[SPARK-28535] Flaky test: JobCancellationSuite."interruptible iterator of shuffle reader"

• Type: Test
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.3.3, 2.4.3, 3.0.0
• Fix Version/s: 2.3.4, 2.4.4, 3.0.0
• Component/s: Tests
• Assignee: Marcelo Masiero Vanzin
• Reporter: Marcelo Masiero Vanzin
• Created: 26/Jul/19 23:59
• Updated: 01/Sep/19 06:30
• Resolved: 27/Jul/19 18:07

Description

This is the same flakiness as in SPARK-23881, except the fix there didn't really take, at least on our build machines.

org.scalatest.exceptions.TestFailedException: 10000 was not less than 10000
      at org.scalatest.Assertions$class.newAssertionFailedException(Assertions.scala:528)
      at org.scalatest.FunSuite.newAssertionFailedException(FunSuite.scala:1560)
      at org.scalatest.Assertions$AssertionsHelper.macroAssert(Assertions.scala:501)

Since that bug is short on explanations, the issue is that there's a race between the thread posting the "stage completed" event to the listener which unblocks the test, and the thread killing the task in the executor. If the even arrives first, it will unblock task execution, and there's a chance that all elements will actually be processed before the executor has a chance to stop the task.

Issue Links

• relates to: SPARK-23881 Flaky test: JobCancellationSuite."interruptible iterator of shuffle reader"
• links to: GitHub Pull Request #25270