test_get_normally_distributed_number is flaky

[devos50]

Since the test_get_normally_distributed_number test depends on randomness, it can be flaky and fail when 'unlucky' numbers are drawn that do not pass the Shapiro test. I saw this test failing in one of my runs:

2022-10-18T07:43:01.6956570Z =================================== FAILURES ===================================
2022-10-18T07:43:01.6957363Z _____________________ test_get_normally_distributed_number _____________________
2022-10-18T07:43:01.6958298Z 
2022-10-18T07:43:01.6958890Z     def test_get_normally_distributed_number():
2022-10-18T07:43:01.6959807Z         """
2022-10-18T07:43:01.6960279Z         To test if the random number is from normal distribution, we do Shapiro test and check that
2022-10-18T07:43:01.6961011Z         p-value is higher than 0.05.
2022-10-18T07:43:01.6961267Z         """
2022-10-18T07:43:01.6961633Z         random_numbers = [get_normally_distributed_number_with_zero_mean() for _ in range(100)]
2022-10-18T07:43:01.6962039Z         shapiro_test = shapiro(random_numbers)
2022-10-18T07:43:01.6962386Z >       assert shapiro_test.pvalue > 0.05
2022-10-18T07:43:01.6962673Z E       assert 0.005304783117026091 > 0.05
2022-10-18T07:43:01.6963047Z E        +  where 0.005304783117026091 = ShapiroResult(statistic=0.9616765379905701, pvalue=0.005304783117026091).pvalue
2022-10-18T07:43:01.6963294Z 
2022-10-18T07:43:01.6963496Z src/tribler/core/utilities/tests/test_utilities.py:270: AssertionError
2022-10-18T07:43:01.6964125Z =============================== warnings summary ===============================

I also managed to make this test fail locally, by running this particular test hundreds of times:

seq 1000 | xargs -Iz pytest tribler/core/utilities/tests/test_utilities.py::test_get_normally_distributed_number

================================================================================================================================== FAILURES ==================================================================================================================================
____________________________________________________________________________________________________________________ test_get_normally_distributed_number ____________________________________________________________________________________________________________________

    def test_get_normally_distributed_number():
        """
        To test if the random number is from normal distribution, we do Shapiro test and check that
        p-value is higher than 0.05.
        """
        random_numbers = [get_normally_distributed_number_with_zero_mean() for _ in range(100)]
        shapiro_test = shapiro(random_numbers)
>       assert shapiro_test.pvalue > 0.05
E       assert 0.03620334342122078 > 0.05
E        +  where 0.03620334342122078 = ShapiroResult(statistic=0.972777783870697, pvalue=0.03620334342122078).pvalue

tribler/core/utilities/tests/test_utilities.py:270: AssertionError

I'm not in favour of having such a test. I'm also questioning its value since we're essentially testing whether the random.normalvariate method gives correct numbers.

[drew2a]

@xoriole seems to be a code owner of these tests.

I confirm that test_get_normally_distributed_number probably is the most unstable test at the moment.

[drew2a]

Again: https://github.com/Tribler/tribler/actions/runs/3297913958/jobs/5439341261 here #7101

[xoriole]

#7103 resolves this