NullPointerException on connection disposal

[SerCeMan]

Hey, folks!

We've noticed the following error periodically when the connections are being disposed. If I'm reading the exception correctly, it seems like the collection stores null at some point.

java.lang.NullPointerException: null
    at java.lang.Iterable.forEach(Iterable.java:75)
    at io.rsocket.core.RSocketResponder.cleanUpSendingSubscriptions(RSocketResponder.java:274)
    at io.rsocket.core.RSocketResponder.cleanup(RSocketResponder.java:264)
    at io.rsocket.core.RSocketResponder.tryTerminate(RSocketResponder.java:166)
    at io.rsocket.core.RSocketResponder.tryTerminateOnConnectionClose(RSocketResponder.java:159)
    at reactor.core.publisher.LambdaMonoSubscriber.onComplete(LambdaMonoSubscriber.java:132)
    at reactor.core.publisher.MonoProcessor$NextInner.onComplete(MonoProcessor.java:518)
    at reactor.core.publisher.MonoProcessor.onNext(MonoProcessor.java:308)
    at reactor.core.publisher.MonoProcessor.onComplete(MonoProcessor.java:265)
    at io.rsocket.internal.BaseDuplexConnection.dispose(BaseDuplexConnection.java:23)
    at ...

Steps to Reproduce

It seems that the error isn't easily reproducible. We're only seeing very rarely, once or twice a day.

Your Environment

• RSocket version(s) used: 1.0.1
• Other relevant libraries versions (eg. netty, ...): none
• Platform (eg. JVM version (javar -version) or Node version (node --version)): 13
• OS and version (eg uname -a): ubuntu 20.04

[OlegDokuka]

Hey @SerCeMan!

Thank you for reporting that!
I will have a look.

On the root-cause... Not sure we put null somewhere, it is more likely a racing between cleanup during disposal and a stream that does self-removal from the same collection when it is terminated.

So it may be that the iterator has seen an element but a collection has already been shrunken so once a next is called, the element being observed is null, hence we see NPE.

[rstoyanchev]

This should be resolved now.

[Cesarla]

Hi @OlegDokuka, @rstoyanchev can these improvements be back-ported to rsocket 1.1.X? I am facing the same problems with the version 1.1.0

[OlegDokuka]

@Cesarla any reproducer to share?

[Cesarla]

It happens while cleaning channel-based communication:

java.lang.NullPointerException: null
at java.lang.Iterable.forEach (Iterable.java:75)
at io.rsocket.core.RSocketResponder.cleanUpSendingSubscriptions (RSocketResponder.java:186)
at io.rsocket.core.RSocketResponder.doOnDispose (RSocketResponder.java:174)
at io.rsocket.core.RSocketResponder.tryTerminate (RSocketResponder.java:108)
at io.rsocket.core.RSocketResponder.tryTerminateOnConnectionClose (RSocketResponder.java:101)
at reactor.core.publisher.LambdaMonoSubscriber.onComplete (LambdaMonoSubscriber.java:135)
at reactor.core.publisher.NextProcessor$NextInner.onComplete (NextProcessor.java:403)
at reactor.core.publisher.NextProcessor.tryEmitValue (NextProcessor.java:185)
at reactor.core.publisher.NextProcessor.tryEmitEmpty (NextProcessor.java:137)
at reactor.core.publisher.NextProcessor.onComplete (NextProcessor.java:132)
at io.rsocket.resume.ResumableDuplexConnection.dispose (ResumableDuplexConnection.java:213)
at io.rsocket.resume.ServerRSocketSession.onNext (ServerRSocketSession.java:215)
at io.rsocket.resume.ServerRSocketSession.onNext (ServerRSocketSession.java:40)
at com.XXXX.eventstream.MdcContextLifter.onNext (MdcContextLifter.java:29)
at reactor.core.publisher.MonoDelay$MonoDelayRunnable.run (MonoDelay.java:119)
at reactor.core.scheduler.SchedulerTask.call (SchedulerTask.java:68)
at reactor.core.scheduler.SchedulerTask.call (SchedulerTask.java:28)
at java.util.concurrent.FutureTask.run (FutureTask.java:264)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run (ScheduledThreadPoolExecutor.java:304)
at java.util.concurrent.ThreadPoolExecutor.runWorker (ThreadPoolExecutor.java:1128)
at java.util.concurrent.ThreadPoolExecutor$Worker.run (ThreadPoolExecutor.java:628)
at java.lang.Thread.run (Thread.java:834)

[OlegDokuka]

yeah. It is actually IntObjectMap bug that may send unused values so they are null which endups with NPE

[yschimke]

Is this still happening?

[rstoyanchev]

@Cesarla this should be fixed in 1.1.1 snapshots.