Tornado 高并发源码分析之四--- HTTPServer 与 TCPServer 对象

 1 class HTTPServer(TCPServer, httputil.HTTPServerConnectionDelegate):
 2 #继承自TCPServer 和 httputil.HTTPServerConnectionDelegate
 3 
 4 def __init__(self, request_callback, no_keep_alive=False, io_loop=None,
 5              xheaders=False, ssl_options=None, protocol=None,
 6              decompress_request=False,
 7              chunk_size=None, max_header_size=None,
 8              idle_connection_timeout=None, body_timeout=None,
 9              max_body_size=None, max_buffer_size=None):
10     self.request_callback = request_callback   #获取 application对象
11     self.no_keep_alive = no_keep_alive
12     self.xheaders = xheaders
13     self.protocol = protocol
14     self.conn_params = HTTP1ConnectionParameters(   #获取基本的http 参数
15         decompress=decompress_request,
16         chunk_size=chunk_size,
17         max_header_size=max_header_size,
18         header_timeout=idle_connection_timeout or 3600,
19         max_body_size=max_body_size,
20         body_timeout=body_timeout)
21     TCPServer.__init__(self, io_loop=io_loop, ssl_options=ssl_options,   传递HTTPServer 给TCPServer对象，初始化
22                        max_buffer_size=max_buffer_size,
23                        read_chunk_size=chunk_size)
24     self._connections = set()
25 
26 def handle_stream(self, stream, address):  
27  #请求到来时，真正的数据处理方法
28     context = _HTTPRequestContext(stream, address,
29                                   self.protocol)
30     conn = HTTP1ServerConnection(   #创建HTTP1ServerConnection对象，用来获取链接
31         stream, self.conn_params, context)
32     self._connections.add(conn)
33     conn.start_serving(self)   #启动获取数据服务， 传递自己为代理
34 
35 def start_request(self, server_conn, request_conn): #将会在handle_stream添加进去的代理中执行
36     return _ServerRequestAdapter(self, request_conn)   #创建_ServerRequestAdapter对象， 开始接受封装数据

 1 class _ServerRequestAdapter(httputil.HTTPMessageDelegate):
 2     """负责数据流整理，将数据最后封装成 HTTPRequest对象
 3     """
 4     def __init__(self, server, connection):
 5         self.server = server
 6         self.connection = connection
 7         self.request = None
 8         if isinstance(server.request_callback, httputil.HTTPServerConnectionDelegate):
 9             self.delegate = server.request_callback.start_request(connection)
10             self._chunks = None
11         else:
12             self.delegate = None
13             self._chunks = []
14 
15     def headers_received(self, start_line, headers):   #接受请求头部数据
16         if self.server.xheaders:
17             self.connection.context._apply_xheaders(headers)
18         if self.delegate is None:
19             self.request = httputil.HTTPServerRequest(  #创建HTTPServerRequest对象
20                 connection=self.connection, start_line=start_line,
21                 headers=headers)
22         else:
23             return self.delegate.headers_received(start_line, headers)
24 
25     def data_received(self, chunk):  #不停的接受数据，放到_chunks里面
26         if self.delegate is None:
27             self._chunks.append(chunk)
28         else:
29             return self.delegate.data_received(chunk)
30 
31     def finish(self):  #request_callback本身就是一个application对象，在加上括号，那就是执行这个对象，也是就是调用application 的 __call__方法，并传入了封装好的request对象
32         if self.delegate is None:
33             self.request.body = b‘‘.join(self._chunks)
34             self.request._parse_body()
35             self.server.request_callback(self.request)
36         else:
37             self.delegate.finish()
38         self._cleanup()

 1 TCPServer 启动一个 Socket 服务端口监听有三中方式
 2 1. `listen`: 单进程启动一个监听
 3         server = TCPServer()
 4         server.listen(8888)
 5         IOLoop.instance().start()
 6 
 7 2. `bind`/`start`: 多进程启动一个监听，取决于start(1)的进程数 
 8         server = TCPServer()
 9         server.bind(8888)
10         server.start(0)  # Forks multiple sub-processes
11         IOLoop.instance().start()
12 
13 3. `add_sockets`: 另外一种方式启动多进程
14         sockets = bind_sockets(8888)
15         tornado.process.fork_processes(0)
16         server = TCPServer()
17         server.add_sockets(sockets)
18         IOLoop.instance().start()
19 
20 class TCPServer(object):
21 
22 def listen(self, port, address=""):   #单进程启动
23     sockets = bind_sockets(port, address=address)    #绑定 socket 监听端口
24     self.add_sockets(sockets)      #添加监听
25 
26 def add_sockets(self, sockets):
27     if self.io_loop is None:
28         self.io_loop = IOLoop.current()  #获取 IOLoop 单例
29     for sock in sockets:
30         self._sockets[sock.fileno()] = sock
31  #将封装好的 self._handle_connection 回调函数与sock对象一起绑定到IOLoop 中， 这一步很重要，_handle_connection 是封装之后的回调函数，在 IOLoop 中会回调它
32         add_accept_handler(sock, self._handle_connection,   io_loop=self.io_loop)  #add_accept_handler方法封装在netutil 文件中
33 
34 def _handle_connection(self, connection, address):
35  #创建IOStream对象，并封装回调函数， 这个方法并不会在这里执行，只是在这里封装好之后，被当做参数传递与IOLoop 绑定 .......
36     try:
37         if self.ssl_options is not None:   #https 相关
38             stream = SSLIOStream(connection, io_loop=self.io_loop,
39                                  max_buffer_size=self.max_buffer_size,
40                                  read_chunk_size=self.read_chunk_size)
41         else:
42             stream = IOStream(connection, io_loop=self.io_loop,    #创建一个 IOStream， 主要用来读取HTTP 数据流
43                               max_buffer_size=self.max_buffer_size,
44                               read_chunk_size=self.read_chunk_size)
45  #调用继承至 TCPServer的类的handle_stream方法（也就是HTTPServer 的handle_stream 方法），传入IOStream和address， 
46         self.handle_stream(stream, address)  
47     except Exception:
48         app_log.error("Error in connection callback", exc_info=True)
49 
50 #*******下面两个方法是用于多进程的方式启动 TCPServer 可以不看********
51 def bind(self, port, address=None, family=socket.AF_UNSPEC, backlog=128):
52         #绑定socket
53     sockets = bind_sockets(port, address=address, family=family,
54                            backlog=backlog)
55     if self._started:
56         self.add_sockets(sockets)
57     else:
58         self._pending_sockets.extend(sockets)
59 
60 def start(self, num_processes=1):
61 #多进程方式启动TCPServer，也就是一个 tornado 运行时，启动多个进程，同时监听一个端口，一般不推荐这么使用，而是采用 Supervisor 启动多个tornado进程，分别监听不同的端口
62     assert not self._started
63     self._started = True
64     if num_processes != 1:
65         process.fork_processes(num_processes)   #fork 子进程
66     sockets = self._pending_sockets
67     self._pending_sockets = []
68     self.add_sockets(sockets)