How to properly test this kind of timed concurrent processing service?

Question

I have a service which is used to simulate an IoT hardware component which sends data periodically to a SOAP endpoint:

private static final int DEFAULT_CHUNK_SIZE_IN_BYTES = 100 * 1024;

public void simulate(final int samplingFrequencyMs, final int uploadFrequencyMs) {
    // ...

This method generates a JSON object every time samplingFrequencyMs time passes and calls a SOAP endpoint every time uploadFrequencyMs time passes and chunks the generated data into DEFAULT_CHUNK_SIZE_IN_BYTES size chunks.

So for example if samplingFrequencyMs is 500 and uploadFrequencyMs is 1000 then this method will send 2 generated JSON strings every second to a SOAP endpoint and if the size of the 2 JSON strings is 200 Kbyte then it will send 2 chunks.

There is also a method which stops the simulation:

public void stopSimulation() {
    // ...

I've written a test for this simulator:

@Test
public void shouldSendProperAmountOfChunksWhenInvokingSimulate() throws InterruptedException
{
    // ...

    // when
    underTest.simulate(100, 500);
    Thread.sleep(750);

    underTest.stopSimulation();

    Thread.sleep(500);

    // then
    verify(requestServiceMock, times(EXPECTED_REQUEST_CALL_TIMES)).doRequest(uploadBinary, header);
}

but this test fails 1-2 times out of 100 because the service was not invoked EXPECTED_REQUEST_CALL_TIMES times so it is not stable. How can I test this when I have to verify the times a method was called but there is timing involved? I used 750 because it is between 500 and 1000 and I thought that this should do the trick but I think there must be some testing best practice for this kind of situation.

Answer 1

What you need to do is abstract away the concept of the system time from your CUT. A typical approach is to inject a Clock object that you can control. Your real implementation will delegate to the system clock, but within the test you have exact control over what's going on.