IllegalStateException at the end of CEP test about flink-spector

[lofifnc]

Do you get this exception with every run of the test?

Currently the LocalFlinkCluster will get shut down after every test. Flink seems to have registered some timer/scheduler with Akka internally. Which is now complaining because the ActorSystem gets shutdown and still has an active timer. I've never seen this exception but I think It can be ignored. I will catch this exception in the Future.

[debelyoo]

Yes, I get it with every run of the test

[lofifnc]

Ok, I'm going to have a detailed look into this on the weekend. The fix should be part of the release planned on monday.

[lofifnc]

I'm unable to reproduce this error. I will catch it but it would be interesting to see where this comes from. What Flink version are you using?

[debelyoo]

I'm using Flink 1.3.1, and Akka 2.4.16

[lofifnc]

I'm still unable to reproduce this. Nonetheless I've included a fix in the latest 0.8.1 release.
Could you test this and provide feedback if this resolves you problem?

[debelyoo]

I tested with release 0.8.1, but the exception still occurs.

However, the exception is not linked to flinkspector. I investigated a bit more, and this exception is actually linked to the Kafka version. Flink uses its own version of akka (forked from akka 2.3.X). I'm using version 2.4.19 and the exception seems to come from a library version conflict. I tested with akka 2.3.9 and the test completes without any exception.

Thanks for your time investigating anyway !

[whipit]

I am running flink version 1.4.1 with flink spector 0.8.1. And I do get this exception.

@debelyoo -- "...I investigated a bit more, and this exception is actually linked to the Kafka version." Do you mean Akka version here?

[lofifnc]

I think Akka version is meant. Are you using your own Akka dependency? Or maybe another dependency is using a different Akka version than Flink.
Could you also try using flink spector version 0.8.3?

[whipit]

AFAIK, I don't explicitly rely on Akka...and the only thing using akka is Flink.

So I just downgraded to flink version 1.3.2 where the issue does not occur. I'll try upgrading and retrying with flinkspector 0.8.3.

FWIW - I also added this SO question: https://stackoverflow.com/questions/49012005/flink-cluster-throws-illegalstateexception-cannot-enqueue-after-timer-shutdown

I referenced this issue in the question. The problem seems to impact some other areas too, so would be good to find out how to resolve long term.

[whipit]

Upgrading to flinkspector 0.8.3 with flink 1.4.1 seems to have fixed it. I have not been able to reproduce the issue. Here's my complete gradle dependency settings

    testCompile group: 'io.flinkspector', name: 'flinkspector-datastream_2.11', version: '0.8.3'
    compile group: 'org.apache.flink', name: 'flink-java', version: '1.4.1'
    compile group: 'org.apache.flink', name: 'flink-streaming-java_2.11', version: '1.4.1'
    compile group: 'org.apache.flink', name: 'flink-clients_2.11', version: '1.4.1'

    compile group: 'org.apache.flink', name: 'flink-connector-kafka-0.11_2.11', version: '1.4.1'
    compile group: 'org.apache.flink', name: 'flink-metrics-dropwizard', version: '1.4.1'
    compile group: 'org.apache.flink', name: 'flink-statebackend-rocksdb_2.11', version: '1.4.1'
    compile group: 'org.apache.flink', name: 'flink-avro', version: '1.4.1'

    compile group: 'com.fasterxml.jackson.core', name: 'jackson-core', version: '2.9.2'
    compile group: 'com.fasterxml.jackson.core', name: 'jackson-databind', version: '2.9.2'
    compile group: 'com.fasterxml.jackson.core', name: 'jackson-annotations', version: '2.9.2'