【模块】：Concurrent

concurrent 模块

回顾：

　　对于python来说，作为解释型语言，Python的解释器必须做到既安全又高效。我们都知道多线程编程会遇到的问题，解释器要留意的是避免在不同的线程操作内部共享的数据，同时它还要保证在管理用户线程时保证总是有最大化的计算资源。而python是通过使用全局解释器锁来保护数据的安全性：

　　python代码的执行由python虚拟机来控制，即Python先把代码（.py文件）编译成字节码（字节码在Python虚拟机程序里对应的是PyCodeObject对象，.pyc文件是字节码在磁盘上的表现形式），交给字节码虚拟机，然后虚拟机一条一条执行字节码指令，从而完成程序的执行。python在设计的时候在虚拟机中，同时只能有一个线程执行。同样地，虽然python解释器中可以运行多个线程，但在任意时刻，只有一个线程在解释器中运行。而对python虚拟机的访问由全局解释器锁来控制，正是这个锁能保证同一时刻只有一个线程在运行

多线程执行方式：

• 设置GIL(global interpreter lock).
• 切换到一个线程执行。
• 运行：
•     a，指定数量的字节码指令。
•     b，线程主动让出控制（可以调用time.sleep(0)）。
• 把线程设置为睡眠状态。
• 解锁GIL.
• 再次重复以上步骤。

　　GIL的特性，也就导致了python不能充分利用多核cpu。而对面向I/O的（会调用内建操作系统C代码的）程序来说，GIL会在这个I/O调用之前被释放，以允许其他线程在这个线程等待I/O的时候运行。如果线程并为使用很多I/O操作，它会在自己的时间片一直占用处理器和GIL。这也就是所说的：I/O密集型python程序比计算密集型的程序更能充分利用多线程的好处。

总之，不要使用python多线程，使用python多进程进行并发编程，就不会有GIL这种问题存在，并且也能充分利用多核cpu

 

threading使用回顾：

import threading
import time

def run(n):
    semaphore.acquire()
    time.sleep(2)
    print("run the thread: %s" % n)
    semaphore.release()

if __name__ == '__main__':
    start_time = time.time()
    thread_list = []
    semaphore = threading.BoundedSemaphore(5)  # 信号量，最多允许5个线程同时运行
    for i in range(20):
        t = threading.Thread(target=run, args=(i,))
        t.start()
        thread_list.append(t)
    for t in thread_list:
        t.join()

    used_time = time.time() - start_time
    print('用时',used_time)

# 用时 8.04102110862732

　　

ThreadPoolExecutor多并发：

1、submit

import time
from concurrent import futures


def run(n):
    time.sleep(2)
    print("run the thread: %s" % n)

if __name__ == '__main__':
    start = time.time()
    with futures.ThreadPoolExecutor(5) as executor:
        for i in range(20):
            executor.submit(run,i)     

    print(time.time()-start)

# 8.006775379180908

 2、map

import time
from concurrent import futures


def run(n):
    time.sleep(2)
    print("run the thread: %s" % n)

if __name__ == '__main__':
    start = time.time()
    with futures.ThreadPoolExecutor(5) as executor:
        executor.map(run,range(20))

    print(time.time()-start)

# 8.006775379180908　

executor.submit 和 futures.as_completed 这个组合比executor.map 更灵活，因为 submit 方法能处理不同的可调用对象和参数，而 executor.map 只能处理参数不同的同一个可调用对象。此外，传给 futures.as_completed 函数的期物集合可以来自多个 Executor 实例，例如一些由 ThreadPoolExecutor 实例创建，另一些由ProcessPoolExecutor创建

ProcessPoolExecutor多并发：

1、submit

import time
from concurrent import futures

import time
from concurrent import futures


def run(n):
    time.sleep(2)
    print("run the thread: %s" % n)


if __name__ == '__main__':
    start = time.time()
    with futures.ProcessPoolExecutor(5) as executor:
        for i in range(20):
            executor.submit(run, i)

    print(time.time() - start)

# 8.365714311599731

2、map

import time
from concurrent import futures

import time
from concurrent import futures


def run(n):
    time.sleep(2)
    print("run the thread: %s" % n)


if __name__ == '__main__':
    start = time.time()
    with futures.ProcessPoolExecutor(5) as executor:
        executor.map(run, range(20))

    print(time.time() - start)

# 8.317736864089966

接口压力测试的脚本

# #!/usr/bin/env python
# # -*- coding:utf-8 -*-

import os
import time
import logging
import requests
import threading
from multiprocessing import Lock,Manager
from concurrent import futures


download_url = 'http://192.168.188.105:8888'
workers = 250
cpu_count = 4

session = requests.Session()

def handle(cost,mutex,contain):
    with mutex:
        min_cost = contain['min_cost']
        max_cost = contain['max_cost']
        hit_count = contain['hit_count']
        average_cost = contain['average_cost']
        if min_cost == 0:
            contain['min_cost'] = cost
        if min_cost > cost:
            contain['min_cost'] = cost
        if max_cost < cost:
            contain['max_cost'] = cost
        average_cost = (average_cost*hit_count + cost) / (hit_count + 1)
        hit_count +=1
        contain['average_cost'] = average_cost
        contain['hit_count'] = hit_count
    logging.info(contain)

def download_one(mutex,contain):
    while True:
        try:
            stime = time.time()
            request = requests.Request(method='GET', url=download_url,)
            prep = session.prepare_request(request)
            response = session.send(prep, timeout=50)
            etime = time.time()
            print(response.status_code)
            logging.info('process[%s] thread[%s] status[%s] cost[%s]',os.getpid(),threading.current_thread().ident,
                         response.status_code,etime-stime)
            handle(float(etime-stime),mutex,contain)
            # time.sleep(1)
        except Exception as e:
            logging.error(e)
            print(e)

def new_thread_pool(mutex,contain):
    with futures.ThreadPoolExecutor(workers) as executor:
        for i in range(workers):
            executor.submit(download_one,mutex,contain)

def subprocess():
    manager = Manager()
    mutex = manager.Lock()
    contain = manager.dict({'average_cost': 0, 'min_cost': 0, 'max_cost': 0, 'hit_count': 0})

    with futures.ProcessPoolExecutor(cpu_count) as executor:
        for i in range(cpu_count):
            executor.submit(new_thread_pool,mutex,contain)

if __name__ == '__main__':
    logging.basicConfig(filename="client.log", level=logging.INFO,
                        format="%(asctime)s  [%(filename)s:%(lineno)d] %(message)s", datefmt="%m/%d/%Y %H:%M:%S [%A]")
    subprocess()