废话不多说, 直接上代码

__auth__ = "aleimu"
__doc__ = "学习 tornado6.0+ 版本与 python3.7+"

import time
import asyncio
import tornado.gen
import tornado.web
import tornado.ioloop
import tornado.httpserver  # tornado 的 HTTP 服务器实现
from tornado.options import define, options
from tornado.httpclient import HTTPClient, AsyncHTTPClient
from requests import get

settings = {'debug': True}
url = "http://127.0.0.1:5000/"  # 这是另个服务, 请求 5s 后返回结果


# RuntimeError: Cannot run the event loop while another loop is running
# 解释:HTTPClient 内部写 loop.run_xxx，因为那是启动 event loop 的命令，通常只再最最最外面用一次，之后的代码都应假设 loop 已经在运转了。def synchronous_fetch(url):
    print("synchronous_fetch")
    try:
        http_client = HTTPClient()
        time.sleep(5)
        response = http_client.fetch(url)
        print(response.body)
    except Exception as e:
        print("Error:" + str(e))
        return str(e)
    http_client.close()
    return response.body


# 替代 synchronous_fetch 的同步请求, 没有内置 loop.run_xxx
def synchronous_get(url):
    response = get(url)
    time.sleep(5)
    print("synchronous_fetch")
    return response.text


# 简单的模拟异步操作, 这里之后应该替换成各种异步库的函数
async def sleep():
    print("start sleep")
    await asyncio.sleep(5)
    print("end sleep")


# 异步请求
async def asynchronous_fetch(url):
    http_client = AsyncHTTPClient()
    response = await http_client.fetch(url)
    print("asynchronous_fetch")
    return response.body


# 测试
class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("Hello, world:%s" % self.request.request_time())
        self.finish()
        print("not finish!")
        return


# 同步阻塞
class synchronous_fetcher(tornado.web.RequestHandler):
    def get(self):
        self.write("%s,%s" % (synchronous_fetch(url), self.request.request_time()))


# 同步阻塞
class synchronous_geter(tornado.web.RequestHandler):
    def get(self):
        self.write("%s,%s" % (synchronous_get(url), self.request.request_time()))


# 异步阻塞, 我以为 curl "127.0.0.1:8888/1" 总耗时希望为 5s, 可是是 25s, 看来异步没搞好, 以下的函数都是基于此改进的
class asynchronous_fetcher_1(tornado.web.RequestHandler):
    async def get(self):
        body = await asynchronous_fetch(url)
        for i in range(3):
            print("skip %s" % i)
            await tornado.gen.sleep(5)
        time.sleep(5)
        print("end request")
        self.write("%s,%s" % (body, self.request.request_time()))

# curl "127.0.0.1:8888/1"
# b'{\n"data":"123"\n}\n',25.026000022888184


# 异步阻塞, 效果同上, 这里只是证明 tornado.gen.sleep(5) 和 asyncio.sleep(5) 效果一致
class asynchronous_fetcher_2(tornado.web.RequestHandler):
    async def get(self):
        body = await asynchronous_fetch(url)  # 关注协程完成后返回的结果
        for i in range(3):
            print("skip %s" % i)
            await sleep()
        time.sleep(5)
        print("end request")
        self.write("%s,%s" % (body, self.request.request_time()))

# curl "127.0.0.1:8888/2"
# b'{\n"data":"123"\n}\n',25.039999961853027


# 异步非阻塞 - 将部分异步操作放入组中, 实现 loop 管理
class asynchronous_fetcher_3(tornado.web.RequestHandler):
    async def get(self):
        body = await asynchronous_fetch(url)
        await asyncio.wait([sleep() for i in range(3)])
        print("end request")
        self.write("%s,%s" % (body, self.request.request_time()))

# curl "127.0.0.1:8888/3"
# b'{\n"data":"123"\n}\n',10.001000165939331

# 异步非阻塞 - 将所有异步操作放入组中, 实现 loop 管理
class asynchronous_fetcher_4(tornado.web.RequestHandler):
    async def get(self):
        task_list = [sleep() for i in range(3)]
        task_list.append(asynchronous_fetch(url))
        body = await asyncio.wait(task_list)  # 将所有异步操作的结果返回, 但是是无序的, 要是需要返回结果的话解析起来比较麻烦
        print("end request:", body)
        # print(type(body), len(body),type(body[0]),len(body[0]),type(body[0]))
        self.write("%s,%s" % ([x.result() for x in body[0] if x.result() is not None][0],
                              self.request.request_time()))
# curl "127.0.0.1:8888/4"
# b'{\n"data":"123"\n}\n',5.006999969482422

def make_app():
    return tornado.web.Application([(r"/", MainHandler),
        (r"/1", asynchronous_fetcher_1),
        (r"/2", asynchronous_fetcher_2),
        (r"/3", asynchronous_fetcher_3),
        (r"/4", asynchronous_fetcher_4),
        (r"/5", synchronous_fetcher),
        (r"/6", synchronous_geter),

    ], **settings)


if __name__ == "__main__":
    print("server start!")
    app = make_app()
    server = tornado.httpserver.HTTPServer(app)
    server.bind(8888)
    server.start(1)  # forks one process per cpu,windows 上无法 fork, 这里默认为 1
    tornado.ioloop.IOLoop.current().start()

总结

1.Tornado 使用单线程事件循环, 写的不好, 会阻塞的非常严重, 比如 synchronous_geter
2.flask+celery 可以完成常见的异步任务
3.await 语法只能出现在通过 async 修饰的函数中
4. 可以看到 tornado.gen.coroutine, tornado.concurrent.run_on_executor,tornado.web.asynchronous,tornado.gen.coroutine 等这些装饰器都不在经常使用了, 都由 async 和 await 代替

参考文档:

https://zhuanlan.zhihu.com/p/27258289   # Python async/await 入门指南
http://www.tornadoweb.org/en/stable/guide/intro.html    # 这个官网
https://www.osgeo.cn/tornado/guide/intro.html   #Tornado 1.0 - Tornado 6.0 的更新说明, 以及 6.0 版本的中文文档, 适合英语不好的人阅读
https://www.osgeo.cn/tornado/releases/v5.0.0.html#  在 Python 3 上，IOLoop 总是包装 asyncio 事件循环。

On Python 3, IOLoop is always a wrapper around the asyncio event loop.
这是我重新复习 tornado 的原因,tornado 放弃了之前自己实现的 tornado.ioloop, 全面拥抱 asyncio 的 event_loop. 这个改动是非常大的,
而且阅读 tornado 的源码可以发现其中大部分函数都支持了类型检验, 和返回值提示, 值得阅读.