Non-Blocking Reactive Streams Foundation for the JVM both implementing a [Reactive Extensions] (http://reactivex.io) inspired API and efficient event streaming support.
3.0 requires Java 8 or + to run.
With Gradle from repo.spring.io or Maven Central repositories (stable releases only):
repositories {
//maven { url 'http://repo.spring.io/snapshot' }
mavenCentral()
}
dependencies {
//compile "io.projectreactor:reactor-core:3.0.6.BUILD-SNAPSHOT"
compile "io.projectreactor:reactor-core:3.0.5.RELEASE"
}New to Reactive Programming or bored of reading already ? Try the Introduction to Reactor Core hands-on !
If you are familiar with RxJava or if you want to check more detailled introduction, be sure to check https://www.infoq.com/articles/reactor-by-example !
A Reactive Streams Publisher with basic flow operators.
- Static factories on Flux allow for source generation from arbitrary callbacks types.
- Instance methods allows operational building, materialized on each Flux#subscribe(), Flux#subscribe() or multicasting operations such as Flux#publish and Flux#publishNext.
Flux in action :
Flux.fromIterable(getSomeLongList())
.mergeWith(Flux.interval(100))
.doOnNext(serviceA::someObserver)
.map(d -> d * 2)
.take(3)
.onErrorResumeWith(errorHandler::fallback)
.doAfterTerminate(serviceM::incrementTerminate)
.subscribe(System.out::println);A Reactive Streams Publisher constrained to ZERO or ONE element with appropriate operators.
- Static factories on Mono allow for deterministic zero or one sequence generation from arbitrary callbacks types.
- Instance methods allows operational building, materialized on each Mono#subscribe() or Mono#get() eventually called.
Mono in action :
Mono.fromCallable(System::currentTimeMillis)
.then(time -> Mono.first(serviceA.findRecent(time), serviceB.findRecent(time)))
.timeout(Duration.ofSeconds(3), errorHandler::fallback)
.doOnSuccess(r -> serviceM.incrementSuccess())
.subscribe(System.out::println);Blocking Mono result :
Tuple2<Long, Long> nowAndLater =
Mono.when(
Mono.just(System.currentTimeMillis()),
Flux.just(1).delay(1).map(i -> System.currentTimeMillis()))
.block();Reactor uses a Scheduler as a contract for arbitrary task execution. It provides some guarantees required by Reactive Streams flows like FIFO execution.
You can use or create efficient schedulers to jump thread on the producing flows (subscribeOn) or receiving flows (publishOn):
Mono.fromCallable( () -> System.currentTimeMillis() )
.repeat()
.publishOn(Schedulers.single())
.log("foo.bar")
.flatMap(time ->
Mono.fromCallable(() -> { Thread.sleep(1000); return time; })
.subscribeOn(Schedulers.parallel())
, 8) //maxConcurrency 8
.subscribe();ParallelFlux can starve your CPU's from any sequence whose work can be subdivided in concurrent
tasks. Turn back into a Flux with ParallelFlux#sequential(), an unordered join or
use abitrary merge strategies via 'groups()'.
Mono.fromCallable( () -> System.currentTimeMillis() )
.repeat()
.parallel(8) //parallelism
.runOn(Schedulers.parallel())
.doOnNext( d -> System.out.println("I'm on thread "+Thread.currentThread()) ).
.sequential()
.subscribe()To bridge a Subscriber or Processor into an outside context that is taking care of
producing non concurrently, use Flux#create, Mono#create, or
FluxProcessor#connectSink().
Flux.create(emitter -> {
ActionListener al = e -> {
emitter.next(textField.getText());
};
// without cancellation support:
button.addActionListener(al);
// with cancellation support:
emitter.setCancellation(() -> {
button.removeListener(al);
});
},
// Overflow (backpressure) handling, default is BUFFER
FluxSink.OverflowStrategy.LATEST)
.timeout(3)
.doOnComplete(() -> System.out.println("completed!"))
.subscribe(System.out::println)The 3 main processor implementations are message relays using 0 (EmitterProcessor) or N threads (TopicProcessor and WorkQueueProcessor). They also use bounded buffers, aka RingBuffer.
A signal broadcaster that will safely handle asynchronous boundaries between N Subscribers (asynchronous or not) and a parent producer.
EmitterProcessor<Integer> emitter = EmitterProcessor.create();
BlockingSink<Integer> sink = emitter.connectSink();
sink.next(1);
sink.next(2);
emitter.subscribe(System.out::println);
sink.next(3); //output : 3
sink.finish();A caching broadcaster that will safely handle asynchronous boundaries between N Subscribers (asynchronous or not) and a parent producer.
Replay capacity in action:
ReplayProcessor<Integer> replayer = ReplayProcessor.create();
BlockingSink<Integer> sink = replayer.connectSink();
sink.submit(1);
sink.submit(2);
replayer.subscribe(System.out::println); //output 1, 2
replayer.subscribe(System.out::println); //output 1, 2
sink.submit(3); //output : ...3 ...3
sink.finish();Note : ReplayProcessor does not explicitly require a call to connectSink() since its demand upstream is constant and unbounded (and it will retain the specified history number of items). That means you could just call onNext on the processor directly without problems.
An asynchronous signal broadcaster dedicating an event loop thread per subscriber and maxing out producing/consuming rate with temporary tolerance to latency peaks. Also supports multi-producing and emission without onSubscribe.
TopicProcessor<Integer> topic = TopicProcessor.create();
topic.subscribe(System.out::println);
topic.onNext(1); //output : ...1
topic.onNext(2); //output : ...2
topic.subscribe(System.out::println); //output : ...1, 2
topic.onNext(3); //output : ...3 ...3
topic.onComplete();Similar to TopicProcessor regarding thread per subscriber but this time exclusively distributing the input data signal to the next available Subscriber. WorkQueueProcessor is also able to replay detected dropped data downstream (error or cancel) to any Subscriber ready.
WorkQueueProcessor<Integer> queue = WorkQueueProcessor.create();
queue.subscribe(System.out::println);
queue.subscribe(System.out::println);
queue.onNext(1); //output : ...1
queue.onNext(2); //output : .... ...2
queue.onNext(3); //output : ...3
queue.onComplete();Most of this cool stuff uses bounded ring buffer implementation under the hood to mitigate signal processing difference between producers and consumers. Now, the operators and processors or any standard reactive stream component working on the sequence will be instructed to flow in when these buffers have free room AND only then. This means that we make sure we both have a deterministic capacity model (bounded buffer) and we never block (request more data on write capacity). Yup, it's not rocket science after all, the boring part is already being worked by us in collaboration with Reactive Streams Commons on going research effort.
"Operator Fusion" (flow optimizers), health state observers, helpers to build custom reactive components, bounded queue generator, hash-wheel timer, converters from/to Java 9 Flow, Publisher and Java 8 CompletableFuture. The reactor-addons repository contains a reactor-test project with test features like the StepVerifier.
https://projectreactor.io/docs/core/release/api/
https://github.com/reactor/lite-rx-api-hands-on
https://www.infoq.com/articles/reactor-by-example
- Everything to jump outside the JVM with the non-blocking drivers from Reactor IPC.
- Reactor Addons include Bus and Pipes event routers plus a handful of extra reactive modules.
Powered by Reactive Streams Commons
Licensed under Apache Software License 2.0
Sponsored by Pivotal