为什么 asyncio 的 run_in_executor 会阻塞 tornado 的 get 处理程序？

Question

我想在 tornado 的异步 GET 请求处理程序中运行一个缓慢的阻塞方法（实际上来自第 3 方库）。让方法只是：

def blocking_method(uid):
    print("slow method started: ", uid)
    time.sleep(10)
    print("slow method done: ", uid)
    return "slow method ({}) result".format(uid)

此外，我更喜欢运行在 asyncio 的事件循环中使用 tornado 服务器：

if __name__ == '__main__':
    tornado.platform.asyncio.AsyncIOMainLoop().install()
    loop = asyncio.get_event_loop()
    loop.run_until_complete(make_app())
    loop.run_forever()

我知道 @run_in_executor 装饰器，但它不适合我，因为我使用 asyncio。对于运行异步协程中的阻塞方法，我应该使用 asyncio.get_event_loop() 的 run_in_executor 方法。这是一个如何做的例子，来自 this 答案：

import asyncio

async def main():
    loop = asyncio.get_event_loop()
    executor = concurrent.futures.ThreadPoolExecutor(max_workers=4)
    future1 = loop.run_in_executor(executor, blocking_method, 1)
    future2 = loop.run_in_executor(executor, blocking_method, 2)
    response1 = await future1
    response2 = await future2
    print(response1)
    print(response2)

if __name__ == '__main__':
    loop = asyncio.get_event_loop()
    loop.run_until_complete(main())

它工作得很好，这是之前脚本的输出：

slow method started:  1
slow method started:  2
slow method done:  2
slow method done:  1
slow method (1) result
slow method (2) result

但是如果我在 tornado 的 RequestHandler 的 async def get 方法中使用完全相同的技术：

class AsyncHandler(tornado.web.RequestHandler):

    async def get(self):
        #  simple counter to distinguish requests
        self.application.counter += 1
        in_msg = "Registered request #{}, working...".format(self.application.counter)
        print(in_msg)
        loop = asyncio.get_event_loop()
        future = loop.run_in_executor(self.application.executor,
                                      blocking_method,
                                      self.application.counter)
        result = await future
        out_msg = "Request processed, result: {}".format(result)
        print(out_msg)
        self.write(out_msg)

它阻止处理程序的方法。换句话说，如果我在多个浏览器选项卡中打开 http://localhost:8888/（让它是两个），那么我期望两个并行工作的请求，输出如下：

Registered request #1, working...
slow method started:  1
Registered request #2, working...
slow method started:  2
slow method done:  1
Request processed, result: slow method (1) result
slow method done:  2
Request processed, result: slow method (2) result

但是请求被执行随后:

Registered request #1, working...
slow method started:  1
slow method done:  1
Request processed, result: slow method (1) result
Registered request #2, working...
slow method started:  2
slow method done:  2
Request processed, result: slow method (2) result

所以，我哪里错了？我应该怎么做才能并行执行请求处理程序？

这是描述我的问题的完整脚本：

import asyncio
import concurrent.futures
import time

import tornado.web
import tornado.platform


def blocking_method(uid):
    print("slow method started: ", uid)
    time.sleep(10)
    print("slow method done: ", uid)
    return "slow method ({}) result".format(uid)


class AsyncHandler(tornado.web.RequestHandler):

    async def get(self):
        #  simple counter to distinguish requests
        self.application.counter += 1
        in_msg = "Registered request #{}, working...".format(self.application.counter)
        print(in_msg)
        loop = asyncio.get_event_loop()
        future = loop.run_in_executor(self.application.executor,
                                      blocking_method,
                                      self.application.counter)
        result = await future
        out_msg = "Request processed, result: {}".format(result)
        print(out_msg)
        self.write(out_msg)

async def make_app():
    handlers = [(r"/", AsyncHandler)]
    app = tornado.web.Application(handlers, debug=True)
    app.executor = concurrent.futures.ThreadPoolExecutor(max_workers=4)
    app.counter = 0
    app.listen(8888)

if __name__ == '__main__':
    tornado.platform.asyncio.AsyncIOMainLoop().install()
    loop = asyncio.get_event_loop()
    loop.run_until_complete(make_app())
    loop.run_forever()

Answer 1

浏览器会识别出您正尝试在两个不同的选项卡中加载同一页面并延迟第二个请求，直到第一个请求完成。

http://www.tornadoweb.org/en/latest/faq.html#why-isn-t-this-example-with-time-sleep-running-in-parallel

为您的 url 添加一些内容，使它们独一无二。而不是http://localhost:8888 in both tabs, load http://localhost:8888/?x=1 in one and http://localhost:8888/?x=2在另一个
使用两种不同的浏览器。例如，Firefox 将能够加载 url，即使在 Chrome 选项卡中正在加载相同的 url。

为什么 asyncio 的 run_in_executor 会阻塞 tornado 的 get 处理程序？

Why asyncio's run_in_executor blocks tornado's get handler?

python

tornado

python-asyncio