tornado 源码分析 之 异步io的实现方式

摘要：前言本文将尝试详细的带大家一步步走完一个异步操作从而了解是如何实现异步的其实本文是对上一篇文的实践和复习主旨在于关注异步的实现所以会忽略掉代码中的一些异常处理文字较多凑合下吧接下来只会贴出部分源码帮助理解希望有耐心的同学打开源码一起跟踪一遍

本文将尝试详细的带大家一步步走完一个异步操作,从而了解tornado是如何实现异步io的.
其实本文是对[上一篇文][1]的实践和复习
主旨在于关注异步io的实现,所以会忽略掉代码中的一些异常处理.文字较多,凑合下吧

接下来只会贴出部分源码,帮助理解,希望有耐心的同学打开tornado源码,一起跟踪一遍吧.

 AsyncHTTPClient 继承 Configurable ,从__new__重看出是单例模式.  
根据 Configurable 的__new__和 AsyncHTTPClient 的 configurable_base 和 configurable_default 得知,
实例化后一定是 SimpleAsyncHTTPClient 的实例

def fetch(self, request, callback=None, raise_error=True, **kwargs):

        if self._closed:
            raise RuntimeError("fetch() called on closed AsyncHTTPClient")
        if not isinstance(request, HTTPRequest):
            request = HTTPRequest(url=request, **kwargs)
        # We may modify this (to add Host, Accept-Encoding, etc),
        # so make sure we don"t modify the caller"s object.  This is also
        # where normal dicts get converted to HTTPHeaders objects.
        request.headers = httputil.HTTPHeaders(request.headers)
        request = _RequestProxy(request, self.defaults)
        future = TracebackFuture()
        if callback is not None:
            callback = stack_context.wrap(callback)

            def handle_future(future):
                exc = future.exception()
                if isinstance(exc, HTTPError) and exc.response is not None:
                    response = exc.response
                elif exc is not None:
                    response = HTTPResponse(
                        request, 599, error=exc,
                        request_time=time.time() - request.start_time)
                else:
                    response = future.result()
                self.io_loop.add_callback(callback, response)
            future.add_done_callback(handle_future)
        ##fetch_impl带上handle_response,重点
        def handle_response(response):
            if raise_error and response.error:
                future.set_exception(response.error)
            else:
                future.set_result(response)
        self.fetch_impl(request, handle_response)
        return future

 fetch 中调用 fetch_impl,fetch_impl 中其中一个参数是 callback ,而代码中的 callback 包含了 future 的 set_result ,
所以,当 callback 被调用时,外部的 yield 操作将被激活,程序会在 ioloop 中调用此 callback ,然后回到原函数的 yield 处,
并且原函数返回此次 qeust 的 future 对象,以便在函数外部增加别的 callback 

def _connection_class(self):
        return _HTTPConnection

def _handle_request(self, request, release_callback, final_callback):
        self._connection_class()(
            self.io_loop, self, request, release_callback,
            final_callback, self.max_buffer_size, self.tcp_client,
            self.max_header_size, self.max_body_size)

在 return 之前,继续查看 fetch_impl 内部是如何处理,根据推测,他一定是将继续处理网络请求,
肯定会将网络请求交由 ioloop 的 epoll 部分处理,设定好处理的 hanlder 再返回
 future.set_result ,接下来继续分析 fetch_impl 内部是如果设置网络请求的.
 fetch_impl 的实现代码中查看,实例化中创建了 tcpclient 对象,这个肯定是关键

根据之前的分析 SimpleAsyncHTTPClient 是单例模式,那他怎么处理各种 http 请求呢?
查看代码得知,他将请求的 request 和 callback 存储在 self.queue 中,
每次 fetch_impl 的时候,一个个 pop 出来处理就好了,这样就能处理n个请求了

一步步跟踪到 _handle_request ,发现最后到了 _HTTPConnection 的实例化中了.
实例化的参数有之前那个包含 future 的 callback .
这样就可以保证 yield 操作可以回到原处了.好了,继续走

class _HTTPConnection(httputil.HTTPMessageDelegate):
    _SUPPORTED_METHODS = set(["GET", "HEAD", "POST", "PUT", "DELETE", "PATCH", "OPTIONS"])

    def __init__(self, io_loop, client, request, release_callback,
                 final_callback, max_buffer_size, tcp_client,
                 max_header_size, max_body_size):
        self.start_time = io_loop.time()
        self.io_loop = io_loop
        self.client = client
        self.request = request
        self.release_callback = release_callback
        self.final_callback = final_callback
        self.max_buffer_size = max_buffer_size
        self.tcp_client = tcp_client
        self.max_header_size = max_header_size
        self.max_body_size = max_body_size
        self.code = None
        self.headers = None
        self.chunks = []
        self._decompressor = None
        # Timeout handle returned by IOLoop.add_timeout
        self._timeout = None
        self._sockaddr = None
        with stack_context.ExceptionStackContext(self._handle_exception):
            self.parsed = urlparse.urlsplit(_unicode(self.request.url))
            if self.parsed.scheme not in ("http", "https"):
                raise ValueError("Unsupported url scheme: %s" %
                                 self.request.url)
            # urlsplit results have hostname and port results, but they
            # didn"t support ipv6 literals until python 2.7.
            netloc = self.parsed.netloc
            if "@" in netloc:
                userpass, _, netloc = netloc.rpartition("@")
            host, port = httputil.split_host_and_port(netloc)
            if port is None:
                port = 443 if self.parsed.scheme == "https" else 80
            if re.match(r"^[.*]$", host):
                # raw ipv6 addresses in urls are enclosed in brackets
                host = host[1:-1]
            self.parsed_hostname = host  # save final host for _on_connect

            if request.allow_ipv6 is False:
                af = socket.AF_INET
            else:
                af = socket.AF_UNSPEC

            ssl_options = self._get_ssl_options(self.parsed.scheme)

            timeout = min(self.request.connect_timeout, self.request.request_timeout)
            if timeout:
                self._timeout = self.io_loop.add_timeout(
                    self.start_time + timeout,
                    stack_context.wrap(self._on_timeout))
            self.tcp_client.connect(host, port, af=af,
                                    ssl_options=ssl_options,
                                    max_buffer_size=self.max_buffer_size,
                                    callback=self._on_connect)

 _HTTPConnection 的实例化中有一堆成员变量,有点晕,
先不管这么多,关注我们的 callback ,和 tcpclient .

一行行往下看,是 host 和 port 的初始化操作 ,http 和 https 是不一样的嘛,当然得处理一下,

终于到了最后,是 tcpclient.connect ,从 connect 的参数中看到 callback=self._on_connect ,
应该是个重要的方法,出去那些字符串处理,发现 self.connection.write_headers(start_line ,  self.request.headers ) ,
这应该是发送 http 头的操作吧,是网络请求,所以这是处理 connect 这个 url 后,发送 http 头的操作.

还是回头看看是如何 connect 的吧,因为这是异步的关键,搞懂了这个,那剩下来的也是同出一则

转到 tcpclient 的代码去看他的实例化和 connect 操作,看来剩下的路还很长呢
 TCPClient 实例化的代码很短,有个 resolver 对象,先不管

    @gen.coroutine
    def connect(self, host, port, af=socket.AF_UNSPEC, ssl_options=None,
                max_buffer_size=None):
        """Connect to the given host and port.

        Asynchronously returns an `.IOStream` (or `.SSLIOStream` if
        ``ssl_options`` is not None).
        """
        addrinfo = yield self.resolver.resolve(host, port, af)
        connector = _Connector(
            addrinfo, self.io_loop,
            functools.partial(self._create_stream, max_buffer_size))
        af, addr, stream = yield connector.start()
        # TODO: For better performance we could cache the (af, addr)
        # information here and re-use it on subsequent connections to
        # the same host. (http://tools.ietf.org/html/rfc6555#section-4.2)
        if ssl_options is not None:
            stream = yield stream.start_tls(False, ssl_options=ssl_options,
                                            server_hostname=host)
        raise gen.Return(stream)

去到 connect 方法里,发现 coroutine 装饰器,并且调用时设置了 callback=self._on_connect ,
所以当这个 coroutine 的 future 被解决时,会调用 self._on_connect ,
你也可以看到 _on_connect 的参数是 stream ,就是 gen.Return(stream )传过去的. 
因为 gen.coroutine 实现时的代码中,
 send 了 value 后,代码继续走,走到 gen.Return (其实这是个 exception ,
就会走到 gen.coroutine 里的 set_result 了.)

第一个 yield  右边是 self.resolver.resolve ,左边是 addrinfo ,是地址信息,
这个异步操作处理的便是解析 url 的地址信息.此处 tornado 默认使用了阻塞的实现,暂时先不看,
以后在新的篇幅补充,主要内容是 run_on_executor 装饰器的内容,
此处其实是同步返回的,因为默认用的是 BlockingResolver 的代码,直接看下一个 yield 

    def __init__(self, addrinfo, io_loop, connect):
        self.io_loop = io_loop
        self.connect = connect

        self.future = Future()
        self.timeout = None
        self.last_error = None
        self.remaining = len(addrinfo)
        self.primary_addrs, self.secondary_addrs = self.split(addrinfo)

 _Connector 实例化,参数有一个 callback ,是本类的 _create_stream ,
并把 self.connect 设置成传过来的 callback 
所以 self.connect 就是 TCPClient._create_stream 了,
成员变量有个 future 实例,我们需要全程高度关注 future 和 callback .

实例化后调用了 start 方法 ,start 内部,调用 try_connect,set_timout ,
根据函数名得知,是 connect 操作和设置超时的操作.最后返回实例化时创建的 future .

def start(self, timeout=_INITIAL_CONNECT_TIMEOUT):
        self.try_connect(iter(self.primary_addrs))
        self.set_timout(timeout)
        return self.future

    def try_connect(self, addrs):
        try:
            af, addr = next(addrs)
        except StopIteration:
            # We"ve reached the end of our queue, but the other queue
            # might still be working.  Send a final error on the future
            # only when both queues are finished.
            if self.remaining == 0 and not self.future.done():
                self.future.set_exception(self.last_error or
                                          IOError("connection failed"))
            return
        future = self.connect(af, addr)
        future.add_done_callback(functools.partial(self.on_connect_done,
                                                   addrs, af, addr))

 future  =  self.connect(af ,  addr ),执行了 TCPClient._create_stream 方法,
返回 future ,并且设置 future 的 callback=on_connect_done 

    def _create_stream(self, max_buffer_size, af, addr):
        # Always connect in plaintext; we"ll convert to ssl if necessary
        # after one connection has completed.
        stream = IOStream(socket.socket(af),
                          io_loop=self.io_loop,
                          max_buffer_size=max_buffer_size)
        return stream.connect(addr)

实例化 IOStream ,执行并返回 stream.connect,stream.connect 返回的 future 便是 try_connect 中的 future ,
所以,进去看看 stream.connect 内部是怎么”解决”这个 future 的.

def connect(self, address, callback=None, server_hostname=None):
        self._connecting = True
        if callback is not None:
            self._connect_callback = stack_context.wrap(callback)
            future = None
        else:
            future = self._connect_future = TracebackFuture()
        try:
            self.socket.connect(address)
        except socket.error as e:

            if (errno_from_exception(e) not in _ERRNO_INPROGRESS and
                    errno_from_exception(e) not in _ERRNO_WOULDBLOCK):
                if future is None:
                    gen_log.warning("Connect error on fd %s: %s",
                                    self.socket.fileno(), e)
                self.close(exc_info=True)
                return future
        self._add_io_state(self.io_loop.WRITE)
        return future

 self._connecting  =  True  设置此实例正在连接中,连接完毕设置成 false 
如果没有 callback 传入,生成 future 对象, 刚才返回的 future 记录在这个实例的成员变量 self._connect_future 中.
然后执行 socket 的 connect 操作,因为 socket 设置成非阻塞,
所以此处会立即返回,不会阻塞,当连接成功时,缓冲区可写,失败时缓冲区可读可写.这是基础知识,详情百度.
然后调用 self._add_io_state ,返回 future 

    def _add_io_state(self, state):

        if self.closed():
            # connection has been closed, so there can be no future events
            return
        if self._state is None:
            self._state = ioloop.IOLoop.ERROR | state
            with stack_context.NullContext():
                self.io_loop.add_handler(
                    self.fileno(), self._handle_events, self._state)
        elif not self._state & state:
            self._state = self._state | state
            self.io_loop.update_handler(self.fileno(), self._state)

终于到了这一步,要用 epoll 了!!!根据实例化的代码得知 self._state=None ,
会走 self.io_loop.add_handler 这步,根据我之前发的[文章][2],会将当前的 fd ,当前实例的 _handle_events ,和写,错误操作注册到 epoll 中
接着!!!!!终于走完了这个 yield 的流程了!!!!!!

 请一定弄清 future 是怎么传递的,每个 future 管理的 callback 是什么操作.
  _HTTPConnection  中  tcpclient.connect 一个 future   ,callback=self._on_connect  .
 他将在 raise   gen.Return(stream )时被添加到 ioloop 执行. 
  tcpclient.connect 内部的  connector.start 一个 future , 
  callback 是 on_connect_done ,他将在 poll 检测到 write 事件时,被添加到 ioloop 执行

    def start(self):
        if self._running:
            raise RuntimeError("IOLoop is already running")
        self._setup_logging()
        if self._stopped:
            self._stopped = False
            return
        old_current = getattr(IOLoop._current, "instance", None)
        IOLoop._current.instance = self
        self._thread_ident = thread.get_ident()
        self._running = True

        old_wakeup_fd = None
        if hasattr(signal, "set_wakeup_fd") and os.name == "posix":

            try:
                old_wakeup_fd = signal.set_wakeup_fd(self._waker.write_fileno())
                if old_wakeup_fd != -1:
                    signal.set_wakeup_fd(old_wakeup_fd)
                    old_wakeup_fd = None
            except ValueError:
                old_wakeup_fd = None

        try:
            while True:
                with self._callback_lock:
                    callbacks = self._callbacks
                    self._callbacks = []

                due_timeouts = []

                if self._timeouts:
                    now = self.time()
                    while self._timeouts:
                        if self._timeouts[0].callback is None:

                            heapq.heappop(self._timeouts)
                            self._cancellations -= 1
                        elif self._timeouts[0].deadline <= now:
                            due_timeouts.append(heapq.heappop(self._timeouts))
                        else:
                            break
                    if (self._cancellations > 512
                            and self._cancellations > (len(self._timeouts) >> 1)):

                        self._cancellations = 0
                        self._timeouts = [x for x in self._timeouts
                                          if x.callback is not None]
                        heapq.heapify(self._timeouts)

                for callback in callbacks:
                    self._run_callback(callback)
                for timeout in due_timeouts:
                    if timeout.callback is not None:
                        self._run_callback(timeout.callback)
                callbacks = callback = due_timeouts = timeout = None

                if self._callbacks:
                    poll_timeout = 0.0
                elif self._timeouts:

                    poll_timeout = self._timeouts[0].deadline - self.time()
                    poll_timeout = max(0, min(poll_timeout, _POLL_TIMEOUT))
                else:

                    poll_timeout = _POLL_TIMEOUT

                if not self._running:
                    break

                if self._blocking_signal_threshold is not None:

                    signal.setitimer(signal.ITIMER_REAL, 0, 0)

                try:
                    event_pairs = self._impl.poll(poll_timeout)
                except Exception as e:

                    if errno_from_exception(e) == errno.EINTR:
                        continue
                    else:
                        raise

                if self._blocking_signal_threshold is not None:
                    signal.setitimer(signal.ITIMER_REAL,
                                     self._blocking_signal_threshold, 0)

                self._events.update(event_pairs)
                while self._events:
                    fd, events = self._events.popitem()
                    try:
                        fd_obj, handler_func = self._handlers[fd]
                        handler_func(fd_obj, events)
                    except (OSError, IOError) as e:
                        if errno_from_exception(e) == errno.EPIPE:
                            # Happens when the client closes the connection
                            pass
                        else:
                            self.handle_callback_exception(self._handlers.get(fd))
                    except Exception:
                        self.handle_callback_exception(self._handlers.get(fd))
                fd_obj = handler_func = None

        finally:
            # reset the stopped flag so another start/stop pair can be issued
            self._stopped = False
            if self._blocking_signal_threshold is not None:
                signal.setitimer(signal.ITIMER_REAL, 0, 0)
            IOLoop._current.instance = old_current
            if old_wakeup_fd is not None:
                signal.set_wakeup_fd(old_wakeup_fd)

接下来 tornado 终于也回到了 ioloop 代码中了(泪奔)!!当连接成功时,该 fd 的缓冲区可写,
 epoll 收到 fd 的 write 操作通知~进入到了 epoll 的 loop 中处理.然后!回到刚才注册的 _handle_events 了!
注意这个 _handle_events 是 IOStream 的实例里的 _handle_events ,他有刚才我们处理的所有信息哦~

接下来看 _handle_events 的代码,看他如果解决掉 future 

    def _handle_events(self, fd, events):
        if self.closed():
            gen_log.warning("Got events for closed stream %s", fd)
            return
        try:
            if self._connecting:
                # Most IOLoops will report a write failed connect
                # with the WRITE event, but SelectIOLoop reports a
                # READ as well so we must check for connecting before
                # either.
                self._handle_connect()
            if self.closed():
                return
            if events & self.io_loop.READ:
                self._handle_read()
            if self.closed():
                return
            if events & self.io_loop.WRITE:
                self._handle_write()
            if self.closed():
                return
            if events & self.io_loop.ERROR:
                self.error = self.get_fd_error()
                # We may have queued up a user callback in _handle_read or
                # _handle_write, so don"t close the IOStream until those
                # callbacks have had a chance to run.
                self.io_loop.add_callback(self.close)
                return
            state = self.io_loop.ERROR
            if self.reading():
                state |= self.io_loop.READ
            if self.writing():
                state |= self.io_loop.WRITE
            if state == self.io_loop.ERROR and self._read_buffer_size == 0:
                # If the connection is idle, listen for reads too so
                # we can tell if the connection is closed.  If there is
                # data in the read buffer we won"t run the close callback
                # yet anyway, so we don"t need to listen in this case.
                state |= self.io_loop.READ
            if state != self._state:
                assert self._state is not None, 
                    "shouldn"t happen: _handle_events without self._state"
                self._state = state
                self.io_loop.update_handler(self.fileno(), self._state)
        except UnsatisfiableReadError as e:
            gen_log.info("Unsatisfiable read, closing connection: %s" % e)
            self.close(exc_info=True)
        except Exception:
            gen_log.error("Uncaught exception, closing connection.",
                          exc_info=True)
            self.close(exc_info=True)
            raise

    def _handle_connect(self):
        err = self.socket.getsockopt(socket.SOL_SOCKET, socket.SO_ERROR)
        if err != 0:
            self.error = socket.error(err, os.strerror(err))
            # IOLoop implementations may vary: some of them return
            # an error state before the socket becomes writable, so
            # in that case a connection failure would be handled by the
            # error path in _handle_events instead of here.
            if self._connect_future is None:
                gen_log.warning("Connect error on fd %s: %s",
                                self.socket.fileno(), errno.errorcode[err])
            self.close()
            return
        if self._connect_callback is not None:
            callback = self._connect_callback
            self._connect_callback = None
            self._run_callback(callback)
        if self._connect_future is not None:
            future = self._connect_future
            self._connect_future = None
            future.set_result(self)
        self._connecting = False

判断是否在连接中,当然是了,刚才我也强调过了,
然后进入 _handle_connect,_handle_connect 先判断 connect 有没成功,
成功了就是设置 _connect_future 的 result,set_result(self ),把 self(iostream )设置进去了!   
然后 _connect_future 的 callbacks 会在下一次循环被 ioloop 消化掉!!

一步步返回看,这个 future 正是我们之前的那个 yiled 操作的右边的返回的 future ,
所以刚才 _Connector.try_connect 设置的 callback   ,on_connect_done 会在 ioloop 的 callback 里执行. 
根据上一篇[文章][3]讲的 coroutine 的源码得知,此 future 里还有 Runner.run 的 callback 哦~
所以 ,run 里 send 了 vaule 到 gen . 
终于终于!!程序回到了刚才的 yield 处了!!!!!

tornado正是如此实现异步io的

感觉一直讲完整个操作不太现实,剩下的大家还是自己跟踪吧,道理跟这个流程类似.
 yield 操作右边,一定是返回一个 future (旧版本貌似是 YieldPoint ,因为没看过旧版,所以不太清楚) ,
然后在返回 future 之前,设置好 fd 的 handler ,和其他的解析工作,然后等待 epoll 检测到关心的 io   event ,
在 io 的 handler 里把 future 解决,从而回到 yield 处~ 核心就是 ioloop 三部分 ,future,gen.coroutine . 
相互配合完成异步操作. 跟踪几遍消化一下,就可以写 tornado 的扩展了. 

祝大家武运亨通