# Federation

Federation is an optional module which can be included in your configuration to enroll with a federated schema.

# Dependencies

caliban-federation only depends on caliban-core and is very unobtrusive.

To use, add the following line to your build.sbt file:

libraryDependencies += "com.github.ghostdogpr" %% "caliban-federation" % "1.1.1"

# Federating

Federation allows graphs to become part of a larger graph without having to share models or create brittle schema stitching code at the gateway level.

You can read more about federation and why it may be useful here (opens new window).

Federation creates a wrapper over your existing schema so that it can add the necessary hooks to support interaction with the gateway.

If you already have a graph you can add federation simply by calling the wrapper function federate

import caliban.federation._

val schema: GraphQL[R] = graphQL(RootResolver(Queries(
  characters = List(Character("Amos"))

val federatedSchema: GraphQL[R] = federate(schema)

This will wrap the bare minimum schema additions around your API so that the gateway will recognize your schema. To actually enable entity resolution you will need to do a bit of leg work.

First, any types that will be "resolvable" need to be annotated with a @key directive. You can use a helper function found in the federation package to help with that.

case class Character(name: String)

The "name" field is a field selector minus the outer braces.

If you need to extend a type from another service, you will need to define a stub version of it in the current service and annotate it with the @extends annotation

@GQLDirective(Key("season episode")) 
case class Episode(@GQLDirective(External) season: Int, @GQLDirective(External) episode: Int, cast: List[Character])

Note the additional annotations we needed in this case. Extend is needed to tell the gateway that this type is defined within another service, while the External flags these fields as being owned by the other service (there are several other annotations available that you are encouraged to read about).

Once you have annotated your types, you need to tell Federation how to resolve those types. Federation uses a slightly different mechanism in resolving types from a standard GraphQL query, so for each type that you wish to support, you will need to add an EntityResolver:

EntityResolver[CharacterService, CharacterArgs, Character](args => 

EntityResolvers like normal field resolvers also supports a "metadata" variant which can be used to inspect the requested fields and potentially optimize the resulting query. You can use the provided helper method if you need to access the metadata field:

EntityResolver.fromMetadata[CharacterArgs](field => args => {
  if (field.fields.forall(_.name == "name")) ZQuery.succeed(Character(args.name, Nil, None))
  else ZQuery.fromEffect(characters.getCharacter(args.name))

In the above we need to define an resolver which takes an R environment type, an A which has an implicit ArgBuilder and an Option[Out] where Out has an implicit Schema[R, Out] available. Creating the above we can now add these resolvers to our federated schema like so:

federate(schema, aResolver, additionalResolvers:_*)

You can now use the resulting GraphQL[R] to start querying. You can also see the full code example here (opens new window)

# Tracing

Federated tracing is slightly different from standard apollo-tracing thus it comes with its own wrapper defined in the caliban-federation module.

import caliban.federation.tracing.ApolloFederatedTracing

val api = federate(schema, resolver, additionalResolvers: _*) @@ ApolloFederatedTracing.wrapper

In federated tracing the gateway communicates with the implementing service via a header apollo-federation-include-trace, for now the only value it can send is ftv1. Thus if you detect this header then you should enable tracing otherwise you can disable it.

If you are using one of the wrappers you are done, they will automatically detect when the gateway enables tracing on a request. However, if you are calling the interpreter.execute independently or you have some other custom set up you will need to add one more step to enable tracing.

If you wish to enable it manually (after detecting the header with your preferred framework) you can call: request.withFederatedTracing which will return a new GraphQLRequest that informs the wrapper that it should include tracing data as part of the response extensions.