Beispiel #1
0
        //传递过来一个List对象


        //返回的是一个 类
        public static Web_Content Test_Ip(Ip ip, string url)
        {
            //Console.WriteLine(ip.Port);



            //尝试代理Ip  接受请求
            WebProxy proxyObject = new WebProxy(ip.Address, Convert.ToInt32(ip.Port));
            // WebProxy proxyObject = new WebProxy("47.240.100.124");
            //设置请求
            HttpWebRequest Req = (HttpWebRequest)WebRequest.Create(url);     // 61.183.192.5

            Req.UserAgent = @"Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 6.1; Trident/4.0; 
                    QQWubi 133; SLCC2; .NET CLR 2.0.50727; .NET CLR 3.5.30729; .NET 
                        CLR 3.0.30729; Media Center PC 6.0; CIBA; InfoPath.2)";

            Req.Proxy   = proxyObject;   //设置代理
            Req.Method  = "GET";
            Req.Timeout = 5000;

            Web_Content result = new Web_Content();

            try
            {
                HttpWebResponse Resp = (HttpWebResponse)Req.GetResponse();
                Encoding        code = Encoding.GetEncoding("UTF-8");
                using (StreamReader sr = new StreamReader(Resp.GetResponseStream(), code))
                {
                    string str = null;

                    str = sr.ReadToEnd();    //获取得到的网址html返回数据,这里就可以使用某些解析html的dll直接使用了,比如htmlpaser
                    if (str != null)
                    {
                        result.Connect = true;
                        result.Content = str;
                        return(result);
                    }
                }
            }
            catch (Exception e)
            {
                result.Content = "";
                result.Connect = false;
            }



            result.Content = "";
            result.Connect = false;
            return(result);
        }
Beispiel #2
0
        static void Main(string[] args)
        {
            /*获取网页内容*/
            try
            {
                //传递的字符串
                string url  = "http://www.xicidaili.com/nn/1";
                string Html = Get_Html(url);

                /* //获取代理*/
                #region
                string regex = @"<td>(\d+\.\d+\.\d+\.\d+)</td>\s+<td>(\d+)</td>";
                Match  mstr  = Regex.Match(Html, regex);

                //c#高效队列
                //ConcurrentQueue<string> proxyIp = new ConcurrentQueue<string>();

                List <Ip> proxyIp = new List <Ip>();
                //设置dictionary 来存放
                while (mstr.Success)
                {
                    // Response.Write(match.Value + "<br/>");
                    // match = match.NextMatch();
                    //} 获取
                    /* Console.WriteLine(mstr.Groups[1].Value + ":" + mstr.Groups[2].Value);*/


                    /*存放进高效队列*/

                    proxyIp.Add(new Ip(mstr.Groups[1].Value.ToString(), mstr.Groups[2].Value.ToString()));
                    mstr = mstr.NextMatch();
                }

                #endregion
                //输出获得代理ip List中的个数
                Console.WriteLine("从代理网站获取的Ip个数" + proxyIp.Count);
                //处理获取到的Ip 用来测试网页   输入值为List中的Ip,输出值为爬取的网站
                #region

                string Catched_url = "https://baidu.com";
                //string Catched_url = @"http://ditu.92cha.com/dizhen.php?dizhen_ly=china&dizhen_zjs=1&dizhen_zje=6&dizhen_riqis=&dizhen_riqie=&ckwz=";

                int i = 0;
                foreach (Ip ip in proxyIp)
                {
                    Web_Content result = Test_Ip(ip, Catched_url);
                    if (result.Connect == false)
                    {
                        Console.WriteLine(ip.Address + "不可以用");
                        continue;
                    }
                    else
                    {
                        Console.WriteLine(ip.Address + "可以用");


                        //写入文件
                        //StreamWriter.Write

                        //在将文本写入文件前,处理文本行
                        //StreamWriter一个参数默认覆盖
                        //StreamWriter第二个参数为false覆盖现有文件,为true则把文本追加到文件末尾

                        //StreamWriter(Stream stream);推荐使用,灵活便于理解多态
                        //StreamWriter(string  path);写死了就是指向硬盘写文件

                        /*
                         *                  //C#文件流写文件,默认追加FileMode.Append
                         *                  Console.WriteLine("你爬取的网站为"+Catched_url+"\n"+"请输入你保存文件的名字");
                         *                  string FileName = Console.ReadLine();
                         *                  FileName = @"../../"+DateTime.Now.Year.ToString()+"_"+DateTime.Now.Month.ToString()+"_"+DateTime.Now.Day.ToString()+"_"+DateTime.Now.Hour+"_"+DateTime.Now.Minute +"_"+ FileName + ".txt";
                         *                  byte[] myByte = System.Text.Encoding.UTF8.GetBytes(result.Content);
                         *                  using (FileStream fsWrite = new FileStream(FileName, FileMode.Append))
                         *                  {
                         *                      fsWrite.Write(myByte, 0, myByte.Length);
                         *                  };
                         *                          Console.WriteLine("写入完成");
                         *
                         * */
                        i++;
                        continue;
                    }
                }
                #endregion

                Console.WriteLine(i + "可用");
            }


            catch (Exception e)
            {
                //输出当前异常的信息
                Console.WriteLine(e.Message);

                Console.WriteLine(e.ToString());
            }



            Console.ReadKey();
        }