tornado 启动WSGI应用（Flask）使用多线程将同步代码变成异步

1、tornado是单线程的，同时WSGI应用又是同步的，如果我们使用Tornado启动WSGI应用，理论上每次只能处理一个请求都是，任何一个请求有阻塞，都会导致tornado的整个IOLOOP阻塞。如下所示，我们同时发出两个GET请求向http://127.0.0.1:5000/

会发现第一个发出的请求会在大约5s之后返回，而另一个请求会在10s左右返回，我们可以判断，这两个请求是顺序执行的。

from tornado.wsgi import WSGIContainer
from tornado.httpserver import HTTPServer
from tornado.ioloop import IOLoopfrom flask import Flask
import time



app = Flask(__name__)

@app.route('/')
def index():
    time.sleep(5)
    return 'OK'
if __name__ == '__main__':
    http_server = HTTPServer(WSGIContainer(app))
    http_server.listen(5000)
    IOLoop.instance().start()

2、我们知道，tornado实现异步运行同步函数，我们只能使用线程来运行，如下所示：

import tornado.web
import tornado.ioloop
import time
import tornado

class IndexHandler(tornado.web.RequestHandler):
    """主路由处理类"""
    @tornado.gen.coroutine
    def get(self):
        """对应http的get请求方式"""
        loop = tornado.ioloop.IOLoop.instance()
        yield loop.run_in_executor(None,self.sleep)
        self.write("Hello You!")

    def sleep(self):
        time.sleep(5)
        self.write('sleep OK')


if __name__ == "__main__":
    app = tornado.web.Application([
        (r"/", IndexHandler),
    ])
    app.listen(8000)
    tornado.ioloop.IOLoop.current().start()

3、对于这种（使用tornado运行Flask的情况）情况，我们如何做呢，查看 WSGIContainer 的代码我们发现：

我们只需要重写整个方法，将红圈中的部分变为异步运行即可。

代码如下  

loop.run_in_executor 的第一个参数可以为一个ThreadPoolExecutor对象

from flask import Flask
import time
from tornado.wsgi import WSGIContainer
from tornado.httpserver import HTTPServer
from tornado.ioloop import IOLoop

app = Flask(__name__)


@app.route('/')
def index():
    time.sleep(5)
    return 'OK'


import tornado
from tornado import escape
from tornado import httputil
from typing import List, Tuple, Optional, Callable, Any, Dict
from types import TracebackType


class WSGIContainer_With_Thread(WSGIContainer):
    @tornado.gen.coroutine
    def __call__(self, request):
        data = {}  # type: Dict[str, Any]
        response = []  # type: List[bytes]

        def start_response(
                status: str,
                headers: List[Tuple[str, str]],
                exc_info: Optional[
                    Tuple[
                        "Optional[Type[BaseException]]",
                        Optional[BaseException],
                        Optional[TracebackType],
                    ]
                ] = None,
        ) -> Callable[[bytes], Any]:
            data["status"] = status
            data["headers"] = headers
            return response.append

        loop = tornado.ioloop.IOLoop.instance()
        app_response = yield loop.run_in_executor(None, self.wsgi_application, WSGIContainer.environ(request),
                                                  start_response)
        # app_response = self.wsgi_application(
        #     WSGIContainer.environ(request), start_response
        # )
        try:
            response.extend(app_response)
            body = b"".join(response)
        finally:
            if hasattr(app_response, "close"):
                app_response.close()  # type: ignore
        if not data:
            raise Exception("WSGI app did not call start_response")

        status_code_str, reason = data["status"].split(" ", 1)
        status_code = int(status_code_str)
        headers = data["headers"]  # type: List[Tuple[str, str]]
        header_set = set(k.lower() for (k, v) in headers)
        body = escape.utf8(body)
        if status_code != 304:
            if "content-length" not in header_set:
                headers.append(("Content-Length", str(len(body))))
            if "content-type" not in header_set:
                headers.append(("Content-Type", "text/html; charset=UTF-8"))
        if "server" not in header_set:
            headers.append(("Server", "TornadoServer/%s" % tornado.version))

        start_line = httputil.ResponseStartLine("HTTP/1.1", status_code, reason)
        header_obj = httputil.HTTPHeaders()
        for key, value in headers:
            header_obj.add(key, value)
        assert request.connection is not None
        request.connection.write_headers(start_line, header_obj, chunk=body)
        request.connection.finish()
        self._log(status_code, request)


if __name__ == '__main__':
    http_server = HTTPServer(WSGIContainer_With_Thread(app))
    http_server.listen(5000)
    IOLoop.instance().start()

注意：

　　1 、这种方法实际上并没有提高性能，说到底还是使用多线程来运行的，所以推荐如果使用tornado还是和tornado的web框架联合起来写出真正的异步代码，这样才会达到tornado异步IO的高性能目的。