@jonasschnelli The first commit should fix the asyncore issue you encountered. I investigated what happens inside asyncore when it uses select instead of poll, and the bug is a subtle race condition: it appears to me that it's not safe to call close() while asyncore.loop() is running when using select(). Merely increasing the timeout would only reduce the likelihood of the bug triggering, not eliminate it, so instead I reworked the disconnect logic to happen outside of asyncore.loop() to eliminate that bug altogether.
The second commit takes invalidblockrequest.py out of travis until I am able to offer up a proper fix to a different race condition that could cause that test to spuriously fail -- that should be soon, but I worry that this could break travis for people in the meantime. Thanks again @jonasschnelli for reporting this in #5981.
