Workflow & schema customization

You are not satisfied with your generated schema? Follow these steps in order to get your schema closer to your API.

Note

The warnings emitted by ./manage.py spectacular --file schema.yaml --validate are intended as an indicator to where drf-spectacular discovered issues. Sane fallbacks are used wherever possible and some warnings might not even be relevant to you. The remaining issues can be solved with the following steps.

Step 1: queryset and serializer_class

Introspection heavily relies on those two attributes. get_serializer_class() and get_serializer() are also used if available. You can also set those on APIView. Even though this is not supported by DRF, drf-spectacular will pick them up and use them.

Step 2: @extend_schema

Decorate your view functions with the @extend_schema decorator. There is a multitude of override options, but you only need to override what was not properly discovered in the introspection.

class PersonView(viewsets.GenericViewSet):
    @extend_schema(
        parameters=[
          QuerySerializer,  # serializer fields are converted to parameters
          OpenApiParameter("nested", QuerySerializer),  # serializer object is converted to a parameter
          OpenApiParameter("queryparam1", OpenApiTypes.UUID, OpenApiParameter.QUERY),
          OpenApiParameter("pk", OpenApiTypes.UUID, OpenApiParameter.PATH), # path variable was overridden
        ],
        request=YourRequestSerializer,
        responses=YourResponseSerializer,
        # more customizations
    )
    def retrieve(self, request, pk, *args, **kwargs)
        # your code

Note

responses can be detailed further by providing a dictionary instead. This could be for example {201: YourResponseSerializer, ...} or {(200, 'application/pdf'): OpenApiTypes.BINARY, ...}.

Note

For simple responses, you might not go through the hassle of writing an explicit serializer class. In those cases, you can simply specify the request/response with a call to inline_serializer. This lets you conveniently define the endpoint’s schema inline without actually writing a serializer class.

Note

If you want to annotate methods that are provided by the base classes of a view, you have nothing to attach @extend_schema to. In those instances you can use @extend_schema_view to conveniently annotate the default implementations.

class XViewset(mixins.ListModelMixin, viewsets.GenericViewSet):
    @extend_schema(description='text')
    def list(self, request, *args, **kwargs):
        return super().list(request, *args, **kwargs)

is equivalent to

@extend_schema_view(
    list=extend_schema(description='text')
)
class XViewset(mixins.ListModelMixin, viewsets.GenericViewSet):
    ...

Note

You may also use @extend_schema on views to attach annotations to all methods in that view (e.g. tags). Method annotations will take precedence over view annotation.

Step 3: @extend_schema_field and type hints

A custom SerializerField might not get picked up properly. You can inform drf-spectacular on what is to be expected with the @extend_schema_field decorator. It takes either basic types or a Serializer as argument. In case of basic types (e.g. str, int, etc.) a type hint is already sufficient.

@extend_schema_field(OpenApiTypes.BYTE)  # also takes basic python types
class CustomField(serializers.Field):
    def to_representation(self, value):
        return urlsafe_base64_encode(b'\xf0\xf1\xf2')

You can apply it also to the method of a SerializerMethodField.

class ErrorDetailSerializer(serializers.Serializer):
    field_custom = serializers.SerializerMethodField()

    @extend_schema_field(OpenApiTypes.DATETIME)
    def get_field_custom(self, object):
        return '2020-03-06 20:54:00.104248'

Step 4: @extend_schema_serializer

You may also decorate your serializer with @extend_schema_serializer. Mainly used for excluding specific fields from the schema or attaching request/response examples. On rare occasions (e.g. envelope serializers), overriding list detection with many=False may come in handy.

@extend_schema_serializer(
    exclude_fields=('single',), # schema ignore these fields
    examples = [
         OpenApiExample(
            'Valid example 1',
            summary='short summary',
            description='longer description',
            value={
                'songs': {'top10': True},
                'single': {'top10': True}
            },
            request_only=True, # signal that example only applies to requests
            response_only=False, # signal that example only applies to responses
        ),
    ]
)
class AlbumSerializer(serializers.ModelSerializer):
    songs = SongSerializer(many=True)
    single = SongSerializer(read_only=True)

    class Meta:
        fields = '__all__'
        model = Album

Step 5: Extensions

The core purpose of extensions is to make the above customization mechanisms also available for library code. Usually, you cannot easily decorate or modify View, Serializer or Field from libraries. Extensions provide a way to hook into the introspection without actually touching the library.

All extensions work on the same principle. You provide a target_class (import path string or actual class) and then state what drf-spectcular should use instead of what it would normally discover.

Important

The extensions register themselves automatically. Just be sure that the Python interpreter sees them at least once.

It is good practice to collect your extensions in YOUR_MAIN_APP_NAME/schema.py and importing that file in your YOUR_MAIN_APP_NAME/apps.py. Every proper Django app will already have an auto-generated apps.py file. Although not strictly necessary, doing the import in ready() is the most robust approach. It will make sure your environment (e.g. settings) is properly set up prior to loading.

# your_main_app_name/apps.py
class YourMainAppNameConfig(AppConfig):
    default_auto_field = "django.db.models.BigAutoField"
    name = "your_main_app_name"

    def ready(self):
        import your_main_app_name.schema  # noqa: E402

Note

Only the first Extension matching the criteria is used. By setting the priority attribute on your extension, you can influence the matching order (default 0). Built-in Extensions have a priority of -1. If you subclass built-in Extensions, don’t forget to increase the priority.

Replace views with OpenApiViewExtension

Many libraries use @api_view or APIView instead of ViewSet or GenericAPIView. In those cases, introspection has very little to work with. The purpose of this extension is to augment or switch out the encountered view (only for schema generation). Simply extending the discovered class class Fixed(self.target_class) with a queryset or serializer_class attribute will often solve most issues.

class Fix4(OpenApiViewExtension):
    target_class = 'oscarapi.views.checkout.UserAddressDetail'

    def view_replacement(self):
        from oscar.apps.address.models import UserAddress

        class Fixed(self.target_class):
            queryset = UserAddress.objects.none()
        return Fixed

Specify authentication with OpenApiAuthenticationExtension

Authentication classes that do not have 3rd party support will emit warnings and be ignored. Luckily authentication extensions are very easy to implement. Have a look at the default authentication method extensions. A simple custom HTTP header based authentication could be achieved like this:

class MyAuthenticationScheme(OpenApiAuthenticationExtension):
    target_class = 'my_app.MyAuthentication'  # full import path OR class ref
    name = 'MyAuthentication'  # name used in the schema

    def get_security_definition(self, auto_schema):
        return {
            'type': 'apiKey',
            'in': 'header',
            'name': 'api_key',
        }

Declare field output with OpenApiSerializerFieldExtension

This is mainly targeted to custom SerializerField’s that are within library code. This extension is functionally equivalent to @extend_schema_field

class CategoryFieldFix(OpenApiSerializerFieldExtension):
    target_class = 'oscarapi.serializers.fields.CategoryField'

    def map_serializer_field(self, auto_schema, direction):
        # equivalent to return {'type': 'string'}
        return build_basic_type(OpenApiTypes.STR)

Declare serializer magic with OpenApiSerializerExtension

This is one of the more involved extension mechanisms. drf-spectacular uses those to implement polymorphic serializers. The usage of this extension is rarely necessary because most custom Serializer classes stay very close to the default behaviour.

In case your Serializer makes use of a custom ListSerializer (i.e. a custom to_representation()), you can write a dedicated extensions for that. This is usually the case when many=True does not result in a plain list, but rather in augmented object with additional fields (e.g. envelopes).

Declare custom/library filters with OpenApiFilterExtension

This extension only applies to filter and pagination classes and is rarely used. Built-in support for django-filter is realized with this extension. OpenApiFilterExtension replaces the filter’s native get_schema_operation_parameters with your customized version, where you have full access to drf-spectacular’s more advanced introspection features.

Step 6: Postprocessing hooks

The generated schema is still not to your liking? You are no easy customer, but there is one more thing you can do. Postprocessing hooks run at the very end of schema generation. This is how the choice Enum are consolidated into component objects. You can register hooks with the POSTPROCESSING_HOOKS setting.

def custom_postprocessing_hook(result, generator, request, public):
    # your modifications to the schema in parameter result
    return result

Note

Please note that setting POSTPROCESSING_HOOKS will override the default. If you intend to keep the Enum hook, be sure to add 'drf_spectacular.hooks.postprocess_schema_enums' back into the list.

Step 7: Preprocessing hooks

Preprocessing hooks are applied shortly after collecting all API operations and before the actual schema generation starts. They provide an easy mechanism to alter which operations should be represented in your schema. You can exclude specific operations, prefix paths, introduce or hardcode path parameters or modify view initiation. additional hooks with the PREPROCESSING_HOOKS setting.

def custom_preprocessing_hook(endpoints):
    # your modifications to the list of operations that are exposed in the schema
    for (path, path_regex, method, callback) in endpoints:
        pass
    return endpoints

Note

A common use case would be the removal of duplicated {format}-suffixed operations, for which we already provide the drf_spectacular.hooks.preprocess_exclude_path_format hook. You can simply enable this hook by adding the import path string to the PREPROCESSING_HOOKS.

Congratulations

You should now have no more warnings and a spectacular schema that satisfies all your requirements. If that is not the case, feel free to open an issue and make a suggestion for improvement.