爬虫性能:NodeJs VS Python

前言

早就听说Nodejs的异步策略是多么的好,I/O是多么的牛逼……反正就是各种好。今天我就准备给nodejs和python来做个比较。能体现异步策略和I/O优势的项目,我觉得莫过于爬虫了。那么就以一个爬虫项目来一较高下吧。

爬虫项目

众筹网-众筹中项目 http://www.zhongchou.com/brow…,我们就以这个网站为例,我们爬取它所有目前正在众筹中的项目,获得每一个项目详情页的URL,存入txt文件中。

实战比较

python原始版

# -*- coding:utf-8 -*-
'''
Created on 20160827
@author: qiukang
'''
import requests,time
from BeautifulSoup import BeautifulSoup    # HTML

#请求头
headers = {
   'Accept':'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
   'Accept-Encoding':'gzip, deflate, sdch',
   'Accept-Language':'zh-CN,zh;q=0.8',
   'Connection':'keep-alive',
   'Host':'www.zhongchou.com',
   'Upgrade-Insecure-Requests':1,
   'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/49.0.2593.0 Safari/537.36'
}

# 获得项目url列表
def getItems(allpage):
    no = 0
    items = open('pystandard.txt','a')
    for page in range(allpage):
        if page==0:
            url = 'http://www.zhongchou.com/browse/di'
        else:
            url = 'http://www.zhongchou.com/browse/di-p'+str(page+1)
        # print url #①
        r1 = requests.get(url,headers=headers)
        html = r1.text.encode('utf8')
        soup = BeautifulSoup(html);
        lists = soup.findAll(attrs={"class":"ssCardItem"})
        for i in range(len(lists)):
            href = lists[i].a['href']
            items.write(href+"\n")
            no +=1
    items.close()
    return no
    
if __name__ == '__main__':
    start = time.clock()
    allpage = 30
    no = getItems(allpage)
    end = time.clock()
    print('it takes %s Seconds to get %s items '%(end-start,no))

实验5次的结果:

 it takes 48.1727159614 Seconds to get 720 items 
 it takes 45.3397999415 Seconds to get 720 items  
 it takes 44.4811429862 Seconds to get 720 items 
 it takes 44.4619293082 Seconds to get 720 items
 it takes 46.669706593 Seconds to get 720 items 

python多线程版

# -*- coding:utf-8 -*-
'''
Created on 20160827
@author: qiukang
'''
import requests,time,threading
from BeautifulSoup import BeautifulSoup    # HTML

#请求头
headers = {
   'Accept':'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
   'Accept-Encoding':'gzip, deflate, sdch',
   'Accept-Language':'zh-CN,zh;q=0.8',
   'Connection':'keep-alive',
   'Host':'www.zhongchou.com',
   'Upgrade-Insecure-Requests':1,
   'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/49.0.2593.0 Safari/537.36'
}

items = open('pymulti.txt','a')
no = 0
lock = threading.Lock()

# 获得项目url列表
def getItems(urllist):
    # print urllist  #①
    global items,no,lock
    for url in urllist:
        r1 = requests.get(url,headers=headers)
        html = r1.text.encode('utf8')
        soup = BeautifulSoup(html);
        lists = soup.findAll(attrs={"class":"ssCardItem"})
        for i in range(len(lists)):
            href = lists[i].a['href']
            lock.acquire()
            items.write(href+"\n")
            no +=1
            # print no
            lock.release()
    
if __name__ == '__main__':
    start = time.clock()
    allpage = 30
    allthread = 30
    per = (int)(allpage/allthread)
    urllist = []
    ths = []
    for page in range(allpage):
        if page==0:
            url = 'http://www.zhongchou.com/browse/di'
        else:
            url = 'http://www.zhongchou.com/browse/di-p'+str(page+1)
        urllist.append(url)
    for i in range(allthread):
        # print urllist[i*(per):(i+1)*(per)]
        th = threading.Thread(target = getItems,args= (urllist[i*(per):(i+1)*(per)],))
        th.start()
        th.join()
    items.close()
    end = time.clock()
    print('it takes %s Seconds to get %s items '%(end-start,no))
    

实验5次的结果:

it takes 45.5222291114 Seconds to get 720 items 
it takes 46.7097831417 Seconds to get 720 items
it takes 45.5334646156 Seconds to get 720 items 
it takes 48.0242797553 Seconds to get 720 items
it takes 44.804855018 Seconds to get 720 items  

这个多线程并没有优势,经过 #① 的注释与否发现,这个所谓的多线程也是按照单线程运行的。

python改进

单线程

首先我们把解析html的步骤改进一下,分析发现

lists = soup.findAll('a',attrs={"class":"siteCardICH3"})

lists = soup.findAll(attrs={"class":"ssCardItem"})

更好,因为它是直接找 a ,而不是先找 div 再找 div 下的 a
改进后实验5次结果如下,可见有进步:

it takes 41.0018861912 Seconds to get 720 items 
it takes 42.0260390497 Seconds to get 720 items
it takes 42.249635988 Seconds to get 720 items 
it takes 41.295524133 Seconds to get 720 items 
it takes 42.9022894154 Seconds to get 720 items 

多线程

修改 getItems(urllist)getItems(urllist,thno)
函数起止加入 print thno," begin at",time.clock()print thno," end at",time.clock()。结果:

0  begin at 0.00100631078628
0  end at 1.28625832936
1  begin at 1.28703230691
1  end at 2.61739476075
2  begin at 2.61801291642
2  end at 3.92514717937
3  begin at 3.9255829208
3  end at 5.38870235361
4  begin at 5.38921134066
4  end at 6.670658786
5  begin at 6.67125734731
5  end at 8.01520989534
6  begin at 8.01566383155
6  end at 9.42006780585
7  begin at 9.42053340537
7  end at 11.0386755513
8  begin at 11.0391565464
8  end at 12.421359168
9  begin at 12.4218294329
9  end at 13.9932716671
10  begin at 13.9939957256
10  end at 15.3535799145
11  begin at 15.3540870354
11  end at 16.6968289314
12  begin at 16.6972665389
12  end at 17.9798803157
13  begin at 17.9804714125
13  end at 19.326706238
14  begin at 19.3271438455
14  end at 20.8744308886
15  begin at 20.8751017624
15  end at 22.5306500245
16  begin at 22.5311450156
16  end at 23.7781693541
17  begin at 23.7787245279
17  end at 25.1775114499
18  begin at 25.178350742
18  end at 26.5497330734
19  begin at 26.5501776789
19  end at 27.970799259
20  begin at 27.9712727895
20  end at 29.4595075375
21  begin at 29.4599959972
21  end at 30.9507299602
22  begin at 30.9513989679
22  end at 32.2762763982
23  begin at 32.2767182045
23  end at 33.6476256057
24  begin at 33.648137392
24  end at 35.1100517711
25  begin at 35.1104907783
25  end at 36.462657099
26  begin at 36.4632234696
26  end at 37.7908515759
27  begin at 37.7912845182
27  end at 39.4359928956
28  begin at 39.436448698
28  end at 40.9955021593
29  begin at 40.9960871912
29  end at 42.6425665264
it takes 42.6435882327 Seconds to get 720 items 

可见这些线程是真的没有并发执行,而是顺序执行的,并没有达到多线程的目的。问题在哪里呢?原来
我的循环中

th.start()
th.join()

两行代码是紧接着的,所以新的线程会等待上一个线程执行完毕才会start,修改为

for i in range(allthread):
    # print urllist[i*(per):(i+1)*(per)]
    th = threading.Thread(target = getItems,args= (urllist[i*(per):(i+1)*(per)],i))
    ths.append(th)
for th in ths:
    th.start()
for th in ths:
    th.join()

结果:

0  begin at 0.0010814225325
1  begin at 0.00135201143191
2  begin at 0.00191744892518
3  begin at 0.0021311208492
4  begin at 0.00247495536449
5  begin at 0.0027334144167
6  begin at 0.00320601192551
7  begin at 0.00379011072218
8  begin at 0.00425431064445
9  begin at 0.00511692939449
10  begin at 0.0132038052264
11  begin at 0.0165926979253
12  begin at 0.0170886220634
13  begin at 0.0174665134574
14  begin at 0.018348726576
15  begin at 0.0189780790334
16  begin at 0.0201896641572
17  begin at 0.0220576606283
18  begin at 0.0231484138125
19  begin at 0.0238804034387
20  begin at 0.0273901280772
21  begin at 0.0300363009005
22  begin at 0.0362878375422
23  begin at 0.0395512329756
24  begin at 0.0431556637289
25  begin at 0.0459581249682
26  begin at 0.0482254733323
27  begin at 0.0535430117384
28  begin at 0.0584971212607
29  begin at 0.0598136762161
16  end at 65.2657542222
24  end at 66.2951247811
21  end at 66.3849747583
4  end at 66.6230160119
5  end at 67.5501632164
29  end at 67.7516992283
23  end at 68.6985322418
7  end at 69.1060433231
22  end at 69.2743398214
2  end at 69.5523713152
14  end at 69.6454986837
15  end at 69.8333400981
12  end at 69.9508018062
10  end at 70.2860348602
26  end at 70.3670659719
13  end at 70.3847232972
27  end at 70.3941635841
11  end at 70.5132838156
1  end at 70.7272351926
0  end at 70.9115253609
6  end at 71.0876563409
8  end at 71.112480539825
  end at 71.1145248855
3  end at 71.4606034226
19  end at 71.6103622486
18  end at 71.6674453096
20  end at 71.725601862
17  end at 71.7778992318
9  end at 71.7847479301
28  end at 71.7921004837
it takes 71.7931912368 Seconds to get 720 items 

反思

上面的的多线是并发了,可是比单线程运行时间长了太多……我还没找出来原因,猜想是不是beautifulsoup不支持多线程?请各位多多指教。为了验证这个想法,我准备不用beautifulsoup,直接使用字符串查找。首先还是从单线程的修改:

# -*- coding:utf-8 -*-
'''
Created on 20160827
@author: qiukang
'''
import requests,time
from BeautifulSoup import BeautifulSoup    # HTML

#请求头
headers = {
   'Accept':'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
   'Accept-Encoding':'gzip, deflate, sdch',
   'Accept-Language':'zh-CN,zh;q=0.8',
   'Connection':'keep-alive',
   'Host':'www.zhongchou.com',
   'Upgrade-Insecure-Requests':'1',
   'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/49.0.2593.0 Safari/537.36'
}

# 获得项目url列表
def getItems(allpage):
    no = 0
    data = set()
    for page in range(allpage):
        if page==0:
            url = 'http://www.zhongchou.com/browse/di'
        else:
            url = 'http://www.zhongchou.com/browse/di-p'+str(page+1)
        # print url #①
        r1 = requests.get(url,headers=headers)
        html = r1.text.encode('utf8')
        start = 5000    
        while  True:     
            index = html.find("deal-show", start)   
            if index == -1:     
                break    
            # print "http://www.zhongchou.com/deal-show/"+html[index+10:index+19]+"\n"
            # time.sleep(100)
            data.add("http://www.zhongchou.com/deal-show/"+html[index+10:index+19]+"\n")    
            start = index  + 1000 
    items = open('pystandard.txt','a')
    items.write("".join(data))
    items.close()
    return len(data)
    
if __name__ == '__main__':
    start = time.clock()
    allpage = 30
    no = getItems(allpage)
    end = time.clock()
    print('it takes %s Seconds to get %s items '%(end-start,no))

实验3次,结果:

it takes 11.6800132309 Seconds to get 720 items
it takes 11.3621804427 Seconds to get 720 items
it takes 11.6811991567 Seconds to get 720 items  

然后对多线程进行修改:

# -*- coding:utf-8 -*-
'''
Created on 20160827
@author: qiukang
'''
import requests,time,threading
from BeautifulSoup import BeautifulSoup    # HTML

#请求头
header = {
   'Accept':'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
   'Accept-Encoding':'gzip, deflate, sdch',
   'Accept-Language':'zh-CN,zh;q=0.8',
   'Connection':'keep-alive',
   'Host':'www.zhongchou.com',
   'Upgrade-Insecure-Requests':'1',
   'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/49.0.2593.0 Safari/537.36'
}

data = set()
no = 0
lock = threading.Lock()

# 获得项目url列表 
def getItems(urllist,thno):
    # print urllist
    # print thno," begin at",time.clock()
    global no,lock,data
    for url in urllist:
        r1 = requests.get(url,headers=header)
        html = r1.text.encode('utf8')
        start = 5000    
        while  True:     
            index = html.find("deal-show", start)   
            if index == -1:     
                break
            lock.acquire()
            data.add("http://www.zhongchou.com/deal-show/"+html[index+10:index+19]+"\n")    
            start = index  + 1000 
            lock.release()
        
    # print thno," end at",time.clock()
    
if __name__ == '__main__':
    start = time.clock()
    allpage = 30  #页数
    allthread = 10 #线程数
    per = (int)(allpage/allthread)
    urllist = []
    ths = []
    for page in range(allpage):
        if page==0:
            url = 'http://www.zhongchou.com/browse/di'
        else:
            url = 'http://www.zhongchou.com/browse/di-p'+str(page+1)
        urllist.append(url)
    for i in range(allthread):
        # print urllist[i*(per):(i+1)*(per)]
        low = i*allpage/allthread#注意写法
        high = (i+1)*allpage/allthread
        # print low,' ',high
        th = threading.Thread(target = getItems,args= (urllist[low:high],i))
        ths.append(th)
    for th in ths:
        th.start()
    for th in ths:
        th.join()
    items = open('pymulti.txt','a')
    items.write("".join(data))
    items.close()
    end = time.clock()
    print('it takes %s Seconds to get %s items '%(end-start,len(data)))

实验3次,结果:

it takes 1.4781525123 Seconds to get 720 items 
it takes 1.44905954029 Seconds to get 720 items
it takes 1.49297891786 Seconds to get 720 items

可见多线程确实比单线程快好多倍。对于简单的爬取任务而言,用字符串的内置方法比用beautifulsoup解析html快很多。

NodeJs

// npm install request -g #貌似不行,要进入代码所在目录:npm install --save request
// npm install cheerio -g  #npm install --save cheerio

var request = require("request");
var cheerio = require('cheerio');
var fs = require('fs');

var t1 = new Date().getTime();
var allpage = 30;
var urllist = new Array()  
var urldata = "";
var mark = 0;
var no = 0;
for (var i=0; i<allpage; i++) {
    if (i==0) 
        urllist[i] = 'http://www.zhongchou.com/browse/di'
    else
        urllist[i] = 'http://www.zhongchou.com/browse/di-p'+(i+1).toString();
    request(urllist[i],function(error,resp,body){
        if (!error && resp.statusCode==200) {
            getUrl(body);
        }
    });
} 

function getUrl(data) {
    var $ = cheerio.load(data);  //cheerio解析data
    var href = $("a.siteCardICH3").toArray();
    for (var i = href.length - 1; i >= 0; i--) {
        // console.log(href[i].attribs["href"]);
        urldata += (href[i].attribs["href"]+"\n");
        no += 1;
    }    
    mark += 1;
    if (mark==allpage) {
        // console.log(urldata);
        fs.writeFile('./nodestandard.txt',urldata,function(err){
                    if(err) throw err;
        });
        var t2 = new Date().getTime();
        console.log("it takes " + ((t2-t1)/1000).toString() + " Seconds to get " + no.toString() + " items");
    }  
}

实验5次的结果:

it takes 3.949 Seconds to get 720 items
it takes 3.642 Seconds to get 720 items
it takes 3.641 Seconds to get 720 items
it takes 3.938 Seconds to get 720 items
it takes 3.783 Seconds to get 720 items

可见同样是用解析html的方法,nodejs速度完虐python。字符串查找呢?

var request = require("request");
var cheerio = require('cheerio');
var fs = require('fs');

var t1 = new Date().getTime();
var allpage = 30;
var urllist = new Array()  ;
var urldata = new Array();
var mark = 0;
var no = 0;
for (var i=0; i<allpage; i++) {
    if (i==0) 
        urllist[i] = 'http://www.zhongchou.com/browse/di'
    else
        urllist[i] = 'http://www.zhongchou.com/browse/di-p'+(i+1).toString();
    // console.log(urllist[i]);
    request(urllist[i],function(error,resp,body){
        if (!error && resp.statusCode==200) {
            getUrl(body);
        }
    });
} 

function getUrl(data) {
    mark += 1;
    var start = 5000
    while (true) {
        var index1 = data.indexOf("deal-show", start);
        if (index1 == -1)     
            break;
        var url = "http://www.zhongchou.com/deal-show/"+data.substring(index1+10,index1+19)+"\n";
        // console.log(url);
        if (urldata.indexOf(url)==-1) {
            urldata.push(url);
        }
        start = index1 + 1000;
    }
    if (mark==allpage) {//所有页面执行完毕
        // console.log(urldata);
        no = urldata.length;
        fs.writeFile('./nodestandard.txt',urldata.join(""),function(err){
                    if(err) throw err;
        });
        var t2 = new Date().getTime();
        console.log("it takes " + ((t2-t1)/1000).toString() + " Seconds to get " + no.toString() + " items");
    }  
}

实验5次的结果:

it takes 3.695 Seconds to get 720 items
it takes 3.781 Seconds to get 720 items
it takes 3.94 Seconds to get 720 items
it takes 3.705 Seconds to get 720 items
it takes 3.601 Seconds to get 720 items

可见和解析起来的时间是差不多的。

综上

由我自己了解的知识和本实验而言,我的结论是:python用上多线程下载速度能够比过nodejs,但是解析网页这种事python没有nodejs快,毕竟js原生就是为了写网页,而且复杂的爬虫总不能都用字符串去找吧。

2016.9.13-补充

  1. 评论中提到的time.time(),感谢老司机指出我的错误,我在python多线程,字符串查找版本中使用了,实验3次过后依然是快于nodejs版本的平均用时2.3S,不知道是不是您和我的网络环境不一样导致?我准备换个教室试试……至于有没有误导人,我想读者会自己去尝试,得出自己的结论。

  2. Python的确有异步(twisted),nodejs也的确有多进程(child_process),我想追求极致的性能比较还需要对这两种语言有更深入的研究,这个我目前也是半知不解,我会尽快花时间了解,争取实现比较(这里不是追求编程方法的比较,就是单纯的想比较在同一台机器同一个网络下,两种语言能做到的极致。道阻且长啊。)

  3. 还有解析方法,我这里用的是python自带的解析,官网说lxml的确比自带的快,但是我这里换了过后多线程依然没有体现出来优势,所以我还是很疑惑,是不是beautifulsoup不支持多线程?,我在官网没找到相关文档,请各位指教。另外from BeautifulSoup import BeautifulSoup的确是比from bs4 import BeautifulSoup 慢多了,这是BeautifulSoup的版本原因,感谢评论者指出。

    原文作者:MageekChiu
    原文地址: https://segmentfault.com/a/1190000006741610
    本文转自网络文章,转载此文章仅为分享知识,如有侵权,请联系博主进行删除。
点赞