Version 6 pre-stable
This version of Silverstripe CMS has not yet been given a stable release. See the release roadmap for more information. Go to documentation for the most recent stable version.

Getting started

Server Requirements
What you will need to run GraphQL on a web server
Activating the default server
Open up the default server that comes pre-configured with the module
Configuring your schema
Add a basic type to the schema configuration
Building the schema
Turn your schema configuration into executable code
Building a schema with procedural code
Use PHP code to build your schema
Deploying the schema
Deploy your GraphQL schema to a test or production environment

Configuring your schema

GraphQL is a strongly-typed API layer, so having a schema behind it is essential. Simply put:

Queries are just fields on a type called "query". They can take arguments, and they must be resolved.

There's a bit more to it than that, and if you want to learn more about GraphQL, you can read the full documentation, but for now, these three concepts will serve almost all of your needs to get started.

It is strongly recommended that you install the GraphQL devtools module to help with testing your API.

Included in that module is a GraphQLSchemaInitTask task to initialise a basic GraphQL schema to get you started. Instructions for using the task are included in the module's README.md.

Initial setup

To start your first schema, open a new configuration file. Let's call it graphql.yml.

# app/_config/graphql.yml
SilverStripe\GraphQL\Schema\Schema:
  schemas:
    # your schemas here

Let's populate a schema that is pre-configured for us out of the box called "default".

# app/_config/graphql.yml
SilverStripe\GraphQL\Schema\Schema:
  schemas:
    default:
      config:
        # general schema config here
      types:
        # your generic types here
      models:
        # your DataObjects here
      queries:
        # your queries here
      mutations:
        # your mutations here
      enums:
        # your enums here

Avoid config flushes

Because the schema definition is only consumed at build time and never used at runtime, it doesn't make much sense to store it in the configuration layer, because it just means you'll have to flush=1 every time you make a schema update, which will slow down your builds.

It is recommended that you store your schema YAML outside of the _config directory to increase performance and remove the need for flushing when you build your schema.

This doesn't mean there is never a need to flush=1 when building your schema. If you were to add a new schema, make a change to the value of this src attribute, or create new PHP classes, those are still standard config changes which won't take effect without a flush.

We can do this by adding a src key to our app/_config/graphql.yml schema definition that maps to a directory relative to the project root.

# app/_config/graphql.yml
SilverStripe\GraphQL\Schema\Schema:
  schemas:
    default:
      src:
        - app/_graphql

Your src must be an array. This allows further source files to be merged into your schema. This feature can be use to extend the schema of third party modules.

Your directory can also be relative to a module reference, e.g. somevendor/somemodule: _graphql:

# app/_config/graphql.yml
SilverStripe\GraphQL\Schema\Schema:
  schemas:
    default:
      src:
        - app/_graphql
        - module/_graphql
        # The next line would map to `vendor/somevendor/somemodule/_graphql`
        - 'somevendor/somemodule: _graphql'

Now, in the new app/_graphql folder, we can create YAML file definitions.

# app/_graphql/schema.yml

# no schema key needed. it's implied!
config:
  # your schema config here
types:
  # your generic types here
models:
  # your DataObjects here
bulkLoad:
  # your bulk loader directives here
queries:
  # your queries here
mutations:
  # your mutations here
enums:
  # your enums here

Namespacing your schema files

Your schema YAML file will get quite bloated if it's just used as a monolithic source of truth like this. We can tidy this up quite a bit by simply placing the files in directories that map to the keys they populate -- e.g. config/, types/, models/, queries/, mutations/, etc.

There are two approaches to namespacing:

  • By filename
  • By directory name
Namespacing by filename

If the filename is named one of the seven keywords used in the app/_graphql/schema.yml example above, it will be implicitly placed in the corresponding section of the schema - e.g. any configuration added to app/_graphql/config.yml will be implicitly added to SilverStripe\GraphQL\Schema\Schema.schemas.default.config.

This only works in the root source directory (i.e. app/_graphql/some-directory/config.yml will not work).

# app/_graphql/config.yml

# my config here
# app/_graphql/types.yml

# my types here
# app/_graphql/models.yml

# my models here
# app/_graphql/enums.yml

# my enums here
# app/_graphql/bulkLoad.yml

# my bulk loader directives here
Namespacing by directory name

If you use a parent directory name (at any depth) of one of the seven keywords, it will be implicitly placed in the corresponding section of the schema - e.g. any configuration added to a .yml file in app/_graphql/config/ will be implicitly added to SilverStripe\GraphQL\Schema\Schema.schemas.default.config.

The names of the actual files here do not matter. You could for example have a separate file for each of your types, e.g. app/_graphql/types/my-first-type.yml.

# app/_graphql/config/config.yml

# my config here
# app/_graphql/types/types.yml

# my types here
# app/_graphql/models/models.yml

# my models here
# app/_graphql/enums/enums.yml

# my enums here
# app/_graphql/bulkLoad/bulkLoad.yml

# my bulk loader directives here
Going even more granular

These special directories can contain multiple files that will all merge together, so you can even create one file per type, or some other convention. All that matters is that the parent directory name or the filename matches one of the schema keys.

The following are perfectly valid:

  • app/_graphql/config/config.yml maps to SilverStripe\GraphQL\Schema\Schema.schemas.default.config
  • app/_graphql/types/allElementalBlocks.yml maps to SilverStripe\GraphQL\Schema\Schema.schemas.default.types
  • app/_graphql/news-and-blog/models/blog.yml maps to SilverStripe\GraphQL\Schema\Schema.schemas.default.models
  • app/_graphql/mySchema.yml maps to SilverStripe\GraphQL\Schema\Schema.schemas.default

Schema config

Each schema can declare a generic configuration section, config. This is mostly used for assigning or removing plugins and resolvers.

An important subsection of config is modelConfig, where you can configure settings for specific models, e.g. DataObject.

Like the other sections, it can have its own config.yml, or just be added as a config: mapping to a generic schema YAML document.

# app/_graphql/config.yml
modelConfig:
  DataObject:
    plugins:
      inheritance: true
    operations:
      read:
        plugins:
          readVersion: false
          paginateList: false

You can learn more about plugins and resolvers in the query plugins, plugins, building a custom query, and resolver discovery sections.

Defining a basic type

Let's define a generic type for our GraphQL schema.

Generic types don't map to DataObject classes - they're useful for querying more 'generic' data (hence the name). You'll learn more about adding DataObjects in working with DataObjects.

# app/_graphql/types.yml

Country:
  fields:
    name: String
    code: String
    population: Int
    languages: '[String]'

If you're familiar with GraphQL type language, this should look pretty familiar.

There are only a handful of scalar types available in GraphQL by default. They are:

  • String
  • Int
  • Float
  • Boolean

To define a type as a list, you wrap it in brackets: [String], [Int]

To define a type as required (non-null), you add an exclamation mark: String!

Often times, you may want to do both: [String!]!

Look out for the footgun, here. Make sure your bracketed type is in quotes (i.e. '[String]', not [String]), otherwise it's valid YAML that will get parsed as an array!

That's all there is to it! To learn how we can take this further, check out the working with generic types documentation. Otherwise, let's get started on adding some DataObjects.

Further reading

Server Requirements
What you will need to run GraphQL on a web server
Activating the default server
Open up the default server that comes pre-configured with the module
Configuring your schema
Add a basic type to the schema configuration
Building the schema
Turn your schema configuration into executable code
Building a schema with procedural code
Use PHP code to build your schema
Deploying the schema
Deploy your GraphQL schema to a test or production environment