[Tests] Fix flaky test GracefulExecutorServicesShutdownTest

[lhotari]

Motivation

GracefulExecutorServicesShutdownTest's shouldTerminateWhenFutureIsCancelled remains flaky after #10592 .

Example failure: https://github.com/apache/pulsar/pull/10598/checks?check_run_id=2590214027#step:10:1341

Modifications

• fix race condition in test by adding a CountDownLatch to verify that execution has entered the awaitTermination method before the future is cancelled.
• use CompletableFuture for awaitTerminationInterrupted check.

[linlinnn]

What caused the flakiness was that this condition caused the execution to never get to the Thread.sleep when the thread was already interrupted:

Yes, I agree that.
But after call countdown, main thread can cancel the future, that means the Interruption exception can probably be caught by outer catch.
You can test it locally by add some time-consuming action after countdown.

[eolivelli]

@linlinnn I have merged this patch in order to unblock CI.
We can continue the discussion and find a better fix if think that this is not the best way

[linlinnn]

ok

[lhotari]

But after call countdown, main thread can cancel the future, that means the Interruption exception can probably be caught by outer catch.
You can test it locally by add some time-consuming action after countdown.

I don't see how this could happen. Interrupting a thread doesn't stop execution. It sets the interrupted status and interrupts any blocking methods such as sleep or blocking IO with an InterruptedException. Calling countDown will never be interrupted.
Makes sense?

[linlinnn]

@lhotari
Oh, I make a wrong understand with the outer catch InterruptedException in GracefulExecutorServicesTerminationHandler#awaitTermination since executor.awaitTermination has been mocked.
Sorry for my mistake.

[lhotari]

Oh, I make a wrong understand with the outer catch InterruptedException in GracefulExecutorServicesTerminationHandler#awaitTermination since executor.awaitTermination has been mocked.
Sorry for my mistake.

No worries, the mocking was quite unexpected for executor.awaitTermination.

btw. It seems that this test became very flaky after the switch to JDK11. To get the fix to other PRs, it is either necessary to rebase or close and re-open a PR so that the changes get picked up. When re-running a failed build it won't pick up changes made into master branch. New PR builds or reopened PR builds will pick up changes.