zoukankan      html  css  js  c++  java
  • Python3 多线程下载代码

    根据http://www.oschina.net/code/snippet_70229_2407修改而来的增强版。貌似原版源自Axel这个多线程下载工具。

    '''
    Created on 2014-10-24
    
    @author: Maple
    '''
    
    import sys
    import os
    import time
    import getopt
    import urllib.request
    import urllib.parse
    from threading import Thread
    
    #===============================================================================
    # def download(url, output=os.getcwd(), blocks=6, proxies=local_proxies)
    # output:输出文件路径,默认为当前路径
    # blocks:线程数
    # proxies:代理地址
    #===============================================================================
    
    local_proxies = {}#代理地址
    
    class Maple(Thread):
        version = "Mozilla/5.0"
    
        def __init__(self, threadname, url, filename, ranges=0, proxies={}):
            Thread.__init__(self, name=threadname)
            self.name = threadname
            self.url = url
            self.proxies = proxies
            self.filename = filename
            self.ranges = ranges
            self.downloaded = 0
    
        def run(self):
    
            try:
                self.downloaded = os.path.getsize( self.filename )        #获取已下载的文件字节块块,支持断点续传
            except OSError:
                #print 'never downloaded'
                self.downloaded = 0
            opener=GetUrlOpener(self.proxies)                                    #根据代理参数生成相应的url opener
            if self.ranges:                                                                         #ranges为线程需要下载的文件块的字节范围
                # rebuild start poind
                self.startpoint = self.ranges[0] + self.downloaded        #从已下载字节块后的位置开始下载
    
                # This part is completed
                if self.startpoint >= self.ranges[1]:
                    self.downloaded = self.ranges[1] - self.ranges[0]                        
                    print ('Part %s has been downloaded over.' % self.filename)
                    return
                opener.addheaders=[('Range','bytes={}-{}'.format(self.startpoint, self.ranges[1])),('User-agent','Mozilla/5.0')]     #添加请求头部内容,仅下载指定范围的字节,伪装成浏览器请求
                print ('task %s will download from %d to %d' % (self.name, self.startpoint+1, self.ranges[1]+1))
            else:                                       #ranges未指定(文件大小未知,无法切割),从已下载字节块后的位置开始下载剩余全部字节
                self.startpoint = self.downloaded
                opener.addheaders=[('Range','bytes={}-'.format(self.startpoint)),('User-agent','Mozilla/5.0')]
            self.fetchsize = 16384                                                     #每次读取的字节数
            self.urlhandle = opener.open(self.url)                             #打开文件地址
            data = self.urlhandle.read( self.fetchsize )                      
            while data:                                                          #循环读取数据写入临时文件,并更新已下载字节数
                filehandle = open( self.filename, 'ab+' )
                filehandle.write( data )
                filehandle.close()
                self.downloaded += len( data )
                data = self.urlhandle.read( self.fetchsize )
    
    def Sec2Time(second):                        #将秒数转换为标准时间格式。以为有现成的函数,结果愣是没找到
        day=second//(3600*24)
        second-=day*3600*24
        hour=second//3600
        second -=hour*3600
        minute=second//60
        second-=minute*60
        if day == 0:
            if hour == 0:
                if minute == 0:
                    return '{:0.2f}S.'.format(second)
                else:
                    return '{:02}M:{:0.2f}S'.format(minute,second)
            else:
                return '{:02}H:{:02}M:{:0.2f}S'.format(hour,minute,second)
        else:
            return '{:03}D:{:02}H:{:02}M:{:0.2f}S'.format(day,hour,minute,second)
    
    def GetUrlOpener(proxies={}):         #分析代理参数,返回url opener。完整代理格式:user/passwd@http://127.0.0.1:8087。如格式不同,需要修改此分析函数
        if proxies:
            try:
                ap=proxies.split('@')
                if len(ap) > 1:
                    auth=ap[0]
                    addr=ap[1]
                else:
                    addr=ap[0]
                    auth=''
                if '://' in addr:
                    ptype=addr[:addr.find('://')]
                    phost=addr[addr.find('://')+3:]
                else:
                    ptype='http'
                    phost=addr
                proxy={ptype:ptype+'://'+phost}
                proxy_handler = urllib.request.ProxyHandler(proxy)
            except Exception as ex:
                print(ex)
                return urllib.request.build_opener()
            try:
                authlist=auth.split('/')
                if len(authlist) > 1:
                    user=authlist[0]
                    passwd=authlist[1]
                    proxy_auth_handler = urllib.request.ProxyBasicAuthHandler()
                    proxy_auth_handler.add_password('realm',phost,user,passwd)
                    opener = urllib.request.build_opener(proxy_handler,proxy_auth_handler)
                else:
                    opener = urllib.request.build_opener(proxy_handler)
                return opener
            except Exception as ex:
                print(ex)
                return urllib.request.build_opener(proxy_handler)
        else:
           # urlHandler=urllib.request.urlopen(url)
            return urllib.request.build_opener()
    
    def GetUrlFileInfo(url,proxies={}):            #获取要下载的文件的信息,包括文件名,文件类型和文件大小
        scheme, netloc, path, query, fragment = urllib.parse.urlsplit(url)     #分析url
        filename=urllib.parse.unquote(path)       #如果url中的文件名部分存在中文,将其正确解码出来
        filename=filename.split('/')[-1]
        opener=GetUrlOpener(proxies)          #通过网络请求读取响应头部,根据头部获取文件信息。文件名以服务器返回的文件名信息为准
        urlHandler=opener.open(url)
        headers=urlHandler.info()
        if 'Content-Disposition' in headers:       #Content-Disposition字段有可能获取到文件名,不过可能是乱码,没找到解决办法</span>
            disposition=headers.get('Content-Disposition')
            if 'filename=' in disposition:
                filename = disposition.split('filename=')[1]
                if filename[0] == '"' or filename[0] == "'":
                    filename = filename[1:-1]
        filename=urllib.parse.unquote(filename)
        if filename:
            (name,ext)=os.path.splitext(filename)
        else:
            (name,ext)=('Unknown','')
        if 'Content-Length' in headers:          #获取文件长度,如果获取失败,则只能使用单线程下载
            length=int(headers.get('Content-Length'))
        else:
            length=-1            
        (type, kind)=headers.get('Content-Type').split('/') #获取文件类型,备用
        infos=[(name,ext),(type,kind),length]
        return infos
    
    def SpliteBlocks(totalsize, blocknumber):   #根据指定的线程数参数和获取到的文件长度划分各线程的下载范围
        blocksize = totalsize//blocknumber
        ranges = []
        for i in range(0, blocknumber-1):
            ranges.append((i*blocksize, i*blocksize +blocksize - 1))
        ranges.append(( blocksize*(blocknumber-1), totalsize -1 ))
    
        return ranges
    def islive(tasks):                        #检查各线程是否全部下载完成
        for task in tasks:
            if task.isAlive():
                return True
        return False
    
    def download(url, target=os.getcwd(), blocks=6, proxies=local_proxies):               
    
        flag=True
        print('Retrieving resource information...')
    
        url=urllib.parse.quote(url,safe='/%&@=+?$;,:')        #将提供的url编码,非英文字符将被编码为标准格式
        try:
            infos=GetUrlFileInfo(url,proxies)                            #获取文件信息
        except Exception as ex:
            print(ex)
            flag=False
        if flag:  
            if not os.path.exists(target):
                os.makedirs(target)
            size=infos[2]                                                              #获取到的文件大小
            output=os.path.join(target,''.join(infos[0]))         #根据获取到的文件名和指定的保存目录生成完整路径
            type=infos[1][0]
    
            starttime=time.time()                                              #开始计时
            print('Infomation:')
            print('FileName:{0}     FileType:{1}    FileLength:{2}'.format(''.join(infos[0]),'/'.join(infos[1]),infos[2] if int(infos[2]) > 0 else 'Unknown'))   #打印获取到的文件信息
            if size > 0:                                                                               #size大于0表示成功获取文件长度,可以进行多线程下载
                print('Starting multithread download...')
                ranges = SpliteBlocks( size, blocks )
            else:                                                                                         #只能单线程下载,线程数置1,ranges置空,
                print('Starting single thread download...')
                ranges=()
                blocks=1
            threadname = [ infos[0][0]+"_thread_%d" % i for i in range(0, blocks) ]         #生成线程名
            filename = [ infos[0][0]+ "_tmpfile_%d" % i for i in range(0, blocks) ]             #生成各线程的临时文件名
            tasks = []                                                                                                                
            for i in range(0,blocks):                                                                                        #生成下载线程,设置为后台线程后启动,将线程加入到线程列表中
                task = Maple( threadname[i], url, filename[i], ranges[i] if ranges else ranges,proxies)
                task.setDaemon( True )
                task.start()
                tasks.append( task )
    
            time.sleep( 1 )
            downloaded = 0
            while islive(tasks):                                                                                                 #统计线程列表中各线程的状态,输出下载进度
                downloaded = sum( [task.downloaded for task in tasks] )
                if size > 0:
                    process = downloaded/float(size)*100
                    show = '
    Filesize:%d Downloaded:%d Completed:%.2f%%' % (size, downloaded, process)
                else:
                    show = '
    Downloaded:%d ' % downloaded
                sys.stdout.write(show)
                sys.stdout.flush()
                time.sleep( 0.2 )
            endtime=time.time()                                                                                           #下载完成后停止计时
            consuming=Sec2Time(endtime-starttime)
            if size > 0:                                                                                                             #多线程下载的后续处理
                downloadsize = 0
                for i in filename:
                    downloadsize += os.path.getsize(i)
                if downloadsize == size:
                    show = '
    Filesize:%d Downloaded:%d Completed:%.2f%%
    ' % (size, downloadsize,100) 
                else:
                    show = '
    Size is not mathed!
    '
                    flag=False
            else:                                                                                                                   #单线程下载的后续处理
                show = '
    Total Size: %d
    '% downloaded
            sys.stdout.write(show)
            sys.stdout.flush()
            if flag:                                                                                                                #确认下载的临时文件没问题后将各文件整合为最终的目标文件
                print('Integrating files...')
                num=1
                while os.path.exists(output):                                                                      #防止与本地已存在文件重名
                    fname,fext=os.path.splitext(output)
                    if '('+str(num-1)+')'+fext in output:
                        output = output.replace('('+str(num-1)+')'+fext,'('+str(num)+')'+fext)
                    else:
                        fname += '('+str(num)+')'
                        output = fname+fext
                    num +=1
                if len(filename) ==1 :                                                                                  #单线程下载的话,直接将下载的文件重命名为目标文件即可
                    os.rename(filename[0], output)
                else:                                                                                                             #多线程临时文件整合
                    filehandle = open( output, 'wb+' )
                    for i in filename:
                        try:
                            f = open( i, 'rb' )
                            filehandle.write( f.read() )
                            f.close()
                            os.remove(i)
                        except Exception as ex:
                            print(ex)
                    filehandle.close()
                if os.path.exists(output):
                    print('Download Complete!')
                else:
                    print('Failed to generate target file!')
                    try:
                        #os.remove(output)
                        pass
                    except:
                        pass
            else:
                for i in filename:
                    try:
                        os.remove(i)
                        pass
                    except:
                        pass
                print('Download Failed!')
                pass
            print('Consuming: {}
    '.format(consuming))                        #输出耗时
        else:
            print('Failed to retrieve resource information!')
            sys.exit()
    def main(argv):                                                                                #处理传入参数,使用了getopt模块,另外有一个更强大的处理传入参数的模块optparse
        try:
            options,args=getopt.getopt(argv,'hu:f:n:p:',['help','url=','target=','num=','proxy='])
        except Exception as ex:
            print(ex)
            sys.exit()
        num = 2
        url,target,proxies= '','',''
    
        url = 'http://www.pygtk.org/dist/pygtk2-tut.pdf'
        target = '/home/maple/Desktop'
        #proxies = 'http://127.0.0.1:8087'
        #proxies={}
        for name, value in options:
            if name in ('-h','--help'):
                print('No Help ^^')
                sys.exit()
            if name in ('-u','--url'):
                url = value
            if name in ('-t','--target'):
                target = value
            if name in ('-n','--num'):
                num = int(value)
            if name in ('-p','--proxy'):
                proxies=value
        #check args
        download(url,target,num,proxies)
    
    
    if __name__ == '__main__':
        main(sys.argv[1:])

    这段代码在异常处理方面写得有些乱,没怎么关心异常处理,需要时再改吧

    另外多线程下载时,如果使用了代理,会导致下载到的文件与服务器提供的文件大小不符。从而下载失败。我使用的是GoAgent,代理服务器会自动对目标文件进行多线程下载,无视程序指定的下载字节范围。第一个线程就会下载到完整的文件,其他线程会下载冗余内容。没有找到规范的处理办法。变通的处理办法有2种:

    1、将使用了代理的多线程下载强制指定为单线程下载

    2、不进行文件大小的校验,将多线程下载的0号临时文件重命名为目标文件,其他临时文件删除。

    两种方法实现都很简单,但是破坏代码的整体逻辑。没有加入代码中。运行截图:


  • 相关阅读:
    表的管理
    子查询
    sql语句
    基本sql语句与oracle函数
    Visual C# 2008+SQL Server 2005 数据库与网络开发6.1.1 报表服务概述
    Visual C# 2008+SQL Server 2005 数据库与网络开发 5.4 小结
    Visual C# 2008+SQL Server 2005 数据库与网络开发5.3.1 日期时间函数
    Visual C# 2008+SQL Server 2005 数据库与网络开发 5.3 函数
    Visual C# 2008+SQL Server 2005 数据库与网络开发第6章 数据报表
    Visual C# 2008+SQL Server 2005 数据库与网络开发5.2.2 GROUP BY
  • 原文地址:https://www.cnblogs.com/xshrim/p/4048954.html
Copyright © 2011-2022 走看看