首先列一下，sellect、poll、epoll三者的区别
select
select最先于1983年出如今4.2BSD中，它经过一个select()系统调用来监视多个文件描述符的数组，当select()返回后，该数组中就绪的文件描述符便会被内核修改标志位，使得进程能够得到这些文件描述符从而进行后续的读写操做。python
select目前几乎在全部的平台上支持，其良好跨平台支持也是它的一个优势，事实上从如今看来，这也是它所剩很少的优势之一。数组
select的一个缺点在于单个进程可以监视的文件描述符的数量存在最大限制，在Linux上通常为1024，不过能够经过修改宏定义甚至从新编译内核的方式提高这一限制。缓存
另外，select()所维护的存储大量文件描述符的数据结构，随着文件描述符数量的增大，其复制的开销也线性增加。同时，因为网络响应时间的延迟使得大量TCP链接处于非活跃状态，但调用select()会对全部socket进行一次线性扫描，因此这也浪费了必定的开销。服务器
poll
poll在1986年诞生于System V Release 3，它和select在本质上没有多大差异，可是poll没有最大文件描述符数量的限制。网络
poll和select一样存在一个缺点就是，包含大量文件描述符的数组被总体复制于用户态和内核的地址空间之间，而不论这些文件描述符是否就绪，它的开销随着文件描述符数量的增长而线性增大。数据结构
另外，select()和poll()将就绪的文件描述符告诉进程后，若是进程没有对其进行IO操做，那么下次调用select()和poll()的时候将再次报告这些文件描述符，因此它们通常不会丢失就绪的消息，这种方式称为水平触发（Level Triggered）。app
epoll
直到Linux2.6才出现了由内核直接支持的实现方法，那就是epoll，它几乎具有了以前所说的一切优势，被公认为Linux2.6下性能最好的多路I/O就绪通知方法。socket
epoll能够同时支持水平触发和边缘触发（Edge Triggered，只告诉进程哪些文件描述符刚刚变为就绪状态，它只说一遍，若是咱们没有采起行动，那么它将不会再次告知，这种方式称为边缘触发），理论上边缘触发的性能要更高一些，可是代码实现至关复杂。oop
epoll一样只告知那些就绪的文件描述符，并且当咱们调用epoll_wait()得到就绪文件描述符时，返回的不是实际的描述符，而是一个表明就绪描述符数量的值，你只须要去epoll指定的一个数组中依次取得相应数量的文件描述符便可，这里也使用了内存映射（mmap）技术，这样便完全省掉了这些文件描述符在系统调用时复制的开销。性能
另外一个本质的改进在于epoll采用基于事件的就绪通知方式。在select/poll中，进程只有在调用必定的方法后，内核才对全部监视的文件描述符进行扫描，而epoll事先经过epoll_ctl()来注册一个文件描述符，一旦基于某个文件描述符就绪时，内核会采用相似callback的回调机制，迅速激活这个文件描述符，当进程调用epoll_wait()时便获得通知。
Python select
Python的select()方法直接调用操做系统的IO接口，它监控sockets,open files, and pipes(全部带fileno()方法的文件句柄)什么时候变成readable 和writeable, 或者通讯错误，select()使得同时监控多个链接变的简单，而且这比写一个长循环来等待和监控多客户端链接要高效，由于select直接经过操做系统提供的C的网络接口进行操做，而不是经过Python的解释器。
注意：Using Python’s file objects with select() works for Unix, but is not supported under Windows.
接下来经过echo server例子要以了解select 是如何经过单进程实现同时处理多个非阻塞的socket链接的
import select
import socket
import sys
import Queue
# Create a TCP/IP socket
server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server.setblocking(0)
# Bind the socket to the port
server_address = ('localhost', 10000)
print >>sys.stderr, 'starting up on %s port %s' % server_address
server.bind(server_address)
# Listen for incoming connections
server.listen(5)
select()方法接收并监控3个通讯列表， 第一个是全部的输入的data,就是指外部发过来的数据，第2个是监控和接收全部要发出去的data(outgoing data),第3个监控错误信息，接下来咱们须要建立2个列表来包含输入和输出信息来传给select().
# Sockets from which we expect to read
inputs = [ server ] # Sockets to which we expect to write outputs = [ ]
全部客户端的进来的链接和数据将会被server的主循环程序放在上面的list中处理，咱们如今的server端须要等待链接可写（writable)以后才能过来，而后接收数据并返回(所以不是在接收到数据以后就马上返回)，由于每一个链接要把输入或输出的数据先缓存到queue里，而后再由select取出来再发出去。
Connections are added to and removed from these lists by the server main loop. Since this version of the server is going to wait for a socket to become writable before sending any data (instead of immediately sending the reply), each output connection needs a queue to act as a buffer for the data to be sent through it.
# Outgoing message queues (socket:Queue)
message_queues = {}
The main portion of the server program loops, calling select() to block and wait for network activity.
下面是此程序的主循环，调用select()时会阻塞和等待直到新的链接和数据进来
while inputs: # Wait for at least one of the sockets to be ready for processing print >>sys.stderr, '\nwaiting for the next event' readable, writable, exceptional = select.select(inputs, outputs, inputs)
当你把inputs,outputs,exceptional(这里跟inputs共用)传给select()后，它返回3个新的list，咱们上面将他们分别赋值为readable,writable,exceptional, 全部在readable list中的socket链接表明有数据可接收(recv),全部在writable list中的存放着你能够对其进行发送(send)操做的socket链接，当链接通讯出现error时会把error写到exceptional列表中。
select() returns three new lists, containing subsets of the contents of the lists passed in. All of the sockets in the readable list have incoming data buffered and available to be read. All of the sockets in the writable list have free space in their buffer and can be written to. The sockets returned in exceptional have had an error (the actual definition of “exceptional condition” depends on the platform).
Readable list 中的socket 能够有3种可能状态，第一种是若是这个socket是main "server" socket,它负责监听客户端的链接，若是这个main server socket出如今readable里，那表明这是server端已经ready来接收一个新的链接进来了，为了让这个main server能同时处理多个链接，在下面的代码里，咱们把这个main server的socket设置为非阻塞模式。
The “readable” sockets represent three possible cases. If the socket is the main “server” socket, the one being used to listen for connections, then the “readable” condition means it is ready to accept another incoming connection. In addition to adding the new connection to the list of inputs to monitor, this section sets the client socket to not block.
# Handle inputs
for s in readable:
if s is server:
# A "readable" server socket is ready to accept a connection
connection, client_address = s.accept()
print >>sys.stderr, 'new connection from', client_address
connection.setblocking(0)
inputs.append(connection)
# Give the connection a queue for data we want to send
message_queues[connection] = Queue.Queue()
第二种状况是这个socket是已经创建了的链接，它把数据发了过来，这个时候你就能够经过recv()来接收它发过来的数据，而后把接收到的数据放到queue里，这样你就能够把接收到的数据再传回给客户端了。
The next case is an established connection with a client that has sent data. The data is read with recv(), then placed on the queue so it can be sent through the socket and back to the client.
else:
data = s.recv(1024)
if data:
# A readable client socket has data
print >>sys.stderr, 'received "%s" from %s' % (data, s.getpeername())
message_queues[s].put(data)
# Add output channel for response
if s not in outputs:
outputs.append(s)
第三种状况就是这个客户端已经断开了，因此你再经过recv()接收到的数据就为空了，因此这个时候你就能够把这个跟客户端的链接关闭了。
A readable socket without data available is from a client that has disconnected, and the stream is ready to be closed.
else:
# Interpret empty result as closed connection
print >>sys.stderr, 'closing', client_address, 'after reading no data'
# Stop listening for input on the connection
if s in outputs:
outputs.remove(s) #既然客户端都断开了，我就不用再给它返回数据了，因此这时候若是这个客户端的链接对象还在outputs列表中，就把它删掉
inputs.remove(s) #inputs中也删除掉
s.close() #把这个链接关闭掉
# Remove message queue
del message_queues[s]
对于writable list中的socket，也有几种状态，若是这个客户端链接在跟它对应的queue里有数据，就把这个数据取出来再发回给这个客户端，不然就把这个链接从output list中移除，这样下一次循环select()调用时检测到outputs list中没有这个链接，那就会认为这个链接还处于非活动状态
There are fewer cases for the writable connections. If there is data in the queue for a connection, the next message is sent. Otherwise, the connection is removed from the list of output connections so that the next time through the loop select() does not indicate that the socket is ready to send data.
# Handle outputs
for s in writable:
try:
next_msg = message_queues[s].get_nowait()
except Queue.Empty:
# No messages waiting so stop checking for writability.
print >>sys.stderr, 'output queue for', s.getpeername(), 'is empty'
outputs.remove(s)
else:
print >>sys.stderr, 'sending "%s" to %s' % (next_msg, s.getpeername())
s.send(next_msg)
最后，若是在跟某个socket链接通讯过程当中出了错误，就把这个链接对象在inputs\outputs\message_queue中都删除，再把链接关闭掉
# Handle "exceptional conditions"
for s in exceptional:
print >>sys.stderr, 'handling exceptional condition for', s.getpeername()
# Stop listening for input on the connection
inputs.remove(s)
if s in outputs:
outputs.remove(s)
s.close()
# Remove message queue
del message_queues[s]
客户端
下面的这个是客户端程序展现了如何经过select()对socket进行管理并与多个链接同时进行交互，
The example client program uses two sockets to demonstrate how the server with select() manages multiple connections at the same time. The client starts by connecting each TCP/IP socket to the server.
import socket
import sys
messages = [ 'This is the message. ',
'It will be sent ',
'in parts.',
]
server_address = ('localhost', 10000)
# Create a TCP/IP socket
socks = [ socket.socket(socket.AF_INET, socket.SOCK_STREAM),
socket.socket(socket.AF_INET, socket.SOCK_STREAM),
]
# Connect the socket to the port where the server is listening
print >>sys.stderr, 'connecting to %s port %s' % server_address
for s in socks:
s.connect(server_address)
接下来经过循环经过每一个socket链接给server发送和接收数据。
Then it sends one pieces of the message at a time via each socket, and reads all responses available after writing new data.
for message in messages:
# Send messages on both sockets
for s in socks:
print >>sys.stderr, '%s: sending "%s"' % (s.getsockname(), message)
s.send(message)
# Read responses on both sockets
for s in socks:
data = s.recv(1024)
print >>sys.stderr, '%s: received "%s"' % (s.getsockname(), data)
if not data:
print >>sys.stderr, 'closing socket', s.getsockname()
最后服务器端的完整代码以下：
#_*_coding:utf-8_*_
__author__ = 'Alex Li'
import select
import socket
import sys
import queue
# Create a TCP/IP socket
server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server.setblocking(False)
# Bind the socket to the port
server_address = ('localhost', 10000)
print(sys.stderr, 'starting up on %s port %s' % server_address)
server.bind(server_address)
# Listen for incoming connections
server.listen(5)
# Sockets from which we expect to read
inputs = [ server ]
# Sockets to which we expect to write
outputs = [ ]
message_queues = {}
while inputs:
# Wait for at least one of the sockets to be ready for processing
print( '\nwaiting for the next event')
readable, writable, exceptional = select.select(inputs, outputs, inputs)
# Handle inputs
for s in readable:
if s is server:
# A "readable" server socket is ready to accept a connection
connection, client_address = s.accept()
print('new connection from', client_address)
connection.setblocking(False)
inputs.append(connection)
# Give the connection a queue for data we want to send
message_queues[connection] = queue.Queue()
else:
data = s.recv(1024)
if data:
# A readable client socket has data
print(sys.stderr, 'received "%s" from %s' % (data, s.getpeername()) )
message_queues[s].put(data)
# Add output channel for response
if s not in outputs:
outputs.append(s)
else:
# Interpret empty result as closed connection
print('closing', client_address, 'after reading no data')
# Stop listening for input on the connection
if s in outputs:
outputs.remove(s) #既然客户端都断开了，我就不用再给它返回数据了，因此这时候若是这个客户端的链接对象还在outputs列表中，就把它删掉
inputs.remove(s) #inputs中也删除掉
s.close() #把这个链接关闭掉
# Remove message queue
del message_queues[s]
# Handle outputs
for s in writable:
try:
next_msg = message_queues[s].get_nowait()
except queue.Empty:
# No messages waiting so stop checking for writability.
print('output queue for', s.getpeername(), 'is empty')
outputs.remove(s)
else:
print( 'sending "%s" to %s' % (next_msg, s.getpeername()))
s.send(next_msg)
# Handle "exceptional conditions"
for s in exceptional:
print('handling exceptional condition for', s.getpeername() )
# Stop listening for input on the connection
inputs.remove(s)
if s in outputs:
outputs.remove(s)
s.close()
# Remove message queue
del message_queues[s]
客户端的完整代码以下：
__author__ = 'jieli'
import socket
import sys
messages = [ 'This is the message. ',
'It will be sent ',
'in parts.',
]
server_address = ('localhost', 10000)
# Create a TCP/IP socket
socks = [ socket.socket(socket.AF_INET, socket.SOCK_STREAM),
socket.socket(socket.AF_INET, socket.SOCK_STREAM),
]
# Connect the socket to the port where the server is listening
print >>sys.stderr, 'connecting to %s port %s' % server_address
for s in socks:
s.connect(server_address)
for message in messages:
# Send messages on both sockets
for s in socks:
print >>sys.stderr, '%s: sending "%s"' % (s.getsockname(), message)
s.send(message)
# Read responses on both sockets
for s in socks:
data = s.recv(1024)
print >>sys.stderr, '%s: received "%s"' % (s.getsockname(), data)
if not data:
print >>sys.stderr, 'closing socket', s.getsockname()
s.close()
Run the server in one window and the client in another. The output will look like this, with different port numbers.
$ python ./select_echo_server.py
starting up on localhost port 10000
waiting for the next event
new connection from ('127.0.0.1', 55821)
waiting for the next event
new connection from ('127.0.0.1', 55822)
received "This is the message. " from ('127.0.0.1', 55821)
waiting for the next event
sending "This is the message. " to ('127.0.0.1', 55821)
waiting for the next event
output queue for ('127.0.0.1', 55821) is empty
waiting for the next event
received "This is the message. " from ('127.0.0.1', 55822)
waiting for the next event
sending "This is the message. " to ('127.0.0.1', 55822)
waiting for the next event
output queue for ('127.0.0.1', 55822) is empty
waiting for the next event
received "It will be sent " from ('127.0.0.1', 55821)
received "It will be sent " from ('127.0.0.1', 55822)
waiting for the next event
sending "It will be sent " to ('127.0.0.1', 55821)
sending "It will be sent " to ('127.0.0.1', 55822)
waiting for the next event
output queue for ('127.0.0.1', 55821) is empty
output queue for ('127.0.0.1', 55822) is empty
waiting for the next event
received "in parts." from ('127.0.0.1', 55821)
received "in parts." from ('127.0.0.1', 55822)
waiting for the next event
sending "in parts." to ('127.0.0.1', 55821)
sending "in parts." to ('127.0.0.1', 55822)
waiting for the next event
output queue for ('127.0.0.1', 55821) is empty
output queue for ('127.0.0.1', 55822) is empty
waiting for the next event
closing ('127.0.0.1', 55822) after reading no data
closing ('127.0.0.1', 55822) after reading no data
waiting for the next event
The client output shows the data being sent and received using both sockets.
$ python ./select_echo_multiclient.py
connecting to localhost port 10000
('127.0.0.1', 55821): sending "This is the message. "
('127.0.0.1', 55822): sending "This is the message. "
('127.0.0.1', 55821): received "This is the message. "
('127.0.0.1', 55822): received "This is the message. "
('127.0.0.1', 55821): sending "It will be sent "
('127.0.0.1', 55822): sending "It will be sent "
('127.0.0.1', 55821): received "It will be sent "
('127.0.0.1', 55822): received "It will be sent "
('127.0.0.1', 55821): sending "in parts."
('127.0.0.1', 55822): sending "in parts."
('127.0.0.1', 55821): received "in parts."
('127.0.0.1', 55822): received "in parts."
原文：http://pymotw.com/2/select/