There's the following code in pool.py, line 494 and following:

        debug('joining task handler')
        task_handler.join()

        debug('joining result handler')
        task_handler.join()

It seems the last line should read `result_handler.join()` instead.

Additionally, when _terminate() is called, it seems the worker_handler could still run while other threads shut down existing workers, meaning it could start new workers (in _repopulate_pool()) in parallel.

So perhaps the worker_handler should be joined before anything else in _terminate(). It would incur a small latency, though (because of the sleep() call there).