Windowed KTables appear to have a concurrency issue.

[AntonyLittle]

Description

I am seeing the following exception when using Windowed KTables in Streamiz 1.6.0:

fail: Streamiz.Kafka.Net.Processors.StreamThread[0]
      stream-thread[urn.menzies.falcona.interface_tableruntobarcodestable-1424c242-4ed5-48a9-927f-c3e6b8abe430-stream-thread-0] Encountered the following error during processing:
      System.InvalidOperationException: Collection was modified; enumeration operation may not execute.
         at System.Collections.Generic.HashSet`1.Enumerator.MoveNext()
         at Streamiz.Kafka.Net.State.InMemory.InMemoryWindowStore.RemoveExpiredData()
         at Streamiz.Kafka.Net.State.InMemory.InMemoryWindowStore.Fetch(Bytes key, Int64 time)
         at Streamiz.Kafka.Net.State.Metered.MeteredWindowStore`2.<>c__DisplayClass20_0.<Fetch>b__0()
         at Streamiz.Kafka.Net.Crosscutting.ActionHelper.MeasureLatency[T](Func`1 actionToMeasure, Sensor sensor)
         at Streamiz.Kafka.Net.Processors.KStreamWindowAggregateProcessor`4.Process(K key, V value)
         at Streamiz.Kafka.Net.Processors.AbstractProcessor`2.Forward(IEnumerable`1 processors, Action`1 action)
         at Streamiz.Kafka.Net.Processors.AbstractProcessor`2.Forward(K key, V value)
         at Streamiz.Kafka.Net.Processors.AbstractProcessor`2.Process(ConsumeResult`2 record)
         at Streamiz.Kafka.Net.Processors.StreamTask.Process()
         at Streamiz.Kafka.Net.Processors.Internal.TaskManager.Process(Int64 now)
         at Streamiz.Kafka.Net.Crosscutting.ActionHelper.MeasureLatency(Action action)
         at Streamiz.Kafka.Net.Processors.StreamThread.Run()

How to reproduce

Create materializer thusly:

        var materializer
            = InMemoryWindows
                .As<TGroupedKey, TValueResult>(tableName)
                .WithKeySerdes((ISerDes<TGroupedKey>)SerDesHelper.GetSerDes<TGroupedKey>(KafkaConfiguration))
                .WithValueSerdes((ISerDes<TValueResult>)SerDesHelper.GetSerDes<TValueResult>(KafkaConfiguration))
                .WithRetention(TimeSpan.FromMilliseconds(_windowSizeMs));

Topology is created like so:

        builder.Stream<TKey, TValue>(TopicName).GroupBy(Grouper).Aggregate(Initializer, Aggregator, materializer)).Build();

Access store thusly:

        _windowStore ??= TableStream.Store(
            StoreQueryParameters.FromNameAndType(Name, QueryableStoreTypes.WindowStore<TGroupedKey, TValueResult>()));
        _windowStore
                .Fetch(
                    key,
                    now.Subtract(TimeSpan.FromMilliseconds(_windowSizeMs + _windowAdvanceMs)),
                    now.Add(TimeSpan.FromMilliseconds(_windowSizeMs)))
                .ToList();

-- OR --

        _windowStore
                .FetchAll(
                    now.Subtract(TimeSpan.FromMilliseconds(_windowSizeMs + _windowAdvanceMs)),
                    now.Add(TimeSpan.FromMilliseconds(_windowSizeMs)))
                .ToList();

The issue does not occur every time, but only with large volumes of data. I suspect the issue is due to the Streamiz framework calling Fetch() at the same time as our code.

Checklist

Please provide the following information:

• [ ] A complete (i.e. we can run it), minimal program demonstrating the problem. No need to supply a project file.
• [ ] A code snippet with your topology builder (ex: builder.Stream<string, string>("topic").to("an-another-topic");)
• [ ] Streamiz.Kafka.Net nuget version.
• [ ] Apache Kafka version.
• [ ] Client configuration.
• [ ] Operating system.
• [ ] Provide logs (with in debug mode (log4net and StreamConfig.Debug) as necessary in configuration).
• [ ] Critical issue.

[LGouellec]

Hey @AntonyLittle,

Thank you for your issue and interest of Streamiz. I will try to reproduce the issue and fix it as soon as possible.

Quick question : If you have a large amount of data, why not choosing RocksDb as a persistent layer of storage ?

Best regards,

[LGouellec]

@AntonyLittle ,

Are you able to test with this specific branch ? https://github.com/LGouellec/streamiz/tree/fix/concurrent-issue-window

Thanks,

[AntonyLittle]

Certainly! We've had some success with adding mutex locking also, but we are still testing it. Hopeful that your fix will do the job :)

[AntonyLittle]

@LGouellec Appologies for the delay, we've had some urgent issues. We should be able to report back on the status of your fix today or tomorrow.

[LGouellec]

@AntonyLittle No worries at all ;)

[AntonyLittle]

Can confirm that your changes have fixed the issue. Thank you! Are you able to provide a nuget release?

[LGouellec]

Awesome @AntonyLittle ! This fix will be part the 1.7.0 release, coming soon.

[AntonyLittle]

Awesome @AntonyLittle ! This fix will be part the 1.7.0 release, coming soon.

Do you know roughly how soon "soon" might be? Is there any chance of you doing a 1.6.1 release with just the fix added?

[LGouellec]

@AntonyLittle Will be before end of November, I'll publish a 1.7.0-RC1 which will contains this fix in 1 week.