Most test frameworks I have seen that support non-determinism in some way print the random seed at the start of the run, and let you specify the seed when you run the tests yourself. It's a good practice for precisely the reasons you wrote.