重庆分公司,新征程启航
为企业提供网站建设、域名注册、服务器等服务
这篇文章将为大家详细讲解有关基于Java实现多线程下载并允许断点续传的方法,小编觉得挺实用的,因此分享给大家做个参考,希望大家阅读完这篇文章后可以有所收获。
专注于为中小企业提供网站建设、网站制作服务,电脑端+手机端+微信端的三站合一,更高效的管理,为中小企业青秀免费做网站提供优质的服务。我们立足成都,凝聚了一批互联网行业人才,有力地推动了上千企业的稳健成长,帮助中小企业通过网站建设实现规模扩充和转变。
Java是一门面向对象编程语言,可以编写桌面应用程序、Web应用程序、分布式系统和嵌入式系统应用程序。
多线程下载及断点续传的实现是使用 HTTP/1.1 引入的 Range 请求参数,可以访问Web资源的指定区间的内容。虽然实现了多线程及断点续传,但还有很多不完善的地方。
包含四个类:
Downloader: 主类,负责分配任务给各个子线程,及检测进度DownloadFile: 表示要下载的哪个文件,为了能写输入到文件的指定位置,使用 RandomAccessFile 类操作文件,多个线程写同一个文件需要保证线程安全,这里直接调用 getChannel 方法,获取一个文件通道,FileChannel是线程安全的。DownloadTask: 实际执行下载的线程,获取 [lowerBound, upperBound] 区间的数据,当下载过程中出现异常时要通知其他线程(使用 AtomicBoolean),结束下载Logger: 实时记录下载进度,以便续传时知道从哪开始。感觉这里做的比较差,为了能实时写出日志及方便地使用Properties类的load/store方法格式化输入输出,每次都是打开后再关闭。
演示:
随便找一个文件下载:
强行结束程序并重新运行:
日志文件:
断点续传的关键是记录各个线程的下载进度,这里细节比较多,花了很久。只需要记录每个线程请求的Range区间极客,每次成功写数据到文件时,就更新一次下载区间。下面是下载完成后的日志内容。
代码:
Downloader.java
package downloader; import java.io.*; import java.net.*; import java.nio.file.Files; import java.nio.file.Path; import java.util.concurrent.atomic.AtomicBoolean; public class Downloader { private static final int DEFAULT_THREAD_COUNT = 4; // 默认线程数量 private AtomicBoolean canceled; // 取消状态,如果有一个子线程出现异常,则取消整个下载任务 private DownloadFile file; // 下载的文件对象 private String storageLocation; private final int threadCount; // 线程数量 private long fileSize; // 文件大小 private final String url; private long beginTime; // 开始时间 private Logger logger; public Downloader(String url) { this(url, DEFAULT_THREAD_COUNT); } public Downloader(String url, int threadCount) { this.url = url; this.threadCount = threadCount; this.canceled = new AtomicBoolean(false); this.storageLocation = url.substring(url.lastIndexOf('/')+1); this.logger = new Logger(storageLocation + ".log", url, threadCount); } public void start() { boolean reStart = Files.exists(Path.of(storageLocation + ".log")); if (reStart) { logger = new Logger(storageLocation + ".log"); System.out.printf("* 继续上次下载进度[已下载:%.2fMB]:%s\n", logger.getWroteSize() / 1014.0 / 1024, url); } else { System.out.println("* 开始下载:" + url); } if (-1 == (this.fileSize = getFileSize())) return; System.out.printf("* 文件大小:%.2fMB\n", fileSize / 1024.0 / 1024); this.beginTime = System.currentTimeMillis(); try { this.file = new DownloadFile(storageLocation, fileSize, logger); if (reStart) { file.setWroteSize(logger.getWroteSize()); } // 分配线程下载 dispatcher(reStart); // 循环打印进度 printDownloadProgress(); } catch (IOException e) { System.err.println("x 创建文件失败[" + e.getMessage() + "]"); } } /** * 分配器,决定每个线程下载哪个区间的数据 */ private void dispatcher(boolean reStart) { long blockSize = fileSize / threadCount; // 每个线程要下载的数据量 long lowerBound = 0, upperBound = 0; long[][] bounds = null; int threadID = 0; if (reStart) { bounds = logger.getBounds(); } for (int i = 0; i < threadCount; i++) { if (reStart) { threadID = (int)(bounds[i][0]); lowerBound = bounds[i][1]; upperBound = bounds[i][2]; } else { threadID = i; lowerBound = i * blockSize; // fileSize-1 !!!!! fu.ck,找了一下午的错 upperBound = (i == threadCount - 1) ? fileSize-1 : lowerBound + blockSize; } new DownloadTask(url, lowerBound, upperBound, file, canceled, threadID).start(); } } /** * 循环打印进度,直到下载完毕,或任务被取消 */ private void printDownloadProgress() { long downloadedSize = file.getWroteSize(); int i = 0; long lastSize = 0; // 三秒前的下载量 while (!canceled.get() && downloadedSize < fileSize) { if (i++ % 4 == 3) { // 每3秒打印一次 System.out.printf("下载进度:%.2f%%, 已下载:%.2fMB,当前速度:%.2fMB/s\n", downloadedSize / (double)fileSize * 100 , downloadedSize / 1024.0 / 1024, (downloadedSize - lastSize) / 1024.0 / 1024 / 3); lastSize = downloadedSize; i = 0; } try { Thread.sleep(1000); } catch (InterruptedException ignore) {} downloadedSize = file.getWroteSize(); } file.close(); if (canceled.get()) { try { Files.delete(Path.of(storageLocation)); } catch (IOException ignore) { } System.err.println("x 下载失败,任务已取消"); } else { System.out.println("* 下载成功,本次用时"+ (System.currentTimeMillis() - beginTime) / 1000 +"秒"); } } /** * @return 要下载的文件的尺寸 */ private long getFileSize() { if (fileSize != 0) { return fileSize; } HttpURLConnection conn = null; try { conn = (HttpURLConnection)new URL(url).openConnection(); conn.setConnectTimeout(3000); conn.setRequestMethod("HEAD"); conn.connect(); System.out.println("* 连接服务器成功"); } catch (MalformedURLException e) { throw new RuntimeException("URL错误"); } catch (IOException e) { System.err.println("x 连接服务器失败["+ e.getMessage() +"]"); return -1; } return conn.getContentLengthLong(); } public static void main(String[] args) throws IOException { new Downloader("http://js.xiazaicc.com//down2/ucliulanqi_downcc.zip").start(); } }
DownloadTask.java
package downloader; import java.io.*; import java.net.HttpURLConnection; import java.net.URL; import java.nio.ByteBuffer; import java.nio.channels.Channels; import java.nio.channels.ReadableByteChannel; import java.util.concurrent.atomic.AtomicBoolean; class DownloadTask extends Thread { private final String url; private long lowerBound; // 下载的文件区间 private long upperBound; private AtomicBoolean canceled; private DownloadFile downloadFile; private int threadId; DownloadTask(String url, long lowerBound, long upperBound, DownloadFile downloadFile, AtomicBoolean canceled, int threadID) { this.url = url; this.lowerBound = lowerBound; this.upperBound = upperBound; this.canceled = canceled; this.downloadFile = downloadFile; this.threadId = threadID; } @Override public void run() { ReadableByteChannel input = null; try { ByteBuffer buffer = ByteBuffer.allocate(1024 * 1024 * 2); // 2MB input = connect(); System.out.println("* [线程" + threadId + "]连接成功,开始下载..."); int len; while (!canceled.get() && lowerBound <= upperBound) { buffer.clear(); len = input.read(buffer); downloadFile.write(lowerBound, buffer, threadId, upperBound); lowerBound += len; } if (!canceled.get()) { System.out.println("* [线程" + threadId + "]下载完成" + ": " + lowerBound + "-" + upperBound); } } catch (IOException e) { canceled.set(true); System.err.println("x [线程" + threadId + "]遇到错误[" + e.getMessage() + "],结束下载"); } finally { if (input != null) { try { input.close(); } catch (IOException e) { e.printStackTrace(); } } } } /** * 连接WEB服务器,并返回一个数据通道 * @return 返回通道 * @throws IOException 网络连接错误 */ private ReadableByteChannel connect() throws IOException { HttpURLConnection conn = (HttpURLConnection)new URL(url).openConnection(); conn.setConnectTimeout(3000); conn.setRequestMethod("GET"); conn.setRequestProperty("Range", "bytes=" + lowerBound + "-" + upperBound); // System.out.println("thread_"+ threadId +": " + lowerBound + "-" + upperBound); conn.connect(); int statusCode = conn.getResponseCode(); if (HttpURLConnection.HTTP_PARTIAL != statusCode) { conn.disconnect(); throw new IOException("状态码错误:" + statusCode); } return Channels.newChannel(conn.getInputStream()); } }
DownloadFile.java
package downloader; import java.io.IOException; import java.io.RandomAccessFile; import java.nio.ByteBuffer; import java.nio.channels.FileChannel; import java.util.concurrent.atomic.AtomicLong; class DownloadFile { private final RandomAccessFile file; private final FileChannel channel; // 线程安全类 private AtomicLong wroteSize; // 已写入的长度 private Logger logger; DownloadFile(String fileName, long fileSize, Logger logger) throws IOException { this.wroteSize = new AtomicLong(0); this.logger = logger; this.file = new RandomAccessFile(fileName, "rw"); file.setLength(fileSize); channel = file.getChannel(); } /** * 写数据 * @param offset 写偏移量 * @param buffer 数据 * @throws IOException 写数据出现异常 */ void write(long offset, ByteBuffer buffer, int threadID, long upperBound) throws IOException { buffer.flip(); int length = buffer.limit(); while (buffer.hasRemaining()) { channel.write(buffer, offset); } wroteSize.addAndGet(length); logger.updateLog(threadID, length, offset + length, upperBound); // 更新日志 } /** * @return 已经下载的数据量,为了知道何时结束整个任务,以及统计信息 */ long getWroteSize() { return wroteSize.get(); } // 继续下载时调用 void setWroteSize(long wroteSize) { this.wroteSize.set(wroteSize); } void close() { try { file.close(); } catch (IOException e) { e.printStackTrace(); } } }
Logger.java
package downloader; import java.io.*; import java.util.Properties; class Logger { private String logFileName; // 下载的文件的名字 private Properties log; /** * 重新开始下载时,使用该构造函数 * @param logFileName */ Logger(String logFileName) { this.logFileName = logFileName; log = new Properties(); FileInputStream fin = null; try { log.load(new FileInputStream(logFileName)); } catch (IOException ignore) { } finally { try { fin.close(); } catch (Exception ignore) {} } } Logger(String logFileName, String url, int threadCount) { this.logFileName = logFileName; this.log = new Properties(); log.put("url", url); log.put("wroteSize", "0"); log.put("threadCount", String.valueOf(threadCount)); for (int i = 0; i < threadCount; i++) { log.put("thread_" + i, "0-0"); } } synchronized void updateLog(int threadID, long length, long lowerBound, long upperBound) { log.put("thread_"+threadID, lowerBound + "-" + upperBound); log.put("wroteSize", String.valueOf(length + Long.parseLong(log.getProperty("wroteSize")))); FileOutputStream file = null; try { file = new FileOutputStream(logFileName); // 每次写时都清空文件 log.store(file, null); } catch (IOException e) { e.printStackTrace(); } finally { if (file != null) { try { file.close(); } catch (IOException e) { e.printStackTrace(); } } } } /** * 获取区间信息 * ret[i][0] = threadID, ret[i][1] = lowerBoundID, ret[i][2] = upperBoundID * @return */ long[][] getBounds() { long[][] bounds = new long[Integer.parseInt(log.get("threadCount").toString())][3]; int[] index = {0}; log.forEach((k, v) -> { String key = k.toString(); if (key.startsWith("thread_")) { String[] interval = v.toString().split("-"); bounds[index[0]][0] = Long.parseLong(key.substring(key.indexOf("_") + 1)); bounds[index[0]][1] = Long.parseLong(interval[0]); bounds[index[0]++][2] = Long.parseLong(interval[1]); } }); return bounds; } long getWroteSize() { return Long.parseLong(log.getProperty("wroteSize")); } }
关于“基于Java实现多线程下载并允许断点续传的方法”这篇文章就分享到这里了,希望以上内容可以对大家有一定的帮助,使各位可以学到更多知识,如果觉得文章不错,请把它分享出去让更多的人看到。