Flaky opentelemetry-sdk tests on windows / pypy-3.10

[xrmx]

______ TestConcurrentMultiSpanProcessor.test_force_flush_late_by_timeout ______

self = <tests.trace.test_span_processor.TestConcurrentMultiSpanProcessor testMethod=test_force_flush_late_by_timeout>

    def test_force_flush_late_by_timeout(self):
        multi_processor = trace.ConcurrentMultiSpanProcessor(5)
        wait_event = Event()
    
        def delayed_flush(_):
            wait_event.wait()
    
        late_mock = mock.Mock(spec=trace.SpanProcessor)
        late_mock.force_flush = mock.Mock(side_effect=delayed_flush)
        mocks = [mock.Mock(spec=trace.SpanProcessor) for _ in range(0, 4)]
        mocks.insert(0, late_mock)
    
        for mock_processor in mocks:
            multi_processor.add_span_processor(mock_processor)
    
        flushed = multi_processor.force_flush(timeout_millis=25)
        # let the thread executing the late_mock continue
        wait_event.set()
    
        self.assertFalse(flushed)
        for mock_processor in mocks:
>           self.assertEqual(1, mock_processor.force_flush.call_count)
E           AssertionError: 1 != 0

opentelemetry-sdk\tests\trace\test_span_processor.py:486: AssertionError

[saisharan0103]

test_force_flush_late_by_timeout has one processor blocked past a 25ms flush timeout, but the failure shows another processor’s force_flush can still be at call_count == 0 when the assertion runs. For a PR, I would start by inspecting ConcurrentMultiSpanProcessor.force_flush for timeout handling that returns before all submitted flush calls have actually been invoked. The regression test should keep one blocking processor and assert the remaining processors are scheduled/called even when the overall flush result is False.