There are tests involving random number generation where the test fails time to time due to stochastic reasons, with relatively small probability (about 10% in the worst case). It would be an improvement to automatically re-run on failure these tests once again.