Hi, RSocket maintainers!
We noticed that in some cases we observe NPEs on cancelling streams. E.g.
java.lang.NullPointerException
at io.rsocket.util.ByteBufPayload.sliceData(ByteBufPayload.java:149)
at io.rsocket.RSocketRequester$3$1.hookOnNext(RSocketRequester.java:351)
at io.rsocket.RSocketRequester$3$1.hookOnNext(RSocketRequester.java:333)
at reactor.core.publisher.BaseSubscriber.onNext(BaseSubscriber.java:160)
at io.rsocket.internal.RateLimitableRequestPublisher$InnerOperator.onNext(RateLimitableRequestPublisher.java:173)
at reactor.core.publisher.FluxJust$WeakScalarSubscription.request(FluxJust.java:99)
at io.rsocket.internal.RateLimitableRequestPublisher.requestN(RateLimitableRequestPublisher.java:124)
at io.rsocket.internal.RateLimitableRequestPublisher.access$400(RateLimitableRequestPublisher.java:29)
at io.rsocket.internal.RateLimitableRequestPublisher$InnerOperator.onSubscribe(RateLimitableRequestPublisher.java:167)
at reactor.core.publisher.FluxJust.subscribe(FluxJust.java:70)
at reactor.core.publisher.Flux.subscribe(Flux.java:8134)
at io.rsocket.internal.RateLimitableRequestPublisher.subscribe(RateLimitableRequestPublisher.java:74)
at io.rsocket.RSocketRequester$3.accept(RSocketRequester.java:332)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.request(FluxPeekFuseable.java:131)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.request(FluxPeekFuseable.java:137)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.request(FluxPeekFuseable.java:137)
at reactor.core.publisher.FluxDoFinally$DoFinallySubscriber.request(FluxDoFinally.java:150)
at reactor.core.publisher.FluxCancelOn$CancelSubscriber.request(FluxCancelOn.java:111)
at reactor.core.publisher.LambdaSubscriber.onSubscribe(LambdaSubscriber.java:119)
at reactor.core.publisher.FluxCancelOn$CancelSubscriber.onSubscribe(FluxCancelOn.java:70)
at reactor.core.publisher.FluxDoFinally$DoFinallySubscriber.onSubscribe(FluxDoFinally.java:117)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.onSubscribe(FluxPeekFuseable.java:171)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.onSubscribe(FluxPeekFuseable.java:171)
at reactor.core.publisher.FluxPeekFuseable$PeekFuseableSubscriber.onSubscribe(FluxPeekFuseable.java:171)
at reactor.core.publisher.UnicastProcessor.subscribe(UnicastProcessor.java:422)
at reactor.core.publisher.InternalFluxOperator.subscribe(InternalFluxOperator.java:53)
at reactor.core.publisher.FluxDefer.subscribe(FluxDefer.java:54)
at reactor.core.publisher.Flux.subscribe(Flux.java:8134)
at reactor.core.publisher.Flux.subscribeWith(Flux.java:8298)
at reactor.core.publisher.Flux.subscribe(Flux.java:8105)
at reactor.core.publisher.Flux.subscribe(Flux.java:8032)
at reactor.core.publisher.Flux.subscribe(Flux.java:7950)
Recently, I found a way to trigger this by subscribing on a cancelled stream twice, here's a reproducer.
@TestpublicvoidtestCancellation() {
TcpServerTransportserverTransport = TcpServerTransport.create("localhost", 0);
CloseableChannelserver =
RSocketFactory.receive()
.acceptor((setup, rsocket) -> Mono.just(newEchoRSocket()))
.transport(serverTransport)
.start()
.block(Duration.ofSeconds(10));
TcpClientTransporttransport = TcpClientTransport.create(server.address());
RSocketclient =
RSocketFactory.connect().transport(transport).start().block(Duration.ofSeconds(10));
Flux<Payload> channel = client.requestChannel(Flux.just(ByteBufPayload.create("foo")));
Flux<Payload> channel = client.requestChannel(Flux.just(ByteBufPayload.create("foo")));
channel.subscribe().dispose();
// operates on a released payloadchannel.subscribe();
}However, I'm not entirely sure how a patch for this would look like. IIUC, cancelling Flux requests all elements, and the same payload gets processed twice after being released. However, maybe it's expected behaviour since the stream is cancelled.
Hi, RSocket maintainers!
We noticed that in some cases we observe NPEs on cancelling streams. E.g.
Recently, I found a way to trigger this by subscribing on a cancelled stream twice, here's a reproducer.
However, I'm not entirely sure how a patch for this would look like. IIUC, cancelling
Fluxrequests all elements, and the same payload gets processed twice after being released. However, maybe it's expected behaviour since the stream is cancelled.