asyncio：等待来自其他线程的事件

Question

I'm designing an application in Python which should access a machine to perform some (lengthy) tasks. The asyncio module seems to be a good choice for everything that is network-related, but now I need to access the serial port for one specific component. I've implemented kind of an abstraction layer for the actual serial port stuff, but can't figure out how to sensibly integrate this with asyncio.

f = threading.Event()
ch.enqueue_query('2 getnlimit', f)
f.wait()
print(ch.get_query_responses())

Answer 1

By passing in a threading.Event (or anything with a set() method), the caller can perform a blocking wait for the response.

Given the above behavior of your query function, all you need is a thread-safe version of asyncio.Event . It's just 3 lines of code:

import asyncio
class Event_ts(asyncio.Event):
    #TODO: clear() method
    def set(self):
        #FIXME: The _loop attribute is not documented as public api!
        self._loop.call_soon_threadsafe(super().set)

A test for functionality:

def threaded(event):
    import time
    while True:
        event.set()
        time.sleep(1)

async def main():
    import threading
    e = Event_ts()
    threading.Thread(target=threaded, args=(e,)).start()
    while True:
        await e.wait()
        e.clear()
        print('whatever')

asyncio.ensure_future(main())
asyncio.get_event_loop().run_forever()

Answer 2

The simplest way is to do exactly what you suggested - wrap the call to f.wait() in an executor:

@asyncio.coroutine
def do_enqueue():
    f = threading.Event()
    ch.enqueue_query('2 getnlimit', f)
    yield from loop.run_in_executor(None, f.wait)
    print(ch.get_query_responses())

You do incur the overhead of starting up a thread pool (at least for the first call, the pool will stay in memory from that point forward), but any solution that provides an implementation of something like threading.Event() with thread-safe blocking and non-blocking APIs, without relying on any background threads internally, would be quite a bit more work.

Answer 3

The class Event_ts in Huazuo Gao's answer works well in Python up to 3.9 but not in 3.10. It is because in Python 3.10 the private attribute _loop is initially None .

The following code works in Python 3.10 as well as 3.9 and below. (I added the clear() method as well.)

import asyncio
class Event_ts(asyncio.Event):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        if self._loop is None:
            self._loop = asyncio.get_event_loop()

    def set(self):
        self._loop.call_soon_threadsafe(super().set)

    def clear(self):
        self._loop.call_soon_threadsafe(super().clear)