Awesome

graphql-java-reactive

Reactive Streams-based execution strategy for http://github.com/graphql-java/graphql-java.

Overview
Hello World
Migration
License

Overview

While GraphQL allows you to define very rich queries with multiple levels of depth, it is not possible to track the data changes efficiently. Once you received the data, you have to poll your GraphQL API. Moreover, even if a single field changed the whole object will be sent.

Subscription model helps to avoid it, and instead of sending the data it sends the changes.

Hello World

This is "hello world" in graphql-java-reactive:

GraphQLSchema schema = newSchema()
        .query(newObject()
                .name("helloWorldQuery")
                .field(newFieldDefinition()
                        .type(GraphQLString)
                        .name("hello")
                        .dataFetcher(env -> Flowable
                                .interval(1, TimeUnit.SECONDS)
                                .take(3)
                                .map(i -> "world " + i))
                )
        )
        .build();

GraphQL graphQL = new GraphQL(schema, new ReactiveExecutionStrategy());

Publisher<Change> changes = (Publisher<Change>) graphQL.execute("{ hello }").getData();

Flowable.fromPublisher(changes).timeInterval().blockingSubscribe(System.out::println);
// Prints:
// Timed[time=1017, unit=MILLISECONDS, value=Change{path=[], data={hello=world 0}}]
// Timed[time=1000, unit=MILLISECONDS, value=Change{path=[hello], data=world 1}]
// Timed[time=1000, unit=MILLISECONDS, value=Change{path=[hello], data=world 2}]

As you can see, initial object was dispatched as a change with an empty path.
Since hello is defined as an asynchronous publisher with 1 second interval, we received it after 1 second.

Every other change is more specific and changes only hello property.

P.S. This example uses RxJava 2, but you can use any Reactive Streams compatible implementation like Project Reactor, RxJava 1, etc...

Migration

Notable change from the standard execution strategy is that the result is not a value, but Publisher of changes. However, first change always contains fully resolved object.
It means that you can use .take(1).map(it -> it.getData()) to emulate old behavior, but with Reactive types.

DataFetcher might return non-Reactive type. By default only CompletableFuture will be treated as asynchronous type (you can override adapt method from ReactiveExecutionStrategy to add your own types), every other type will be considered as a static value.

License