I've noticed this test failing intermittently on Travis and more consistently on the master branch with Docker. It fails with a ConcurrentModificationException (haven't been able to easily get the entire stack trace thanks to the docker issue and i've not caught it happening in the last few days on Travis). Adding this line of code seems to have made the test pass more consistently:
which would indicate something perhaps fishy with how hosts are being marked dead and iterated. Would be nice to get rid of that little hack.
cc/ Divij Vaidya