Migrations​

1. Start the watch command​

The easiest way to work with your schema in development is by running gel watch. This long-running task will monitor your schema files and automatically apply schema changes in your database as you work.

$ 

gel watch

Initialized. Monitoring "/projects/my-gel-project".

If you get output similar to the output above, you're ready to get started!

2. Write an initial schema​

By convention, your Gel schema is defined inside one or more .gel files that live in a directory called dbschema in the root directory of your codebase.

.
├── dbschema
│   └── default.gel          # schema file (written by you)
└── gel.toml

The schema itself is written using Gel's schema definition language. Edit your dbschema/default.gel and add the following schema inside your module default block:

type User {
  required name: str;
}

type Post {
  required title: str;
  required author: User;
}

It's common to keep your entire schema in a single file, and many users use this default.gel that is created when you start a project. However it's also possible to split their schemas across a number of .gel files.

Once you save your initial schema, assuming it is valid, the watch command will pick it up and apply it to your database.

3. Edit your schema files​

As your application evolves, directly edit your schema files to reflect your desired data model. Try updating your dbschema/default.gel to add a Comment type:

  required author: User;
}

type Comment {
  required content: str;
}

When you save your changes, watch will immediately begin applying your new schema to the database.

Once you have the schema the way you want it, and you're ready to lock it in and commit it to version control, it's time to generate a migration.

4. Generate a migration​

To generate a migration that reflects all your changes, run gel migration create.

$ 

gel migration create

The CLI reads your schema file and sends it to the active Gel instance. The instance compares the file's contents to its current schema state and determines a migration plan. The migration plan is generated by the database itself.

This plan is then presented to you interactively; each detected schema change will be individually presented to you for approval. For each prompt, you have a variety of commands at your disposal. Type y to approve, n to reject, q to cancel the migration, or ? for a breakdown of some more advanced options.

$ 

gel migration create

did you create object type 'default::Comment'? [y,n,l,c,b,s,q,?]
> y
did you create object type 'default::User'? [y,n,l,c,b,s,q,?]
> y
did you create object type 'default::Post'? [y,n,l,c,b,s,q,?]
> y
Created dbschema/migrations/00001.edgeql, id: <hash>

Migration without iteration​

If you want to change the schema, but you already know exactly what you want to change and don't need to iterate on your schema — you want to lock in the migration right away — gel watch might not be the tool you reach for.

Instead, you might use this method:

1. Edit your schema files

2. Create your migration with gel migration create

3. Apply your migration with gel migrate

Since you're not using watch, the schema changes are not applied when you save your schema files. As a result, we need to tack an extra step on the end of the process of applying the migration. That's handled by gel migrate.

$ 

gel migrate

Applied m1virjowa... (00002.edgeql)

Once your migration is applied, you'll see the schema changes reflected in your database.

Data migrations​

Depending on how the schema was changed, data in your database may prevent Gel from applying your schema changes. Imagine we added a required body property to our Post type:

type Post {
  required title: str;
  required body: str;
  required author: User;
}

If we hadn't added any Post objects to our database before this, everything would have worked fine, but it's likely that, in testing out our schema, we did add a Post object. It does not have a body property, but now we've told the database this property is required on all Post objects. The database can't apply this change because existing data would break it.

We have a couple of options here. We could delete all the offending objects.

db> 

delete Post;

{
  default::Post {id: a4a0a40c-d9f5-11ed-8912-1397f7af9fdf},
  default::Post {id: cc051bea-d9f5-11ed-a26d-2b64b6b273a4}
}

Now, if we save the schema again, gel watch will be able to apply it. If we have data in here we don't want to lose though, that's not a good option. In that case, we might drop back to creating and applying the migration outside of gel watch.

To start, run gel migration create. The interactive plan generator will ask you for an EdgeQL expression to map the contents of your database to the new schema.

$ 

gel migration create

did you create property 'body' of object type
'default::Post'? [y,n,l,c,b,s,q,?]
> y
Please specify an expression to populate existing objects in order to make
property 'body' of object type 'default::Post' required:
fill_expr>

Because the body property does not currently exist, the database contains Post objects without it. The expression you provide will be used to assign a body to any Post object that doesn't have one. We'll just provide a simple default: 'No content'.

fill_expr> 'No content'
Created dbschema/migrations/00002.edgeql, id:
m1pjiibv4sa4cao7txpgsbuw2erctmacyrj4qmn45ggapsaztmvxfa

Nice! It accepted our answer and created a new migration file 00002.edgeql. Let's see what the newly created 00002.edgeql file contains.

CREATE MIGRATION m1pjiibv4sa4cao7txpgsbuw2erctmacyrj4qmn45ggapsaztmvxfa
    ONTO m1nlvzbm7buwktkp4vu4shylq6zp2shruokbbssyeidqmmmfqz77yq
{
  ALTER TYPE default::Post {
      CREATE REQUIRED PROPERTY body: std::str {
          SET REQUIRED USING ('No content');
      };
  };
};

We have a CREATE MIGRATION block containing an ALTER TYPE statement to create Post.body as a required property. We can see that our fill expression ('No content') is included directly in the migration file.

Note that we could have provide an arbitrary EdgeQL expression! The following EdgeQL features are often useful:

fill_expr> assert_exists(.body)

fill_expr> assert_single(.sheep)

cast_expr> <bigint>.xp