TestSynchronousMeasurementConsumer.test_collect_deadline failures in CI

[xrmx]

I've started seeing failures in this test today in CI (pypy / windows), since this is asserting a mock I guess there's something wrong in the test.

__________ TestSynchronousMeasurementConsumer.test_collect_deadline ___________

self = <test_measurement_consumer.TestSynchronousMeasurementConsumer testMethod=test_collect_deadline>
mock_callback_options = <MagicMock name='CallbackOptions' id='1661901429728'>
MockMetricReaderStorage = <MagicMock name='MetricReaderStorage' id='1661901430400'>

    @patch(
        "opentelemetry.sdk.metrics._internal."
        "measurement_consumer.CallbackOptions"
    )
    def test_collect_deadline(
        self, mock_callback_options, MockMetricReaderStorage
    ):
        reader_mock = Mock()
        reader_storage_mock = Mock()
        MockMetricReaderStorage.return_value = reader_storage_mock
        consumer = SynchronousMeasurementConsumer(
            SdkConfiguration(
                exemplar_filter=Mock(),
                resource=Mock(),
                metric_readers=[reader_mock],
                views=Mock(),
            )
        )
    
        def sleep_1(*args, **kwargs):
            sleep(1)
            return []
    
        consumer.register_asynchronous_instrument(
            Mock(**{"callback.side_effect": sleep_1})
        )
        consumer.register_asynchronous_instrument(
            Mock(**{"callback.side_effect": sleep_1})
        )
    
        consumer.collect(reader_mock)
    
>       callback_options_time_call = mock_callback_options.mock_calls[
            -1
        ].kwargs["timeout_millis"]
E       KeyError: 'timeout_millis'

[aabmass]

def sleep_1(*args, **kwargs):
sleep(1)
return []

Is someone's fixing up this test, maybe we can reduce or modify the sleep here to speed things up

[ravitheja4531-cell]

Hi, I'd like to work on a fix for this if it's still available.

After looking at the source, I think the root cause is a race condition in the test itself. The test calls consumer.collect(reader_mock) with the default timeout_millis=10_000, then registers two callbacks that each sleep(1). The intent is that after callback 1 runs, remaining_time drops below default_timeout_ns (10s), triggering CallbackOptions(timeout_millis=...) before callback 2.

On slow platforms like PyPy/Windows, callback 1 alone can consume enough time that time_ns() >= deadline_ns after it returns, causing MetricsTimeoutError to be raised before the loop ever constructs a CallbackOptions with timeout_millis. This means mock_calls[-1] ends up being the initial no-arg CallbackOptions() call at the top of collect(), and .kwargs['timeout_millis'] raises KeyError.

My proposed fix is to refactor the test to mock time_ns with controlled return values instead of relying on sleep(1), making the timing deterministic across platforms. I'd also consider simplifying the production logic by removing the default_timeout_ns threshold and always constructing CallbackOptions(timeout_millis=remaining/1e6) per iteration.

Happy to submit a PR if this approach sounds good.

[xrmx]

@ravitheja4531-cell Could you please throw your bot to one issue at a time? thanks