Non deterministic test: OperatorRetryTest.testRetryWithBackpressureParallel

[davidmoten]

This fails for me 5 out of 10 times when I run all tests in OperatorRetryTest in Eclipse on a fast machine.

Here's the stack trace:

java.lang.AssertionError: Data content mismatch: {349=[beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, beginningEveryTime, onSuccessOnly]}
    at org.junit.Assert.fail(Assert.java:93)
    at rx.internal.operators.OperatorRetryTest.testRetryWithBackpressureParallel(OperatorRetryTest.java:752)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:606)
    at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:45)
    at org.junit.internal.runners.model.ReflectiveCallable.run(ReflectiveCallable.java:15)
    at org.junit.runners.model.FrameworkMethod.invokeExplosively(FrameworkMethod.java:42)
    at org.junit.internal.runners.statements.InvokeMethod.evaluate(InvokeMethod.java:20)
    at org.junit.internal.runners.statements.FailOnTimeout$StatementThread.run(FailOnTimeout.java:62)

[akarnokd]

This should be fixed now.

[davidmoten]

Not yet:

java.lang.AssertionError: Data content mismatch: 1465={beginningEveryTime}
    at org.junit.Assert.fail(Assert.java:93)
    at rx.internal.operators.OperatorRetryTest.testRetryWithBackpressureParallel(OperatorRetryTest.java:763)

[davidmoten]

@akarnokd can you reopen this issue please

[davidmoten]

Still seeing this failure 50% of the time:

java.lang.AssertionError: Data content mismatch: 2673={beginningEveryTime}, 4222={beginningEveryTime}
    at org.junit.Assert.fail(Assert.java:93)
    at rx.internal.operators.OperatorRetryTest.testRetryWithBackpressureParallel(OperatorRetryTest.java:763)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:497)
    at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:45)
    at org.junit.internal.runners.model.ReflectiveCallable.run(ReflectiveCallable.java:15)
    at org.junit.runners.model.FrameworkMethod.invokeExplosively(FrameworkMethod.java:42)
    at org.junit.internal.runners.statements.InvokeMethod.evaluate(InvokeMethod.java:20)
    at org.junit.internal.runners.statements.FailOnTimeout$StatementThread.run(FailOnTimeout.java:62)

[akarnokd]

I've tried reproducing this error but it never happened. I've run the test 10k times, played with CPU affinity and other thread disruptions to no avail.

[davidmoten]

Have you got an oldish i5 dual core to try it on? I'll have a look at the
failure myself and see what I can find.

On Mon, 27 Apr 2015 21:53 David Karnok notifications@github.com wrote:

I've tried reproducing this error but it never happened. I've run the test
10k times, played with CPU affinity and other thread disruptions to no
avail.

—
Reply to this email directly or view it on GitHub
#2863 (comment).

[akarnokd]

I've tried on a Core 2 Duo E6600 running Windows 7 32 bit and no failure.

[davidmoten]

I've been poking around in OnSubscribeRedo. More doco would be good though it's not too bad. I'm not sure what isLocked is being used for exactly and curious to not see symmetry in its use inasmuch as it gets set to false once and never set to true through the resubscribe lifecycle. Do you see any problems here @akarnokd?

[akarnokd]

I'd wish I understood OnSubscribeRedo better. Despite the many fixes I added, I'm not sure what and how it does things. isLocked caught my eye too and I don't get it either.

I wanted to rewrite the entire operator but I don't understand the protocol for the supplied Func.

[abersnaze]

Sorry that part isn't documented as well. There is a race condition when the observable of restarts is subscribed to and the redo is initiating the first subscribe. The isLocked is to ignore the onNexts from output observable until an onNext sent in to the input.

[abersnaze]

The first message was from my phone so it might not make as much sense as I hoped.

@davidmoten The problem is that this code has to protect itself from the arbitrary code that can exist in the function. The implementations fall into two buckets those that use the scan operator and those that don't. The scan operator is different in that always emits an onNext of the seed synchronously when subscribed. Most operators only emit in reaction to the upstream events.

This led us to the question: What should initiate the first subscription to the source Observable? If the first subscription is to be initiated by the Observable returned by the function then user functions that doesn't use scan would never start but if Redo initiates the first subscription then functions that use scan will subscribe twice immediately.

The path we picked was to make the Redo do the initial subscribe. We didn't want operator to easily fail to do what was expected if there was a misunderstanding in how the user was supposed to write their function to have a scan in it. We handled the double initial subscriptions by not letting them re-subscribe until the initial subscription had completed (aka isLocked).

@akarnokd maybe it would make more sense if it was broken down a bit. The onNext from the Observable returned from the function are mapped to a re-subscription of the source observable. The cool thing about using the onNext as the trigger for re-subscription is that there are many temporal operators to facilitate batching or delaying onNexts. onError and onCompleted are sent as is downstream.

The input to the function is the Observable of terminal Notifications (retry filter down to only the onErrors and repeat filters down to only the onCompletes). The terminal notifications can be counted or delayed in arbitrary was to produce the output behaviour that is desired for when the re-subscription should

[davidmoten]

Thanks @abersnaze, that helps my understanding a little bit. Stuff like that would be a great addition to the comments in this class when we get it working.