java - react 堆核心 - java.lang.IllegalStateException : Queue full? !在 Hot Publisher (ConnectableFlux) 上

Question

到目前为止，我一直在使用 RxJava，但我开始使用来自 projectreactor.io 的 reactor-core，因为它符合 react 流规范。

在下面的测试中，我创建了一个生成随机数的热通量 (ConnectableFlux)。我立即连接()它并预取 256 个值(我可以在日志中看到 258 个值)。我等待 5 秒来模拟订阅者直到一段时间后才会订阅。

主线程唤醒后，RnApp 订阅 ConnectableFlux，randomNumberGenerator.subscribe(new RnApp());。然后调用 RnApp.onSubscribe() 并请求 10 个元素。之后，引发了一个java.lang.IllegalStateException: Queue full?!异常(调用了RnApp.onError())，为什么？

订户:

public class RnApp implements Subscriber<Float>{

    private Subscription subscription;
    private List<Float> randomNumbers = new ArrayList<Float>();

    @Override
    public void onComplete() {
        System.out.println("onComplete");
    }

    @Override
    public void onError(Throwable err) {
        err.printStackTrace();
    }

    @Override
    public void onNext(Float f) {
        if(this.randomNumbers.size()>=10){
            this.subscription.cancel();
        }else{
            this.randomNumbers.add(f);
        }
    }

    @Override
    public void onSubscribe(Subscription subs) {
        this.subscription = subs;
        this.subscription.request(10);
    }
}

发布者测试:

@Test
public void randomNumberReading() throws InterruptedException {

    CountDownLatch latch = new CountDownLatch(1);
    ConnectableFlux<Float> randomNumberGenerator = ConnectableFlux.<Float>create( (c) -> {
        SecureRandom sr = new SecureRandom();
        int i = 1;
        while(true){
            try {
                Thread.sleep(10);
            } catch (Exception e) {
                e.printStackTrace();
            }
            System.out.println("-----------------------------------------------------"+(i++));
            c.onNext(sr.nextFloat());
        }
    }).log().subscribeOn(Computations.concurrent()).publish();

    randomNumberGenerator.connect();

    Thread.sleep(5000);

    randomNumberGenerator.subscribe(new RnApp());

    latch.await();
}

日志:

11:12:05.125 [main] DEBUG r.core.util.Logger$LoggerFactory - Using Slf4j logging framework
11:12:05.363 [concurrent-1] INFO  reactor.core.publisher.FluxLog -  onSubscribe(io.pivotal.literx.Part10SubscribeOnPublishOn$$Lambda$1/1586600255@29d4caeb)
11:12:05.371 [concurrent-1] INFO  reactor.core.publisher.FluxLog -  request(256)
-----------------------------------------------------1
11:12:06.000 [concurrent-1] INFO  reactor.core.publisher.FluxLog -  onNext(0.39189225)
-----------------------------------------------------2
...
-----------------------------------------------------257
11:12:08.683 [concurrent-1] INFO  reactor.core.publisher.FluxLog -  onNext(0.34729618)
-----------------------------------------------------258
11:12:08.697 [concurrent-1] INFO  reactor.core.publisher.FluxLog -  onNext(0.7729547)
java.lang.IllegalStateException: Queue full?!
    at reactor.core.publisher.FluxPublish$State.onNext(FluxPublish.java:246)
    at reactor.core.publisher.FluxSubscribeOn$SubscribeOnPipeline.onNext(FluxSubscribeOn.java:134)
    at reactor.core.publisher.FluxLog$LoggerBarrier.doNext(FluxLog.java:130)
    at reactor.core.subscriber.SubscriberBarrier.onNext(SubscriberBarrier.java:85)
    at reactor.core.subscriber.SubscriberWithContext.onNext(SubscriberWithContext.java:92)
    at io.pivotal.literx.Part10SubscribeOnPublishOn.lambda$1(Part10SubscribeOnPublishOn.java:132)
    at reactor.core.publisher.FluxGenerate$ForEachBiConsumer.accept(FluxGenerate.java:145)
    at reactor.core.publisher.FluxGenerate$ForEachBiConsumer.accept(FluxGenerate.java:114)
    at reactor.core.publisher.FluxGenerate$SubscriberProxy.request(FluxGenerate.java:245)
    at reactor.core.subscriber.SubscriberBarrier.doRequest(SubscriberBarrier.java:146)
    at reactor.core.publisher.FluxLog$LoggerBarrier.doRequest(FluxLog.java:160)
    at reactor.core.subscriber.SubscriberBarrier.request(SubscriberBarrier.java:135)
    at reactor.core.util.DeferredSubscription.set(DeferredSubscription.java:71)
    at reactor.core.publisher.FluxSubscribeOn$SubscribeOnPipeline.onSubscribe(FluxSubscribeOn.java:129)
    at reactor.core.publisher.FluxLog$LoggerBarrier.doOnSubscribe(FluxLog.java:122)
    at reactor.core.subscriber.SubscriberBarrier.onSubscribe(SubscriberBarrier.java:67)
    at reactor.core.publisher.FluxGenerate.subscribe(FluxGenerate.java:72)
    at reactor.core.publisher.FluxLog.subscribe(FluxLog.java:67)
    at reactor.core.publisher.FluxSubscribeOn$SourceSubscribeTask.run(FluxSubscribeOn.java:363)
    at reactor.core.publisher.Computations$ProcessorWorker.onNext(Computations.java:919)
    at reactor.core.publisher.Computations$ProcessorWorker.onNext(Computations.java:883)
    at reactor.core.publisher.WorkQueueProcessor$QueueSubscriberLoop.run(WorkQueueProcessor.java:842)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(Unknown Source)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(Unknown Source)
    at java.lang.Thread.run(Unknown Source)

Answer 1

与 RxJava 一样，如果您使用 create()，您将自行处理取消和背压。您可以改为从标准运算符构建生成器:

ConnectableFlux<Double> secureRandomFlux = Flux.using(
    () -> new SecureRandom(),
    sr -> Flux.interval(10, TimeUnit.MILLISECONDS)
          .map(v -> sr.nextDouble())
          .onBackpressureDrop()
    sr -> { }
).publish();