When using the Streams API to make servers (i.e. using loop.start_server), the StreamWriter.drain() coroutine can short-circuit into an infinite loop and stop bubbling up any exceptions to the caller. This happens, for example, when a client disconnects early: http://stackoverflow.com/questions/32175217/how-to-detect-write-failure-in-asyncio/

A simple solution for the user is to just do an empty yield before every drain() call, but perhaps this fix should be moved upstream.