标签:
最近,一直在做网络爬虫相关的东西。 看了一下开源C++写的larbin爬虫,仔细阅读了里面的设计思想和一些关键技术的实现。
1、larbin的URL去重用的很高效的bloom filter算法;
2、DNS处理,使用的adns异步的开源组件;
3、对于url队列的处理,则是用部分缓存到内存,部分写入文件的策略。
4、larbin对文件的相关操作做了很多工作
5、在larbin里有连接池,通过创建套接字,向目标站点发送HTTP协议中GET方法,获取内容,再解析header之类的东西
6、大量描述字,通过poll方法进行I/O复用,很高效
7、larbin可配置性很强
8、作者所使用的大量数据结构都是自己从最底层写起的,基本没用STL之类的东西
......
还有很多,以后有时间在好好写篇文章,总结下。
这两天,用python写了个多线程下载页面的程序,对于I/O密集的应用而言,多线程显然是个很好的解决方案。刚刚写过的线程池,也正好可以利用上了。其实用python爬取页面非常简单,有个urllib2的模块,使用起来很方便,基本两三行代码就可以搞定。虽然使用第三方模块,可以很方便的解决问题,但是对个人的技术积累而言没有什么好处,因为关键的算法都是别人实现的,而不是你自己实现的,很多细节的东西,你根本就无法了解。 我们做技术的,不能一味的只是用别人写好的模块或是api,要自己动手实现,才能让自己学习得更多。
我决定从socket写起,也是去封装GET协议,解析header,而且还可以把DNS的解析过程单独处理,例如DNS缓存一下,所以这样自己写的话,可控性更强,更有利于扩展。对于timeout的处理,我用的全局的5秒钟的超时处理,对于重定位(301or302)的处理是,最多重定位3次,因为之前测试过程中,发现很多站点的重定位又定位到自己,这样就无限循环了,所以设置了上限。具体原理,比较简单,直接看代码就好了。
自己写完之后,与urllib2进行了下性能对比,自己写的效率还是比较高的,而且urllib2的错误率稍高一些,不知道为什么。网上有人说urllib2在多线程背景下有些小问题,具体我也不是特别清楚。
先贴代码:
fetchPage.py 使用Http协议的Get方法,进行页面下载,并存储为文件
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
‘‘‘ Created on 2012-3-13 Get Page using GET method Default using HTTP Protocol , http port 80 @author: xiaojay ‘‘‘ import socket import statistics import datetime import threading socket.setdefaulttimeout(statistics.timeout) class Error404(Exception): ‘‘‘Can not find the page.‘‘‘ pass class ErrorOther(Exception): ‘‘‘Some other exception‘‘‘ def __init__( self ,code): #print ‘Code :‘,code pass class ErrorTryTooManyTimes(Exception): ‘‘‘try too many times‘‘‘ pass def downPage(hostname ,filename , trytimes = 0 ): try : #To avoid too many tries .Try times can not be more than max_try_times if trytimes > = statistics.max_try_times : raise ErrorTryTooManyTimes except ErrorTryTooManyTimes : return statistics.RESULTTRYTOOMANY,hostname + filename try : s = socket.socket(socket.AF_INET,socket.SOCK_STREAM) #DNS cache if statistics.DNSCache.has_key(hostname): addr = statistics.DNSCache[hostname] else : addr = socket.gethostbyname(hostname) statistics.DNSCache[hostname] = addr #connect to http server ,default port 80 s.connect((addr, 80 )) msg = ‘GET ‘ + filename + ‘ HTTP/1.0\r\n‘ msg + = ‘Host: ‘ + hostname + ‘\r\n‘ msg + = ‘User-Agent:xiaojay\r\n\r\n‘ code = ‘‘ f = None s.sendall(msg) first = True while True : msg = s.recv( 40960 ) if not len (msg): if f! = None : f.flush() f.close() break # Head information must be in the first recv buffer if first: first = False headpos = msg.index( "\r\n\r\n" ) code,other = dealwithHead(msg[:headpos]) if code = = ‘200‘ : #statistics.fetched_url += 1 f = open ( ‘pages/‘ + str ( abs ( hash (hostname + filename))), ‘w‘ ) f.writelines(msg[headpos + 4 :]) elif code = = ‘301‘ or code = = ‘302‘ : #if code is 301 or 302 , try down again using redirect location if other.startswith( "http" ) : hname, fname = parse(other) downPage(hname,fname,trytimes + 1 ) #try again else : downPage(hostname,other,trytimes + 1 ) elif code = = ‘404‘ : raise Error404 else : raise ErrorOther(code) else : if f! = None :f.writelines(msg) s.shutdown(socket.SHUT_RDWR) s.close() return statistics.RESULTFETCHED,hostname + filename except Error404 : return statistics.RESULTCANNOTFIND,hostname + filename except ErrorOther: return statistics.RESULTOTHER,hostname + filename except socket.timeout: return statistics.RESULTTIMEOUT,hostname + filename except Exception, e: return statistics.RESULTOTHER,hostname + filename def dealwithHead(head): ‘‘‘deal with HTTP HEAD‘‘‘ lines = head.splitlines() fstline = lines[ 0 ] code = fstline.split()[ 1 ] if code = = ‘404‘ : return (code, None ) if code = = ‘200‘ : return (code, None ) if code = = ‘301‘ or code = = ‘302‘ : for line in lines[ 1 :]: p = line.index( ‘:‘ ) key = line[:p] if key = = ‘Location‘ : return (code,line[p + 2 :]) return (code, None ) def parse(url): ‘‘‘Parse a url to hostname+filename‘‘‘ try : u = url.strip().strip( ‘\n‘ ).strip( ‘\r‘ ).strip( ‘\t‘ ) u = u[ 7 :] u = u[ 8 :] if u.find( ‘:80‘ )> 0 : p = u.index( ‘:80‘ ) p2 = p + 3 else : if u.find( ‘/‘ )> 0 : p = u.index( ‘/‘ ) p2 = p else : p = len (u) p2 = - 1 hostname = u[:p] if p2> 0 : filename = u[p2:] else : filename = ‘/‘ return hostname, filename except Exception ,e: print "Parse wrong : " , url print e def PrintDNSCache(): ‘‘‘print DNS dict‘‘‘ n = 1 for hostname in statistics.DNSCache.keys(): print n, ‘\t‘ ,hostname, ‘\t‘ ,statistics.DNSCache[hostname] n + = 1 def dealwithResult(res,url): ‘‘‘Deal with the result of downPage‘‘‘ statistics.total_url + = 1 if res = = statistics.RESULTFETCHED : statistics.fetched_url + = 1 print statistics.total_url , ‘\t fetched :‘ , url if res = = statistics.RESULTCANNOTFIND : statistics.failed_url + = 1 print "Error 404 at : " , url if res = = statistics.RESULTOTHER : statistics.other_url + = 1 print "Error Undefined at : " , url if res = = statistics.RESULTTIMEOUT : statistics.timeout_url + = 1 print "Timeout " ,url if res = = statistics.RESULTTRYTOOMANY: statistics.trytoomany_url + = 1 print e , "Try too many times at" , url if __name__ = = ‘__main__‘ : print ‘Get Page using GET method‘ |
下面,我将利用上一篇的线程池作为辅助,实现多线程下的并行爬取,并用上面自己写的下载页面的方法和urllib2进行一下性能对比。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
|
‘‘‘ Created on 2012-3-16 @author: xiaojay ‘‘‘ import fetchPage import threadpool import datetime import statistics import urllib2 ‘‘‘one thread‘‘‘ def usingOneThread(limit): urlset = open ( "input.txt" , "r" ) start = datetime.datetime.now() for u in urlset: if limit < = 0 : break limit - = 1 hostname , filename = parse(u) res = fetchPage.downPage(hostname,filename, 0 ) fetchPage.dealwithResult(res) end = datetime.datetime.now() print "Start at :\t" , start print "End at :\t" , end print "Total Cost :\t" , end - start print ‘Total fetched :‘ , statistics.fetched_url ‘‘‘threadpoll and GET method‘‘‘ def callbackfunc(request,result): fetchPage.dealwithResult(result[ 0 ],result[ 1 ]) def usingThreadpool(limit,num_thread): urlset = open ( "input.txt" , "r" ) start = datetime.datetime.now() main = threadpool.ThreadPool(num_thread) for url in urlset : try : hostname , filename = fetchPage.parse(url) req = threadpool.WorkRequest(fetchPage.downPage,args = [hostname,filename],kwds = {},callback = callbackfunc) main.putRequest(req) except Exception: print Exception.message while True : try : main.poll() if statistics.total_url > = limit : break except threadpool.NoResultsPending: print "no pending results" break except Exception ,e: print e end = datetime.datetime.now() print "Start at :\t" , start print "End at :\t" , end print "Total Cost :\t" , end - start print ‘Total url :‘ ,statistics.total_url print ‘Total fetched :‘ , statistics.fetched_url print ‘Lost url :‘ , statistics.total_url - statistics.fetched_url print ‘Error 404 :‘ ,statistics.failed_url print ‘Error timeout :‘ ,statistics.timeout_url print ‘Error Try too many times ‘ ,statistics.trytoomany_url print ‘Error Other faults ‘ ,statistics.other_url main.stop() ‘‘‘threadpool and urllib2 ‘‘‘ def downPageUsingUrlib2(url): try : req = urllib2.Request(url) fd = urllib2.urlopen(req) f = open ( "pages3/" + str ( abs ( hash (url))), ‘w‘ ) f.write(fd.read()) f.flush() f.close() return url , ‘success‘ except Exception: return url , None def writeFile(request,result): statistics.total_url + = 1 if result[ 1 ]! = None : statistics.fetched_url + = 1 print statistics.total_url, ‘\tfetched :‘ , result[ 0 ], else : statistics.failed_url + = 1 print statistics.total_url, ‘\tLost :‘ ,result[ 0 ], def usingThreadpoolUrllib2(limit,num_thread): urlset = open ( "input.txt" , "r" ) start = datetime.datetime.now() main = threadpool.ThreadPool(num_thread) for url in urlset : try : req = threadpool.WorkRequest(downPageUsingUrlib2,args = [url],kwds = {},callback = writeFile) main.putRequest(req) except Exception ,e: print e while True : try : main.poll() if statistics.total_url > = limit : break except threadpool.NoResultsPending: print "no pending results" break except Exception ,e: print e end = datetime.datetime.now() print "Start at :\t" , start print "End at :\t" , end print "Total Cost :\t" , end - start print ‘Total url :‘ ,statistics.total_url print ‘Total fetched :‘ , statistics.fetched_url print ‘Lost url :‘ , statistics.total_url - statistics.fetched_url main.stop() if __name__ = = ‘__main__‘ : ‘‘‘too slow‘‘‘ #usingOneThread(100) ‘‘‘use Get method‘‘‘ #usingThreadpool(3000,50) ‘‘‘use urllib2‘‘‘ usingThreadpoolUrllib2( 3000 , 50 ) |
实验分析:
实验数据:larbin抓取下来的3000条url,经过Mercator队列模型(我用c++实现的,以后有机会发个blog)处理后的url集合,具有随机和代表性。使用50个线程的线程池。
实验环境:ubuntu10.04,网络较好,python2.6
存储:小文件,每个页面,一个文件进行存储
PS:由于学校上网是按流量收费的,做网络爬虫,灰常费流量啊!!!过几天,可能会做个大规模url下载的实验,用个几十万的url试试。
实验结果:
使用urllib2 ,usingThreadpoolUrllib2(3000,50)
Start at : 2012-03-16 22:18:20.956054
End at : 2012-03-16 22:22:15.203018
Total Cost : 0:03:54.246964
Total url : 3001
Total fetched : 2442
Lost url : 559
下载页面的物理存储大小:84088kb
使用自己的getPageUsingGet ,usingThreadpool(3000,50)
Start at : 2012-03-16 22:23:40.206730
End at : 2012-03-16 22:26:26.843563
Total Cost : 0:02:46.636833
Total url : 3002
Total fetched : 2484
Lost url : 518
Error 404 : 94
Error timeout : 312
Error Try too many times 0
Error Other faults 112
下载页面的物理存储大小:87168kb
小结: 自己写的下载页面程序,效率还是很不错的,而且丢失的页面也较少。但其实自己考虑一下,还是有很多地方可以优化的,比如文件过于分散,过多的小文件创建和释放定会产生不小的性能开销,而且程序里用的是hash命名,也会产生很多的计算,如果有好的策略,其实这些开销都是可以省略的。另外DNS,也可以不使用python自带的DNS解析,因为默认的DNS解析都是同步的操作,而DNS解析一般比较耗时,可以采取多线程的异步的方式进行,再加以适当的DNS缓存很大程度上可以提高效率。不仅如此,在实际的页面抓取过程中,会有大量的url ,不可能一次性把它们存入内存,而应该按照一定的策略或是算法进行合理的分配。 总之,采集页面要做的东西以及可以优化的东西,还有很多很多。
标签:
原文地址:http://www.cnblogs.com/jokerjason/p/5696862.html