[
https://issues.apache.org/jira/browse/FLINK-19427?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17208634#comment-17208634
]
Kezhu Wang commented on FLINK-19427:
------------------------------------
I could image the execution steps for this failure:
1. After {{fetcher.runOnce()}}, the queue is available for pop.
2. Then {{QueueDrainerThread.go}} pop the last element from the queue and move
the queue to un-available.
3. {{assertTrue(queue.getAvailabilityFuture().isDone())}} fails.
The assertion in {{SplitFetcherTest.testNotifiesWhenGoingIdleConcurrent}} is
similar to {{SourceReaderBase.finishedOrAvailableLater}}, they are all some
kind of end-of-input checking. {{FutureCompletingBlockingQueue}} has no builtin
end-of-input checking, {{SourceReaderBase}} uses external knowledge to check
end-of-input. After reaching end-of-input,
{{SourceReaderBase.elementsQueue.getAvailabilityFuture().isDone()}} is
indeterministic. It could be {{false}} due to last poll, or {{true}} due to no
atomic concurrent {{SplitFetcher.checkAndSetIdle()}}.
I think we can pass this test by run {{SplitFetcher.runOnce}} in separate
thread and merge {{QueueDrainerThread}} to main testing method, this is
actually the way {{SourceReaderBase}} taken. This way, after
{{SplitFetcher.runOnce}} joined, we can safely assert {{SplitFetcher.isIdle()}}
and {{FutureCompletingBlockingQueue.getAvailabilityFuture().isDone()}}. But
then, this test looks almost same to its no concurrent counterpart.
Seems that we are testing wrong assumption, may be we can just delete this test
?
[~roman_khachatryan] [~dian.fu] [~sewen] [~becket_qin] Any thoughts ?
> SplitFetcherTest.testNotifiesWhenGoingIdleConcurrent is instable
> ----------------------------------------------------------------
>
> Key: FLINK-19427
> URL: https://issues.apache.org/jira/browse/FLINK-19427
> Project: Flink
> Issue Type: Bug
> Components: Connectors / Common
> Affects Versions: 1.12.0
> Reporter: Dian Fu
> Priority: Major
> Labels: test-stability
>
> https://dev.azure.com/apache-flink/apache-flink/_build/results?buildId=6983&view=logs&j=8fd975ef-f478-511d-4997-6f15fe8a1fd3&t=ac0fa443-5d45-5a6b-3597-0310ecc1d2ab
> {code}
> 2020-09-26T21:27:46.6223579Z [ERROR]
> testNotifiesWhenGoingIdleConcurrent(org.apache.flink.connector.base.source.reader.fetcher.SplitFetcherTest)
> Time elapsed: 0.602 s <<< FAILURE!
> 2020-09-26T21:27:46.6224448Z java.lang.AssertionError
> 2020-09-26T21:27:46.6224804Z at org.junit.Assert.fail(Assert.java:86)
> 2020-09-26T21:27:46.6225136Z at org.junit.Assert.assertTrue(Assert.java:41)
> 2020-09-26T21:27:46.6225498Z at org.junit.Assert.assertTrue(Assert.java:52)
> 2020-09-26T21:27:46.6225984Z at
> org.apache.flink.connector.base.source.reader.fetcher.SplitFetcherTest.testNotifiesWhenGoingIdleConcurrent(SplitFetcherTest.java:129)
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
