18.5.4. Transports and protocols (callback based API)

Source code: Lib/asyncio/transports.py

Source code: Lib/asyncio/protocols.py 传输

Transports are classes provided by asyncio in order to abstract various kinds of communication channels. You generally won’t instantiate a transport yourself; instead, you will call an AbstractEventLoop method which will create the transport and try to initiate the underlying communication channel, calling you back when it succeeds.

Once the communication channel is established, a transport is always paired with a protocol instance. The protocol can then call the transport’s methods for various purposes.

asyncio currently implements transports for TCP, UDP, SSL, and subprocess pipes. The methods available on a transport depend on the transport’s kind.

传输类属于 线程不安全

在 3.6 版更改: The socket option TCP_NODELAY is now set by default. BaseTransport

class asyncio.BaseTransport

Base class for transports.


Close the transport. If the transport has a buffer for outgoing data, buffered data will be flushed asynchronously. No more data will be received. After all buffered data is flushed, the protocol’s connection_lost() method will be called with None as its argument.


返回 True ,如果传输正在关闭或已经关闭。

3.5.1 新版功能.

get_extra_info(name, default=None)

Return optional transport information. name is a string representing the piece of transport-specific information to get, default is the value to return if the information doesn’t exist.

This method allows transport implementations to easily expose channel-specific information.


Set a new protocol. Switching protocol should only be done when both protocols are documented to support the switch.

3.5.3 新版功能.



3.5.3 新版功能.

在 3.5.1 版更改: 'ssl_object' info was added to SSL sockets. ReadTransport

class asyncio.ReadTransport

Interface for read-only transports.


Pause the receiving end of the transport. No data will be passed to the protocol’s data_received() method until resume_reading() is called.

在 3.6.7 版更改: 这个方法幂等的, 它可以在传输已经暂停或关闭时调用。


Resume the receiving end. The protocol’s data_received() method will be called once again if some data is available for reading.

在 3.6.7 版更改: 这个方法幂等的, 它可以在传输已经准备好读取数据时调用。 WriteTransport

class asyncio.WriteTransport

Interface for write-only transports.


Close the transport immediately, without waiting for pending operations to complete. Buffered data will be lost. No more data will be received. The protocol’s connection_lost() method will eventually be called with None as its argument.


Return True if the transport supports write_eof(), False if not.




Get the high- and low-water limits for write flow control. Return a tuple (low, high) where low and high are positive number of bytes.

使用 set_write_buffer_limits() 设置限制。

3.4.2 新版功能.

set_write_buffer_limits(high=None, low=None)

Set the high- and low-water limits for write flow control.

These two values (measured in number of bytes) control when the protocol’s pause_writing() and resume_writing() methods are called. If specified, the low-water limit must be less than or equal to the high-water limit. Neither high nor low can be negative.

pause_writing() is called when the buffer size becomes greater than or equal to the high value. If writing has been paused, resume_writing() is called when the buffer size becomes less than or equal to the low value.

The defaults are implementation-specific. If only the high-water limit is given, the low-water limit defaults to an implementation-specific value less than or equal to the high-water limit. Setting high to zero forces low to zero as well, and causes pause_writing() to be called whenever the buffer becomes non-empty. Setting low to zero causes resume_writing() to be called only once the buffer is empty. Use of zero for either limit is generally sub-optimal as it reduces opportunities for doing I/O and computation concurrently.

Use get_write_buffer_limits() to get the limits.


将一些 data 字节串写入传输。



将数据字节串的列表(或任意可迭代对象)写入传输。 这在功能上等价于在可迭代对象产生的每个元素上调用 write(),但其实现可能更为高效。


Close the write end of the transport after flushing buffered data. Data may still be received.

This method can raise NotImplementedError if the transport (e.g. SSL) doesn’t support half-closes. DatagramTransport

DatagramTransport.sendto(data, addr=None)

data 字节串发送到 addr (基于传输的目标地址) 所给定的远端对等方。 如果 addrNone,则将数据发送到传输创建时给定的目标地址。



Close the transport immediately, without waiting for pending operations to complete. Buffered data will be lost. No more data will be received. The protocol’s connection_lost() method will eventually be called with None as its argument. BaseSubprocessTransport

class asyncio.BaseSubprocessTransport

将子进程的进程 ID 以整数形式返回。


返回对应于整数文件描述符 fd 的通信管道的传输:

  • 0: 标准输入 (stdin) 的可读流式传输,如果子进程创建时未设置 stdin=PIPE 则为 None

  • 1: 标准输出writable streaming transport of the standard output (stdout) 的可写流式传输,如果子进程创建时未设置 stdout=PIPE 则为 None

  • 2: 标准错误 (stderr) 的可写流式传输,如果子进程创建时未设置 stderr=PIPE 则为 None

  • 其他 fd: None


Return the subprocess returncode as an integer or None if it hasn’t returned, similarly to the subprocess.Popen.returncode attribute.


Kill the subprocess, as in subprocess.Popen.kill().

在 POSIX 系统中,函数会发送 SIGKILL 到子进程。 在 Windows 中,此方法是 terminate() 的别名。


发送 signal 编号到子进程,与 subprocess.Popen.send_signal() 一样。


Ask the subprocess to stop, as in subprocess.Popen.terminate(). This method is an alias for the close() method.

在 POSIX 系统中,此方法会发送 SIGTERM 到子进程。 在 Windows 中,则会调用 Windows API 函数 TerminateProcess() 来停止子进程。


Ask the subprocess to stop by calling the terminate() method if the subprocess hasn’t returned yet, and close transports of all pipes (stdin, stdout and stderr). 协议

asyncio provides base classes that you can subclass to implement your network protocols. Those classes are used in conjunction with transports (see below): the protocol parses incoming data and asks for the writing of outgoing data, while the transport is responsible for the actual I/O and buffering.

When subclassing a protocol class, it is recommended you override certain methods. Those methods are callbacks: they will be called by the transport on certain events (for example when some data is received); you shouldn’t call them yourself, unless you are implementing a transport.


All callbacks have default implementations, which are empty. Therefore, you only need to implement the callbacks for the events in which you are interested. Protocol classes

class asyncio.Protocol

The base class for implementing streaming protocols (for use with e.g. TCP and SSL transports).

class asyncio.DatagramProtocol

The base class for implementing datagram protocols (for use with e.g. UDP transports).

class asyncio.SubprocessProtocol

The base class for implementing protocols communicating with child processes (through a set of unidirectional pipes). Connection callbacks

These callbacks may be called on Protocol, DatagramProtocol and SubprocessProtocol instances:



The transport argument is the transport representing the connection. You are responsible for storing it somewhere (e.g. as an attribute) if you need to.



方法的参数是一个异常对象或为 None。 后者意味着收到了常规的 EOF,或者连接被连接的一端取消或关闭。

connection_made() and connection_lost() are called exactly once per successful connection. All other callbacks will be called between those two methods, which allows for easier resource management in your protocol implementation.

The following callbacks may be called only on SubprocessProtocol instances:

SubprocessProtocol.pipe_data_received(fd, data)

Called when the child process writes data into its stdout or stderr pipe. fd is the integer file descriptor of the pipe. data is a non-empty bytes object containing the data.

SubprocessProtocol.pipe_connection_lost(fd, exc)

Called when one of the pipes communicating with the child process is closed. fd is the integer file descriptor that was closed.


子进程退出时被调用。 Streaming protocols

The following callbacks are called on Protocol instances:


当收到数据时被调用。 data 为包含入站数据的非空字节串对象。


Whether the data is buffered, chunked or reassembled depends on the transport. In general, you shouldn’t rely on specific semantics and instead make your parsing generic and flexible enough. However, data is always received in the correct order.


Called when the other end signals it won’t send any more data (for example by calling write_eof(), if the other end also uses asyncio).

This method may return a false value (including None), in which case the transport will close itself. Conversely, if this method returns a true value, closing the transport is up to the protocol. Since the default implementation returns None, it implicitly closes the connection.


Some transports such as SSL don’t support half-closed connections, in which case returning true from this method will not prevent closing the connection.

data_received() can be called an arbitrary number of times during a connection. However, eof_received() is called at most once and, if called, data_received() won’t be called after it.

状态机: Datagram protocols

The following callbacks are called on DatagramProtocol instances.

DatagramProtocol.datagram_received(data, addr)

当接收到数据报时被调用。 data 是包含传入数据的字节串对象。 addr 是发送数据的对等端地址;实际的格式取决于具体传输。


当前一个发送或接收操作引发 OSError 时被调用。 excOSError 的实例。

This method is called in rare conditions, when the transport (e.g. UDP) detects that a datagram couldn’t be delivered to its recipient. In many conditions though, undeliverable datagrams will be silently dropped. Flow control callbacks

These callbacks may be called on Protocol, DatagramProtocol and SubprocessProtocol instances:


Called when the transport’s buffer goes over the high-water mark.


Called when the transport’s buffer drains below the low-water mark.

pause_writing() and resume_writing() calls are paired – pause_writing() is called once when the buffer goes strictly over the high-water mark (even if subsequent writes increases the buffer size even more), and eventually resume_writing() is called once when the buffer size reaches the low-water mark.


If the buffer size equals the high-water mark, pause_writing() is not called – it must go strictly over. Conversely, resume_writing() is called when the buffer size is equal or lower than the low-water mark. These end conditions are important to ensure that things go as expected when either mark is zero.


On BSD systems (OS X, FreeBSD, etc.) flow control is not supported for DatagramProtocol, because send failures caused by writing too many packets cannot be detected easily. The socket always appears ‘ready’ and excess packets are dropped; an OSError with errno set to errno.ENOBUFS may or may not be raised; if it is raised, it will be reported to DatagramProtocol.error_received() but otherwise ignored. Coroutines and protocols

Coroutines can be scheduled in a protocol method using ensure_future(), but there is no guarantee made about the execution order. Protocols are not aware of coroutines created in protocol methods and so will not wait for them.

To have a reliable execution order, use stream objects in a coroutine with yield from. For example, the StreamWriter.drain() coroutine can be used to wait until the write buffer is flushed. Protocol examples TCP echo client protocol

TCP echo client using the AbstractEventLoop.create_connection() method, send data and wait until the connection is closed:

import asyncio

class EchoClientProtocol(asyncio.Protocol):
    def __init__(self, message, loop):
        self.message = message
        self.loop = loop

    def connection_made(self, transport):
        print('Data sent: {!r}'.format(self.message))

    def data_received(self, data):
        print('Data received: {!r}'.format(data.decode()))

    def connection_lost(self, exc):
        print('The server closed the connection')
        print('Stop the event loop')

loop = asyncio.get_event_loop()
message = 'Hello World!'
coro = loop.create_connection(lambda: EchoClientProtocol(message, loop),
                              '', 8888)

The event loop is running twice. The run_until_complete() method is preferred in this short example to raise an exception if the server is not listening, instead of having to write a short coroutine to handle the exception and stop the running loop. At run_until_complete() exit, the loop is no longer running, so there is no need to stop the loop in case of an error.


The TCP echo client using streams example uses the asyncio.open_connection() function. TCP echo server protocol

TCP echo server using the AbstractEventLoop.create_server() method, send back received data and close the connection:

import asyncio

class EchoServerClientProtocol(asyncio.Protocol):
    def connection_made(self, transport):
        peername = transport.get_extra_info('peername')
        print('Connection from {}'.format(peername))
        self.transport = transport

    def data_received(self, data):
        message = data.decode()
        print('Data received: {!r}'.format(message))

        print('Send: {!r}'.format(message))

        print('Close the client socket')

loop = asyncio.get_event_loop()
# Each client connection will create a new protocol instance
coro = loop.create_server(EchoServerClientProtocol, '', 8888)
server = loop.run_until_complete(coro)

# Serve requests until Ctrl+C is pressed
print('Serving on {}'.format(server.sockets[0].getsockname()))
except KeyboardInterrupt:

# Close the server

Transport.close() can be called immediately after WriteTransport.write() even if data are not sent yet on the socket: both methods are asynchronous. yield from is not needed because these transport methods are not coroutines.


The TCP echo server using streams example uses the asyncio.start_server() function. UDP echo client protocol

UDP echo client using the AbstractEventLoop.create_datagram_endpoint() method, send data and close the transport when we received the answer:

import asyncio

class EchoClientProtocol:
    def __init__(self, message, loop):
        self.message = message
        self.loop = loop
        self.transport = None

    def connection_made(self, transport):
        self.transport = transport
        print('Send:', self.message)

    def datagram_received(self, data, addr):
        print("Received:", data.decode())

        print("Close the socket")

    def error_received(self, exc):
        print('Error received:', exc)

    def connection_lost(self, exc):
        print("Socket closed, stop the event loop")
        loop = asyncio.get_event_loop()

loop = asyncio.get_event_loop()
message = "Hello World!"
connect = loop.create_datagram_endpoint(
    lambda: EchoClientProtocol(message, loop),
    remote_addr=('', 9999))
transport, protocol = loop.run_until_complete(connect)
loop.close() UDP echo server protocol

UDP echo server using the AbstractEventLoop.create_datagram_endpoint() method, send back received data:

import asyncio

class EchoServerProtocol:
    def connection_made(self, transport):
        self.transport = transport

    def datagram_received(self, data, addr):
        message = data.decode()
        print('Received %r from %s' % (message, addr))
        print('Send %r to %s' % (message, addr))
        self.transport.sendto(data, addr)

loop = asyncio.get_event_loop()
print("Starting UDP server")
# One protocol instance will be created to serve all client requests
listen = loop.create_datagram_endpoint(
    EchoServerProtocol, local_addr=('', 9999))
transport, protocol = loop.run_until_complete(listen)

except KeyboardInterrupt:

loop.close() Register an open socket to wait for data using a protocol

Wait until a socket receives data using the AbstractEventLoop.create_connection() method with a protocol, and then close the event loop

import asyncio
    from socket import socketpair
except ImportError:
    from asyncio.windows_utils import socketpair

# Create a pair of connected sockets
rsock, wsock = socketpair()
loop = asyncio.get_event_loop()

class MyProtocol(asyncio.Protocol):
    transport = None

    def connection_made(self, transport):
        self.transport = transport

    def data_received(self, data):
        print("Received:", data.decode())

        # We are done: close the transport (it will call connection_lost())

    def connection_lost(self, exc):
        # The socket has been closed, stop the event loop

# Register the socket to wait for data
connect_coro = loop.create_connection(MyProtocol, sock=rsock)
transport, protocol = loop.run_until_complete(connect_coro)

# Simulate the reception of data from the network
loop.call_soon(wsock.send, 'abc'.encode())

# Run the event loop

# We are done, close sockets and the event loop


The watch a file descriptor for read events example uses the low-level AbstractEventLoop.add_reader() method to register the file descriptor of a socket.

The register an open socket to wait for data using streams example uses high-level streams created by the open_connection() function in a coroutine.