用 Java 实现断点续传参考 (HTTP)

断点续传的原理

其实断点续传的原理很简单,就是在 Http 的请求上和一般的下载有所不同而已。        打个比方,浏览器请求服务器上的一个文时,所发出的请求如下:        假设服务器域名为 wwww.sjtu.edu.cn,文件名为 down.zip。 GET /down.zip HTTP/1.1         Accept: image/gif, image/x-xbitmap, image/jpeg, image/pjpeg, application/vnd.ms-         excel, application/msword, application/vnd.ms-powerpoint, */*         Accept-Language: zh-cn         Accept-Encoding: gzip, deflate         User-Agent: Mozilla/4.0 (compatible; MSIE 5.01; Windows NT 5.0)         Connection: Keep-Alive     

服务器收到请求后,按要求寻找请求的文件,提取文件的信息,然后返回给浏览器,返回信息如下:

200         Content-Length=106786028         Accept-Ranges=bytes         Date=Mon, 30 Apr 2001 12:56:11 GMT         ETag=W/"02ca57e173c11:95b"        Content-Type=application/octet-stream         Server=Microsoft-IIS/5.0         Last-Modified=Mon, 30 Apr 2001 12:56:11 GMT     

所谓断点续传,也就是要从文件已经下载的地方开始继续下载。所以在客户端浏览器传给 Web 服务器的时候要多加一条信息 -- 从哪里开始。        下面是用自己编的一个"浏览器"来传递请求信息给 Web 服务器,要求从 2000070 字节开始。        GET /down.zip HTTP/1.0         User-Agent: NetFox         RANGE: bytes=2000070-         Accept: text/html, image/gif, image/jpeg, *; q=.2, */*; q=.2     

仔细看一下就会发现多了一行 RANGE: bytes=2000070-         这一行的意思就是告诉服务器 down.zip 这个文件从 2000070 字节开始传,前面的字节不用传了。        服务器收到这个请求以后,返回的信息如下:        206         Content-Length=106786028         Content-Range=bytes 2000070-106786027/106786028         Date=Mon, 30 Apr 2001 12:55:20 GMT         ETag=W/"02ca57e173c11:95b"        Content-Type=application/octet-stream         Server=Microsoft-IIS/5.0         Last-Modified=Mon, 30 Apr 2001 12:55:20 GMT     

和前面服务器返回的信息比较一下,就会发现增加了一行:        Content-Range=bytes 2000070-106786027/106786028         返回的代码也改为 206 了,而不再是 200 了。    

知道了以上原理,就可以进行断点续传的编程了。

Java 实现断点续传的关键几点

  1. (1) 用什么方法实现提交 RANGE: bytes=2000070-。          当然用最原始的 Socket 是肯定能完成的,不过那样太费事了,其实 Java 的 net 包中提供了这种功能。代码如下:         
    URL url = new URL("http://www.sjtu.edu.cn/down.zip");           HttpURLConnection httpConnection =  (HttpURLConnection)url.openConnection();          
    // 设置 User-Agent           httpConnection.setRequestProperty("User-Agent","NetFox");           // 设置断点续传的开始位置          httpConnection.setRequestProperty("RANGE","bytes=2000070");           // 获得输入流          InputStream input = httpConnection.getInputStream();          

    从输入流中取出的字节流就是 down.zip 文件从 2000070 开始的字节流。 大家看,其实断点续传用 Java 实现起来还是很简单的吧。 接下来要做的事就是怎么保存获得的流到文件中去了。

  2. 保存文件采用的方法。          我采用的是 IO 包中的 RandAccessFile 类。          操作相当简单,假设从 2000070 处开始保存文件,代码如下:          RandomAccess oSavedFile = new  RandomAccessFile("down.zip","rw");           long nPos = 2000070;           // 定位文件指针到 nPos 位置          oSavedFile.seek(nPos);           byte[] b = new byte[1024];           int nRead;           // 从输入流中读入字节流,然后写到文件中          while((nRead=input.read(b,0,1024)) > 0)           {           oSavedFile.write(b,0,nRead);           }          

怎么样,也很简单吧。 接下来要做的就是整合成一个完整的程序了。包括一系列的线程控制等等。

断点续传内核的实现

主要用了 6 个类,包括一个测试类。        SiteFileFetch.java 负责整个文件的抓取,控制内部线程 (FileSplitterFetch 类 )。        FileSplitterFetch.java 负责部分文件的抓取。        FileAccess.java 负责文件的存储。        SiteInfoBean.java 要抓取的文件的信息,如文件保存的目录,名字,抓取文件的 URL 等。        Utility.java 工具类,放一些简单的方法。        TestMethod.java 测试类。       

下面是源程序:

/* 

 /*

 * SiteFileFetch.java 

 */ 

 package NetFox; 

 import java.io.*; 

 import java.net.*; 

 public class SiteFileFetch extends Thread { 

 SiteInfoBean siteInfoBean = null; // 文件信息 Bean 

 long[] nStartPos; // 开始位置

 long[] nEndPos; // 结束位置

 FileSplitterFetch[] fileSplitterFetch; // 子线程对象

 long nFileLength; // 文件长度

 boolean bFirst = true; // 是否第一次取文件

 boolean bStop = false; // 停止标志

 File tmpFile; // 文件下载的临时信息

 DataOutputStream output; // 输出到文件的输出流

 public SiteFileFetch(SiteInfoBean bean) throws IOException 

 { 

 siteInfoBean = bean; 

 //tmpFile = File.createTempFile ("zhong","1111",new File(bean.getSFilePath())); 

 tmpFile = new File(bean.getSFilePath()+File.separator + bean.getSFileName()+".info");

 if(tmpFile.exists ()) 

 { 

 bFirst = false; 

 read_nPos(); 

 } 

 else 

 { 

 nStartPos = new long[bean.getNSplitter()]; 

 nEndPos = new long[bean.getNSplitter()]; 

 } 

 } 

 public void run() 

 { 

 // 获得文件长度

 // 分割文件

 // 实例 FileSplitterFetch 

 // 启动 FileSplitterFetch 线程

 // 等待子线程返回

 try{ 

 if(bFirst) 

 { 

 nFileLength = getFileSize(); 

 if(nFileLength == -1) 

 { 

 System.err.println("File Length is not known!"); 

 } 

 else if(nFileLength == -2) 

 { 

 System.err.println("File is not access!"); 

 } 

 else 

 { 

 for(int i=0;i<nStartPos.length;i++) 

 { 

 nStartPos[i] = (long)(i*(nFileLength/nStartPos.length)); 

 } 

 for(int i=0;i<nEndPos.length-1;i++) 

 { 

 nEndPos[i] = nStartPos[i+1]; 

 } 

 nEndPos[nEndPos.length-1] = nFileLength; 

 } 

 } 

 // 启动子线程

 fileSplitterFetch = new FileSplitterFetch[nStartPos.length]; 

 for(int i=0;i<nStartPos.length;i++) 

 { 

 fileSplitterFetch[i] = new FileSplitterFetch(siteInfoBean.getSSiteURL(), 

 siteInfoBean.getSFilePath() + File.separator + siteInfoBean.getSFileName(), 

 nStartPos[i],nEndPos[i],i); 

 Utility.log("Thread " + i + " , nStartPos = " + nStartPos[i] + ", nEndPos = " 

 + nEndPos[i]); 

 fileSplitterFetch[i].start(); 

 } 

 // fileSplitterFetch[nPos.length-1] = new FileSplitterFetch(siteInfoBean.getSSiteURL(),

 siteInfoBean.getSFilePath() + File.separator 

 + siteInfoBean.getSFileName(),nPos[nPos.length-1],nFileLength,nPos.length-1); 

 // Utility.log("Thread " +(nPos.length-1) + ",nStartPos = "+nPos[nPos.length-1]+",

 nEndPos = " + nFileLength); 

 // fileSplitterFetch[nPos.length-1].start(); 

 // 等待子线程结束

 //int count = 0; 

 // 是否结束 while 循环

 boolean breakWhile = false; 

 while(!bStop) 

 { 

 write_nPos(); 

 Utility.sleep(500); 

 breakWhile = true; 

 for(int i=0;i<nStartPos.length;i++) 

 { 

 if(!fileSplitterFetch[i].bDownOver) 

 { 

 breakWhile = false; 

 break; 

 } 

 } 

 if(breakWhile) 

 break; 

 //count++; 

 //if(count>4) 

 // siteStop(); 

 } 

 System.err.println("文件下载结束!"); 

 } 

 catch(Exception e){e.printStackTrace ();} 

 } 

 // 获得文件长度

 public long getFileSize() 

 { 

 int nFileLength = -1; 

 try{ 

 URL url = new URL(siteInfoBean.getSSiteURL()); 

 HttpURLConnection httpConnection = (HttpURLConnection)url.openConnection ();

 httpConnection.setRequestProperty("User-Agent","NetFox"); 

 int responseCode=httpConnection.getResponseCode(); 

 if(responseCode>=400) 

 { 

 processErrorCode(responseCode); 

 return -2; //-2 represent access is error 

 } 

 String sHeader; 

 for(int i=1;;i++) 

 { 

 //DataInputStream in = new DataInputStream(httpConnection.getInputStream ()); 

 //Utility.log(in.readLine()); 

 sHeader=httpConnection.getHeaderFieldKey(i); 

 if(sHeader!=null) 

 { 

 if(sHeader.equals("Content-Length")) 

 { 

 nFileLength = Integer.parseInt(httpConnection.getHeaderField(sHeader)); 

 break; 

 } 

 } 

 else 

 break; 

 } 

 } 

 catch(IOException e){e.printStackTrace ();} 

 catch(Exception e){e.printStackTrace ();} 

 Utility.log(nFileLength); 

 return nFileLength; 

 } 

 // 保存下载信息(文件指针位置)

 private void write_nPos() 

 { 

 try{ 

 output = new DataOutputStream(new FileOutputStream(tmpFile)); 

 output.writeInt(nStartPos.length); 

 for(int i=0;i<nStartPos.length;i++) 

 { 

 // output.writeLong(nPos[i]); 

 output.writeLong(fileSplitterFetch[i].nStartPos); 

 output.writeLong(fileSplitterFetch[i].nEndPos); 

 } 

 output.close(); 

 } 

 catch(IOException e){e.printStackTrace ();} 

 catch(Exception e){e.printStackTrace ();} 

 } 

 // 读取保存的下载信息(文件指针位置)

 private void read_nPos() 

 { 

 try{ 

 DataInputStream input = new DataInputStream(new FileInputStream(tmpFile)); 

 int nCount = input.readInt(); 

 nStartPos = new long[nCount]; 

 nEndPos = new long[nCount]; 

 for(int i=0;i<nStartPos.length;i++) 

 { 

 nStartPos[i] = input.readLong(); 

 nEndPos[i] = input.readLong(); 

 } 

 input.close(); 

 } 

 catch(IOException e){e.printStackTrace ();} 

 catch(Exception e){e.printStackTrace ();} 

 } 

 private void processErrorCode(int nErrorCode) 

 { 

 System.err.println("Error Code : " + nErrorCode); 

 } 

 // 停止文件下载

 public void siteStop() 

 { 

 bStop = true; 

 for(int i=0;i<nStartPos.length;i++) 

 fileSplitterFetch[i].splitterStop(); 

 } 

 }

 /* 

 **FileSplitterFetch.java 

 */ 

 package NetFox; 

 import java.io.*; 

 import java.net.*; 

 public class FileSplitterFetch extends Thread { 

 String sURL; //File URL 

 long nStartPos; //File Snippet Start Position 

 long nEndPos; //File Snippet End Position 

 int nThreadID; //Thread's ID 

 boolean bDownOver = false; //Downing is over 

 boolean bStop = false; //Stop identical 

 FileAccessI fileAccessI = null; //File Access interface 

 public FileSplitterFetch(String sURL,String sName,long nStart,long nEnd,int id)

 throws IOException 

 { 

 this.sURL = sURL; 

 this.nStartPos = nStart; 

 this.nEndPos = nEnd; 

 nThreadID = id; 

 fileAccessI = new FileAccessI(sName,nStartPos); 

 } 

 public void run() 

 { 

 while(nStartPos < nEndPos && !bStop) 

 { 

 try{ 

 URL url = new URL(sURL); 

 HttpURLConnection httpConnection = (HttpURLConnection)url.openConnection (); 

 httpConnection.setRequestProperty("User-Agent","NetFox"); 

 String sProperty = "bytes="+nStartPos+"-"; 

 httpConnection.setRequestProperty("RANGE",sProperty); 

 Utility.log(sProperty); 

 InputStream input = httpConnection.getInputStream(); 

 //logResponseHead(httpConnection); 

 byte[] b = new byte[1024]; 

 int nRead; 

 while((nRead=input.read(b,0,1024)) > 0 && nStartPos < nEndPos 

 && !bStop) 

 { 

 nStartPos += fileAccessI.write(b,0,nRead); 

 //if(nThreadID == 1) 

 // Utility.log("nStartPos = " + nStartPos + ", nEndPos = " + nEndPos); 

 } 

 Utility.log("Thread " + nThreadID + " is over!"); 

 bDownOver = true; 

 //nPos = fileAccessI.write (b,0,nRead); 

 } 

 catch(Exception e){e.printStackTrace ();} 

 } 

 } 

 // 打印回应的头信息

 public void logResponseHead(HttpURLConnection con) 

 { 

 for(int i=1;;i++) 

 { 

 String header=con.getHeaderFieldKey(i); 

 if(header!=null) 

 //responseHeaders.put(header,httpConnection.getHeaderField(header)); 

 Utility.log(header+" : "+con.getHeaderField(header)); 

 else 

 break; 

 } 

 } 

 public void splitterStop() 

 { 

 bStop = true; 

 } 

 } 

 

 /* 

 **FileAccess.java 

 */ 

 package NetFox; 

 import java.io.*; 

 public class FileAccessI implements Serializable{ 

 RandomAccessFile oSavedFile; 

 long nPos; 

 public FileAccessI() throws IOException 

 { 

 this("",0); 

 } 

 public FileAccessI(String sName,long nPos) throws IOException 

 { 

 oSavedFile = new RandomAccessFile(sName,"rw"); 

 this.nPos = nPos; 

 oSavedFile.seek(nPos); 

 } 

 public synchronized int write(byte[] b,int nStart,int nLen) 

 { 

 int n = -1; 

 try{ 

 oSavedFile.write(b,nStart,nLen); 

 n = nLen; 

 } 

 catch(IOException e) 

 { 

 e.printStackTrace (); 

 } 

 return n; 

 } 

 } 

 

 /* 

 **SiteInfoBean.java 

 */ 

 package NetFox; 

 public class SiteInfoBean { 

 private String sSiteURL; //Site's URL 

 private String sFilePath; //Saved File's Path 

 private String sFileName; //Saved File's Name 

 private int nSplitter; //Count of Splited Downloading File 

 public SiteInfoBean() 

 { 

 //default value of nSplitter is 5 

 this("","","",5); 

 } 

 public SiteInfoBean(String sURL,String sPath,String sName,int nSpiltter)

 { 

 sSiteURL= sURL; 

 sFilePath = sPath; 

 sFileName = sName; 

 this.nSplitter = nSpiltter; 

 } 

 public String getSSiteURL() 

 { 

 return sSiteURL; 

 } 

 public void setSSiteURL(String value) 

 { 

 sSiteURL = value; 

 } 

 public String getSFilePath() 

 { 

 return sFilePath; 

 } 

 public void setSFilePath(String value) 

 { 

 sFilePath = value; 

 } 

 public String getSFileName() 

 { 

 return sFileName; 

 } 

 public void setSFileName(String value) 

 { 

 sFileName = value; 

 } 

 public int getNSplitter() 

 { 

 return nSplitter; 

 } 

 public void setNSplitter(int nCount) 

 { 

 nSplitter = nCount; 

 } 

 } 

 

 /* 

 **Utility.java 

 */ 

 package NetFox; 

 public class Utility { 

 public Utility() 

 { 

 } 

 public static void sleep(int nSecond) 

 { 

 try{ 

 Thread.sleep(nSecond); 

 } 

 catch(Exception e) 

 { 

 e.printStackTrace (); 

 } 

 } 

 public static void log(String sMsg) 

 { 

 System.err.println(sMsg); 

 } 

 public static void log(int sMsg) 

 { 

 System.err.println(sMsg); 

 } 

 } 

 

 /* 

 **TestMethod.java 

 */ 

 package NetFox; 

 public class TestMethod { 

 public TestMethod() 

 { ///xx/weblogic60b2_win.exe 

 try{ 

 SiteInfoBean bean = new SiteInfoBean("http://localhost/xx/weblogic60b2_win.exe",

     "L:\\temp","weblogic60b2_win.exe",5); 

 //SiteInfoBean bean = new SiteInfoBean("http://localhost:8080/down.zip","L:\\temp",

     "weblogic60b2_win.exe",5); 

 SiteFileFetch fileFetch = new SiteFileFetch(bean); 

 fileFetch.start(); 

 } 

 catch(Exception e){e.printStackTrace ();} 

 } 

 public static void main(String[] args) 

 { 

 new TestMethod(); 

 } 

 }

 

你可能感兴趣的:(java)