能看懂的aiohttp(简单版)

asyncio可以实现单线程并发IO操作。如果仅用在客户端，发挥的威力不大。如果把asyncio用在服务器端，例如Web服务器，由于HTTP连接就是IO操作，因此可以用单线程+coroutine实现多用户的高并发支持。 
asyncio实现了TCP、UDP、SSL等协议，aiohttp则是基于asyncio实现的HTTP框架。

使用协程的异步请求以其低时消耗和对硬件的高利用而著称，翻看了很多论坛，发现协程在进行爬虫以及高频网络请求时的耗时比单多进程和单多线程还要好。本文将使用requests和使用aiohttp+asyncio进行比较，比较一下具体使用协程和不使用协程能差距多少。

本文测试所使用目标网址是廖雪峰老师python3教程的评论页 目前一共有2318页。

1.使用requests获取单个网址

import requests,time

r=requests.get(url="https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page=1")
print(len(r.text)) #结果 40007 用时0.3s

2.使用 aiohttp+asyncio 获取单个网址
import aiohttp, asyncio

async def fn():
async with aiohttp.get(url='https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page=1') as resp:
text=await resp.text()
print(len(text))

loop = asyncio.get_event_loop()
loop.run_until_complete(fn()) #结果 40007 用时0.3s

看样对于单个请求来说，阻塞还是不阻塞没有实际意义，因为最终结果都是要等待网络访问，所以实际上对于整个需求来说，都是一个阻塞模型。而且async的方式代码量相对大。
3.使用requests获取50个网址

import requests

a=[len(requests.get(url="https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page={}".format(i)).text) for i in range(1,51)]
print(a) #结果 [40007, 40444, 40367, 40820, 40534, 40505, 40735, 40454, 40768, 40636, 40600, 40888, 41277, 41390, 41222, 40899, 40853, 40616, 40654, 40870, 41249, 40840, 40782, 41326, 41136, 40511, 40504, 40609, 41038, 41054, 40486, 40556, 41083, 40975, 40861, 40877, 40166, 40899, 40598, 40920, 40902, 40994, 40735, 40714, 41064, 40719, 40991, 40748, 40652, 40799]
用时：5.3s

4.使用aiohttp+asyncio获取50个网址

import aiohttp, asyncio

async def fn(num):
async with aiohttp.get(url='https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page={}'.format(num)) as resp:
text=await resp.text()
result.append(len(text))

result=[]
loop = asyncio.get_event_loop()
tasks=[asyncio.ensure_future(fn(i)) for i in range(1,51)]
loop.run_until_complete(asyncio.wait(tasks))
print(result)
#结果：[40636, 41326, 40853, 41277, 41249, 40735, 40616, 40454, 40888, 40899, 40007, 40768, 40486, 40870, 40820, 40444, 40367, 41136, 40609, 40975, 40504, 40166, 40920, 40598, 40556, 40652, 41083, 40735, 40799, 40899, 40902, 40748, 41064, 40505, 40654, 40511, 40600, 40534, 41390, 40782, 41222, 40840, 41038, 40714, 41054, 40877, 40719, 40991, 40861, 40994]
#用时：3.4s

有趣的是，协程最后返回的list的顺序和requests的完全顺序不一样，多试几次就发现，协程每次返回的顺序都不一样，因为它是异步的，不一定这些网络请求任务中哪个先完成。而requests每次得到的都是相同顺序的list。同时，服务器响应越慢，协程的优势体现的越明显，节省了很多网络等待的时间，而requests会阻塞的更久。
5.发送2318个请求

既然页面总共有2318个，我们不如全部使用爬一下，老方法requests:

import requests
a=[len(requests.get(url="https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page={}".format(i)).text) for i in range(1,2319)]
print(a)
#用时：497s

aiohttp+asyncio:
由于同时并发2318个请求并交给网卡去处理会出现负荷过重，aiohttp默认同时最大支持1024个协程的进行，但是考虑到硬件的承受能力，我们采用维护一个“协程池”的方法。使用 asyncio.Semaphore来控制同时进行的最大IO量。每当有一个协程完成时，便加入一个新的协程：

import aiohttp, asyncio

async def fn(num,sem):
async with sem:
async with aiohttp.get(url="https://www.liaoxuefeng.com/discuss/001409195742008d822b26cf3de46aea14f2b7378a1ba91000?page={}".format(num)) as resp:
text=await resp.text()
result.append(len(text))

loop=asyncio.get_event_loop()
result=[]
sem=asyncio.Semaphore(100) #维持100个信号量
tasks=[ asyncio.ensure_future(fn(i,sem)) for i in range(1,2319)]
loop.run_until_complete(asyncio.wait(tasks))
print(result)
#用时 预计200s左右

在用async测试的过程中，可能并发的请求服务器压力太大，廖雪峰老师的网站出现了504网关超时报错，第二天测试的时候，刚进行到一半服务器又崩溃了，服务器应该没做优化，这样一点并发流量怎么就垮了。

参考:https://blog.csdn.net/getcomputerstyle/article/details/78438246

　　https://blog.csdn.net/getcomputerstyle/article/details/71515331

上面是粘贴复制的  里面代码有问题   

async def fn(num):
    async with aiohttp.request('get',
            url='http://www.ccgp-beijing.gov.cn/xxgg/sjzfcggg/sjzbjggg/index_%s.html'% num) as resp:
        

 第一种写法

import aiohttp, asyncio
from lxml import etree
import requests
import time
import os
import hashlib
from apscheduler.schedulers.blocking import BlockingScheduler  #定时任务
headers = {"User-Agent": "Mozilla/5.0"}
pat=os.path.dirname(os.path.abspath(__file__))
ll=['技术研发']
async def main(pool):  # 启动
    sem = asyncio.Semaphore(pool)
    async with aiohttp.ClientSession() as session:  # 给所有的请求，创建同一个session
        tasks = []
        for i in range(1,70):
            url=control_sem(sem,'http://www.ccgp-beijing.gov.cn/xxgg/sjzfcggg/sjzbjggg/index_%s.html'% i, session)
            tasks.append(url)
        await asyncio.wait(tasks)

async def control_sem(sem, url, session):  # 限制信号量
    async with sem:
        await fetch(url, session)

async def fetch(url, session):  # 开启异步请求
    async with session.get(url) as resp:
        page= await resp.text()
        tree = etree.HTML(page)
        ul_list = tree.xpath('//ul[@class="xinxi_ul"]/li')
        for ul in ul_list:
            name = ul.xpath('./a/text()')[0].strip()
            print(name)
            for ii in ll:
                if ii in name:
                    new_url = 'http://www.ccgp-beijing.gov.cn/xxgg/sjzfcggg/sjzbjggg/' + 
                              ul.xpath('./a/@href')[0].split('/')[-1]
                    datatime = ul.xpath('./span/text()')[0].strip()
                    pag = requests.get(url=new_url, headers=headers).text
                    source = name + datatime
                    source_id = hashlib.md5(source.encode()).hexdigest()  # 设置唯一id 做去重
                    filename = os.path.join(pat, 'a', '%s.html') % source_id
                    with open(filename, 'w')as f:
                        f.write(pag)
                    print( name, datatime, new_url,source)

def start():
    s = time.time()
    # loop = asyncio.get_event_loop()  #加win定时就换下面2行代码
    loop = asyncio.new_event_loop()
    asyncio.set_event_loop(loop)
    loop.run_until_complete(main(pool=5))
    e = time.time() - s
    print('>>>>>>>>>>>>>>>>>>',e)

if __name__ == '__main__':
    print('等待中......')
    scheduler = BlockingScheduler()
    scheduler.add_job(start, 'cron', hour=16, minute=43)  # 'interval', seconds=40
    try:
        scheduler.start()
    except (KeyboardInterrupt, SystemExit):
        pass
    except Exception as e:
        print(e)




第二种写法

import aiohttp, asyncio
from conf.setting import *
from lxml import etree
import requests
import hashlib
import time

async def fn(num, sem):
    async with sem:
        async with aiohttp.request('get',url='http://www.ccgp-beijing.gov.cn/xxgg/sjzfcggg/sjzbjggg/index_%s.html'% num,)as resp:
            page = await resp.text()
            tree = etree.HTML(page)
            ul_list = tree.xpath('//ul[@class="xinxi_ul"]/li')
            for ul in ul_list:
                name = ul.xpath('./a/text()')[0].strip()
                print(name)
                for ii in ll:
                    if ii in name:
                        new_url = 'http://www.ccgp-beijing.gov.cn/xxgg/sjzfcggg/sjzbjggg/' + 
                                  ul.xpath('./a/@href')[0].split('/')[-1]
                        datatime = ul.xpath('./span/text()')[0].strip()
                        pag = requests.get(url=new_url, headers=headers).text
                        source = name + datatime
                        source_id = hashlib.md5(source.encode()).hexdigest()  # 设置唯一id 做去重
                        pat = os.path.dirname(os.path.abspath(__file__))
                        filename = os.path.join(pat, 'a', '%s.html') % source_id
                        with open(filename, 'w')as f:
                            f.write(pag)
                        print(name, datatime, new_url, source)





loop = asyncio.get_event_loop()
s=time.time()
sem = asyncio.Semaphore(100)  # 维持100个信号量
tasks = [asyncio.ensure_future(fn(i, sem)) for i in range(1, 143)]
loop.run_until_complete(asyncio.wait(tasks))
e=time.time()-s

print('>>>>>>>>>>>>>>>>>>>>>',e)