多进程和多线程的区别

• Python多线程的操作，由于有GIL锁的存在，使得其运行效率并不会很高，无法充分利用 多核cpu 的优势，只有在I/O密集形的任务逻辑中才能实现并发。
• 使用多进程来编写同样消耗cpu（一般是计算）的逻辑，对于 多核cpu 来说效率会好很多。
• 操作系统对进程的调度代价要比线程调度要大的多。

多线程和多进程使用案例对比

1.用多进程和多线程两种方式来运算 斐波那契数列，这里都依赖 concurrent.futures 模块提供的线/进程池。

import time
from concurrent.futures import ThreadPoolExecutor
from concurrent.futures import ProcessPoolExecutor
from concurrent.futures import as_completed
def fib(n):
 return 1 if n <= 2 else fib(n-1) + fib(n-2)
if __name__ == '__main__':
 # with ProcessPoolExecutor(3) as executor:
 with ThreadPoolExecutor(3) as executor:
 all_task = [executor.submit(fib, n) for n in range(25, 35)]
 start_time = time.time()
 for future in as_completed(all_task):
 data = future.result()
 # todo
 
 end_time = time.time()
 
 
 print("time consuming by threads: {0}s".format(end_time-start_time))
 # print("time consuming by processes: {0}s".format(end_time-start_time))

两种方式的运行结果对比：

# result:
# time consuming by threads: 4.823292016983032s
# time consuming by processes: 3.3890748023986816s

可以看到，对于高计算量的任务，多进程要比多线程更加高效。同时，从这个例子中还能看出，通过concurrent.futures模块使用线程池和进程池的方式的接口和使用逻辑是一样的，不过在使用多进程时，对于Windows的操作平台，相关逻辑一定要放在main中，Linux不受约束。

2.用多进程和多线程两种方式来模拟 I/O密集操作，I/O操作 的特点就是 cpu 要耗费大量的时间进行等待数据，这里用sleep()进行模拟即可。

整体的操作方式不变，修改过的逻辑如下：

def random_sleep(n):
 time.sleep(n)
 return n
 
...
# 8 个线程，每个休眠两秒，模拟 I/O
with ProcessPoolExecutor(8) as executor:
# with ThreadPoolExecutor(8) as executor:
 all_task = [executor.submit(random_sleep, 2) for i in range(30)]

# result:
# time consuming by threads: 8.002903699874878s
# time consuming by processes: 8.34946894645691s

多进程编程

直接使用

import time
import multiprocessing
def read(times):
 time.sleep(times)
 print("process reading...")
 return "read for {0}s".format(times)
def write(times):
 time.sleep(times)
 print("process writing...")
 return "write for {0}s".format(times)
if __name__ == '__main__':
 read_process = multiprocessing.Process(target=read, args=(1,))
 write_process = multiprocessing.Process(target=write, args=(2,))
 read_process.start()
 write_process.start()
 print("read_process id {rid}".format(rid=read_process.pid))
 print("write_process id {wid}".format(wid=write_process.pid))
 read_process.join()
 write_process.join()
 print("done")
 
# result:
# read_process id 7064
# write_process id 836
# process reading...
# process writing...
# done

可以看出，关于多线程的逻辑和多线程的使用方式以类似的，要注意在Windows操作系统上，和进程有关的逻辑要写在if __name__ == '__main__'中。其他的一些方法请参阅 官方文档。

使用原生进程池

import time
import multiprocessing
def read(times):
 time.sleep(times)
 print("process reading...")
 return "read for {0}s".format(times)
def write(times):
 time.sleep(times)
 print("process writing...")
 return "write for {0}s".format(times)
if __name__ == '__main__':
 # multiprocessing.cpu_count() 获取cpu的核心数
 pool = multiprocessing.Pool(multiprocessing.cpu_count())
 
 read_result = pool.apply_async(read, args=(2,))
 write_result = pool.apply_async(write, args=(3,))
 # 关闭进程池，不再接受新的任务提交，否则 join() 出错
 pool.close()
 # 等待进程池中提交的所有任务完成
 pool.join()
 print(read_result.get())
 print(write_result.get())
# result:
# process reading...
# process writing...
# read for 2s
# write for 3s

使用imap()，所有任务顺序执行:

pool = multiprocessing.Pool(multiprocessing.cpu_count())
 
for result in pool.imap(read, [2, 1, 3]):
 print(result)
 
# result:
# process reading...
# process reading...
# read for 2s
# read for 1s
# process reading...
# read for 3s

使用imap_unordered()，哪个任务先完成就先返回结果：

for result in pool.imap_unordered(read, [1, 5, 3]):
 print(result)
# process reading...
# read for 1s
# process reading...
# read for 3s
# process reading...
# read for 5s

使用concurrent.futures中的ProcessPoolExecutor

这个在多线程和多进程对比的时提到过，因为和多线程的使用方式一样，这里就不多赘述，可以参阅 官方文档 给出的例子

进程间通信

进程通信和线程通信有些区别，在线程通信中各种提供的锁的机制和全局变量在这里不再适用，我们要选取新的工具来完成进程通信任务。

使用multiprocessing.Queue

使用逻辑是和多线程中的Queue是一样的，详细方法。这种通信方式不能用在通过Pool进程池创建的进程中

import multiprocessing
import time
def plus(queue):
 for i in range(6):
 num = queue.get() + 1
 queue.put(num)
 print(num)
 time.sleep(1)
def subtract(queue):
 for i in range(6):
 num = queue.get() - 1
 queue.put(num)
 print(num)
 time.sleep(2)
if __name__ == '__main__':
 queue = multiprocessing.Queue(1)
 queue.put(0)
 plus_process = multiprocessing.Process(target=plus, args=(queue,))
 subtract_process = multiprocessing.Process(target=subtract, args=(queue,))
 plus_process.start()
 subtract_process.start()
 
# result:
# 1
# 1
# 2
# 2
# 3
# 3
# 0
# 1
# 2
# 2
# 1
# 0

使用Manager()中的Queue

Manager()会返回一个在进程间进行同步管理的一个对象，它提供了多种在进程间共享数据的形式。

import multiprocessing
import time
def plus(queue):
 for i in range(6):
 num = queue.get() + 1
 queue.put(num)
 print(num)
 time.sleep(1)
def subtract(queue):
 for i in range(6):
 num = queue.get() - 1
 queue.put(num)
 print(num)
 time.sleep(2)
if __name__ == '__main__':
 queue = multiprocessing.Manager().Queue(1) # 创建方式有些奇特
 # queue = multiprocessing.Queue() # 这时用这个就行不通了
 pool = multiprocessing.Pool(2)
 queue.put(0)
 pool.apply_async(plus, args=(queue,))
 pool.apply_async(subtract, args=(queue,))
 pool.close()
 pool.join()
 
# result:
# 0
# 1
# 1
# 2
# 2
# 3
# -1
# 0
# 1
# 2
# 1
# 0

使用Manager()中的list()

多个进程可以共享全局的list，因为是进程间共享，所以用锁的机制保证它的安全性。这里的Manager().Lock不是前面线程级别的Lock，它可以保证进程间的同步。

import multiprocessing as mp
import time
def add_person(waiting_list, name_list, lock):
 lock.acquire()
 for name in name_list:
 waiting_list.append(name)
 time.sleep(1)
 print(waiting_list)
 lock.release()
def get_person(waiting_list, lock):
 lock.acquire()
 if waiting_list:
 name = waiting_list.pop(0)
 print("get {0}".format(name))
 lock.release()
if __name__ == '__main__':
 waiting_list = mp.Manager().list()
 lock = mp.Manager().Lock() # 使用 lock 限制进程对全局量的访问
 name_list = ["MetaTian", "Rity", "Anonymous"]
 add_process = mp.Process(target=add_person, args=(waiting_list, name_list, lock))
 get_process = mp.Process(target=get_person, args=(waiting_list, lock))
 add_process.start()
 get_process.start()
 add_process.join()
 get_process.join()
 print(waiting_list)
 
# result:
# ['MetaTian']
# ['MetaTian', 'Rity']
# ['MetaTian', 'Rity', 'Anonymous']
# get MetaTian
# ['Rity', 'Anonymous']

Manager()中还有更多的进程间通信的工具，可以参阅官方文档。

使用Pipe

Pipe只能适用于两个进程间的通信，它的性能高于Queue，Pipe()会返回两个Connection对象，使用这个对象可以在进程间进行数据的发送和接收，非常像前面讲过的socket对象。关于Connection

import multiprocessing
def plus(conn):
 default_num = 0
 for i in range(3):
 num = 0 if i == 0 else conn.recv()
 conn.send(num + 1)
 print("plus send: {0}".format(num+1))
def subtract(conn):
 for i in range(3):
 num = conn.recv()
 conn.send(num-1)
 print("subtract send: {0}".format(num-1))
if __name__ == '__main__':
 conn_plus, conn_sbtract = multiprocessing.Pipe()
 plus_process = multiprocessing.Process(target=plus, args=(conn_plus,))
 subtract_process = multiprocessing.Process(target=subtract, args=(conn_sbtract,))
 plus_process.start()
 subtract_process.start()
# result:
# plus send: 1
# subtract send: 0
# plus send: 1
# subtract send: 0
# plus send: 1
# subtract send: 0

send()可以连续发送数据，recv()将另一端发送的数据陆续取出，如果没有取到数据，则进入等待状态。

注：喜欢python + qun：839383765 可以获取Python各类免费最新入门学习资料！