graphene-federation

Federation implementation for graphene


Keywords
graphene, graphql, gql, federation
License
MIT
Install
pip install graphene-federation==0.0.2

Documentation

graphene-federation

Federation support for Graphene Logo Graphene following the Apollo Federation specifications.

PyPI version Unit Tests Status Coverage Status Integration Tests Status

This repository is heavily based on the repo it was forked from... Huge thanks to Preply for setting up the foundations.

WARNING: This version is not compatible with graphene version below v3. If you need to use a version compatible with graphene v2 I recommend using the version 1.0.0 of graphene_federation.


Supported Features

At the moment it supports:

  • sdl (_service on field): enable to add schema in federation (as is)
  • @key decorator (entity support): enable to perform queries across service boundaries (you can have more than one key per type)
  • @extend: extend remote types
  • external(): mark a field as external
  • requires(): mark that field resolver requires other fields to be pre-fetched
  • provides()/@provides: annotate the expected returned fieldset from a field on a base type that is guaranteed to be selectable by the gateway.

Each type which is decorated with @key or @extend is added to the _Entity union. The __resolve_reference method can be defined for each type that is an entity. Note that since the notation with double underscores can be problematic in Python for model inheritance this resolver method can also be named _resolve_reference (the __resolve_reference method will take precedence if both are declared).

This method is called whenever an entity is requested as part of the fulfilling a query plan. If not explicitly defined, the default resolver is used. The default resolver just creates instance of type with passed fieldset as kwargs, see entity.get_entity_query for more details

  • You should define __resolve_reference, if you need to extract object before passing it to fields resolvers (example: FileNode)
  • You should not define __resolve_reference, if fields resolvers need only data passed in fieldset (example: FunnyText) Read more in official documentation.

Example

Here is an example of implementation based on the Apollo Federation introduction example. It implements a federation schema for a basic e-commerce application over three services: accounts, products, reviews.

Accounts

First add an account service that expose a User type that can then be referenced in other services by its id field:

from graphene import Field, ID, ObjectType, String
from graphene_federation import build_schema, key

@key("id")
class User(ObjectType):
    id = Int(required=True)
    username = String(required=True)

    def __resolve_reference(self, info, **kwargs):
        """
        Here we resolve the reference of the user entity referenced by its `id` field.
        """
        return User(id=self.id, email=f"user_{self.id}@mail.com")

class Query(ObjectType):
    me = Field(User)

schema = build_schema(query=Query)

Product

The product service exposes a Product type that can be used by other services via the upc field:

from graphene import Argument, ID, Int, List, ObjectType, String
from graphene_federation import build_schema, key

@key("upc")
class Product(ObjectType):
    upc = String(required=True)
    name = String(required=True)
    price = Int()

    def __resolve_reference(self, info, **kwargs):
        """
        Here we resolve the reference of the product entity referenced by its `upc` field.
        """
        return Product(upc=self.upc, name=f"product {self.upc}")

class Query(ObjectType):
    topProducts = List(Product, first=Argument(Int, default_value=5))

schema = build_schema(query=Query)

Reviews

The reviews service exposes a Review type which has a link to both the User and Product types. It also has the ability to provide the username of the User. On top of that it adds to the User/Product types (that are both defined in other services) the ability to get their reviews.

from graphene import Field, ID, Int, List, ObjectType, String
from graphene_federation import build_schema, extend, external, provides

@extend("id")
class User(ObjectType):
    id = external(Int(required=True))
    reviews = List(lambda: Review)

    def resolve_reviews(self, info, *args, **kwargs):
        """
        Get all the reviews of a given user. (not implemented here)
        """
        return []

@extend("upc")
class Product(ObjectType):
    upc = external(String(required=True))
    reviews = List(lambda: Review)

# Note that both the base type and the field need to be decorated with `provides` (on the field itself you need to specify which fields get provided).
@provides
class Review(ObjectType):
    body = String()
    author = provides(Field(User), fields="username")
    product = Field(Product)

class Query(ObjectType):
    review = Field(Review)

schema = build_schema(query=Query)

Federation

Note that each schema declaration for the services is a valid graphql schema (it only adds the _Entity and _Service types). The best way to check that the decorator are set correctly is to request the service sdl:

from graphql import graphql

query = """
query {
    _service {
        sdl
    }
}
"""

result = graphql(schema, query)
print(result.data["_service"]["sdl"])

Those can then be used in a federated schema.

You can find more examples in the unit / integration tests and examples folder.

There is also a cool example of integration with Mongoengine.


Known issues

  1. decorators will not work properly on fields with custom names for example some_field = String(name='another_name')

Contributing

  • You can run the unit tests by doing: make tests.
  • You can run the integration tests by doing make integration-build && make integration-test.
  • You can get a development environment (on a Docker container) with make dev-setup.
  • You should use black to format your code.

The tests are automatically run on Travis CI on push to GitHub.