml
│           ├── public_artists.yaml
│           ├── public_genres.yaml
│           ├── public_media_types.yaml
│           ├── public_playlists.yaml
│           ├── public_playlist_track.yaml
│           ├── public_tracks.yaml
│           └── tables.yaml
├── query_collections.yaml
├── remote_schemas.yaml
├── rest_endpoints.yaml
└── version.yaml

Notice the introduction of a new databases directory. As the name suggests, this directory will have the Metadata related to all your databases.

Now let's looks at the contents of databases/databases.yaml

- name: s1
  kind: postgres
  configuration:
    connection_info:
      database_url:
        from_env: DATABASE_URL
      pool_settings:
        idle_timeout: 180
        max_connections: 50
        retries: 1
  tables: '!include s1/tables/tables.yaml'
  functions: '!include s1/functions/functions.yaml'

We can see that we have a database called s1, with it's configuration information and other metadata. Notice the peculiarity of values of tables and functions keys. CLI now uses a special syntax to "include" contents from another file to use it as the value for a particular key. Here it is !include s1/tables/tables.yaml , which means that CLI will try to fill in value of tables from s1/tables/tables.yaml.

Now, If we look at the contents of s1/tables/tables.yaml (as shown below) we understand that it is array whose values are again similarly sourced from different files.

- '!include public_albums.yaml'
- '!include public_artists.yaml'
- '!include public_genres.yaml'
- '!include public_media_types.yaml'
- '!include public_playlist_track.yaml'
- '!include public_playlists.yaml'
- '!include public_tracks.yaml'

Let's look at public_albums.yaml from the first line above.

array_relationships:
  - name: tracks
    using:
      foreign_key_constraint_on:
        column: album_id
        table:
          name: tracks
          schema: public
object_relationships:
  - name: artist
    using:
      foreign_key_constraint_on: artist_id
table:
  name: albums
  schema: public

This file specifies the Metadata related to a table called tracks in public schema. This means that each table will have it's own file for it's Metadata specifically. This makes understanding and managing Metadata related to tables a lot easier.

Also note that Metadata related to custom functions is also managed in a similar fashion.

When doing a hasura metadata apply CLI will inline !include directives with the appropriate contents.

Note

Currently the CLI looks for !include directives in tables and functions keys only.

Migrations directory & Seeds directory

There are some changes to migrations and seeds project directories in config v3. Projects will have child directories corresponding to Migrations for each connected database. Look at the following example, the migrations directory has a sub directory default which corresponds to the connected database.

migrations
└── default
  └── 1613987232674_init
      └── up.sql

Changes needed in existing workflows

With the introduction of multiple databases and it's configuration being stored in metadata, it brings a change of workflow with it.

The steps to apply Migrations and Metadata to a new Hasura instance will be:

# first apply Metadata, this will connect Hasura to the configured databases.
hasura metadata apply
# now we can apply Migrations to the connected databases.
hasura migrate apply --all-databases
# follow it with a Metadata reload to make sure Hasura is aware of any newly created database objects.
hasura metadata reload

The reason why we have to do metadata apply first instead of migrate apply (that we are used to doing in config v2) is that if we do a migrate apply first then Hasura might not yet be aware about the databases it has to connect to as that information is present in metadata. Earlier we could not start Hasura without a connected database, but now we can.

As we are now applying Metadata before Migrations, your Metadata might be in an inconsistent state after the initial metadata apply till the metadata reload step as some database objects referred to in Metadata might not be available till the Migrations are applied.

Also, hasura seeds and hasura migrate commands now accept a required flag --database-name.

Note

As the database connection configuration is now picked up from metadata, it is important to ensure that all the environments use the same env vars to store the database connection strings and the Metadata being applied also uses the appropriate env vars.

Upgrade steps

The latest version Hasura CLI comes with a convenience script used to upgrade your CLI project to use config v3. Note that this process is completely independent from your Hasura GraphQL Engine server update process.

Pre update checklist / notes

1. Config V3 is expected to be used with Hasura GraphQL Engine versions v2.0.0-alpha.1 and above.
2. During the update process CLI uses the server as the source of truth, so make sure your server is up to date.
3. The update process replaces project Metadata with Metadata on the server.

Step 1: Make a backup

Make sure you take a backup of your Hasura Project before upgrading to config v3.

Step 2: Upgrade to the latest CLI

Run:

hasura update-cli

If you are updating to a beta/alpha release

Run:

hasura update-cli --version <version>

Step 3: Upgrade Hasura Project to v3

In your project directory, run:

hasura scripts update-project-v3

Your project directory and config.yaml should be updated to v3.

The update script will ask for the name of database the current Migrations and seeds correspond to.

Continue using config v2

It is possible to continue using config v2 with Hasura versions v2.0.0 and above if you would like to do so.

While using config v2 with Hasura versions v2.0.0 and above, as we have Metadata and Migrations for only a single database, the server assumes that they belong to a database named default and attempts to apply them to it.

Hence, to continue using config v2 we need to connect a database to Hasura GraphQL Engine with the name default and then run any metadata and Migrations commands.

• If you have connected your database using the HASURA_GRAPHQL_DATABASE_URL env var, the database will be added with the name default automatically.
• Else you can connect a database with the name default following the steps here.

Post this, the Metadata and Migration commands should work as usual.

Note

Ensure the version key in config.yaml is set to 2 while using config v2