Flaky test: `NIOPOSIXTests.EventLoopTest.testCancelledScheduledTasksDoNotHoldOnToRunClosure()`

[dnadoba]

I have seen it fail in CI and locally a couple of times now. It is also reliably reproducible if run repeatedly after ~500-2000 iterations.

swift-nio/Tests/NIOPosixTests/EventLoopTest.swift

Lines 807 to 831 in 712e207

	func testCancelledScheduledTasksDoNotHoldOnToRunClosure() {
	let group = MultiThreadedEventLoopGroup(numberOfThreads: 1)
	defer {
	XCTAssertNoThrow(try group.syncShutdownGracefully())
	}
	class Thing {}
	weak var weakThing: Thing? = nil
	func make() -> Scheduled<Never> {
	let aThing = Thing()
	weakThing = aThing
	return group.next().scheduleTask(in: .hours(1)) {
	preconditionFailure("this should definitely not run: \(aThing)")
	}
	}
	let scheduled = make()
	scheduled.cancel()
	assert(weakThing == nil, within: .seconds(1))
	XCTAssertThrowsError(try scheduled.futureResult.wait()) { error in
	XCTAssertEqual(EventLoopError.cancelled, error as? EventLoopError)
	}
	}

I have played a bit around with the time and testInterval but this doesn't seem to really help make this test reliable.
The memory debugger shows that an instance to weakThing is still hold by the EventLoop which it should not.

Needs further investigation.

[Lukasa]

I think this is simply a timing issue: eventually the closure will execute, which will drop the ref.