如果用HttpURLConnection类的方法打开连接，然后用InputStream类获得输入流，再用BufferedInputStream构造出带缓冲区的输入流，如果网速太慢的话，无论缓冲区设置多大，听起来都是断断续续的，达不到真正缓冲的目的。于是尝试编写代码实现用缓冲方式读取远程文件，以下贴出的代码是我写的MP3解码器的一部分。我是不怎么赞同使用多线程下载的，加之有的链接下载速度本身就比较快，所以在下载速度足够的情况下，就让下载线程退出，直到只剩下一个下载线程。多线程中的同步、HttpURLConnection的超时阻塞等因素都会使代码看起来异常复杂。
问题模型 数组buf[]，N个“写”线程从文件顺序读入固定长度数据写入buf[]；一个“读”线程按照文件内容顺序从缓冲区读取，一次读任意长度。一个写线程发生错误或文件读完，这个读线程和所有写线程退出。
针对上述问题，简要介绍一下实现多线程环形缓冲的方法。将缓冲区buf[]分为16块，每块32K，下载线程负责向缓冲区写数据，每次写一块；每次读小于32K的任意字节。线程同步：各个写线程互斥等待空闲块，用信号量机制分配空闲块；各写线程并发填写buf[]；读线程和各写线程并发使用buf[]。
为突出主题略去了一些和本文无关的代码。
一、HttpReader类功能：HTTP协议从指定URL读取数据。
package instream;import java.io.IOException;import java.io.InputStream;import java.net.HttpURLConnection;import java.net.URL;public final class HttpReader { public static final int MAX_RETRY = 10; private URL url; private HttpURLConnection httpConnection; private InputStream in_stream; private long cur_pos; //决定seek方法中是否执行文件读取定位 private int connect_timeout; private int read_timeout; public HttpReader(URL u) { this(u, 5000, 5000); } public HttpReader(URL u, int connect_timeout, int read_timeout) { this.connect_timeout = connect_timeout; this.read_timeout = read_timeout; url = u; } public int read(byte[] b, int off, int len) throws IOException { int r = in_stream.read(b, off, len); cur_pos += r; return r; } /* * 抛出异常通知重试. * 例如响应码503可能是由某种暂时的原因引起的,同一IP频繁的连接请求会遭服务器拒绝. */ public void seek(long start_pos) throws IOException { if (start_pos == cur_pos && in_stream != null) return; if (httpConnection != null) { httpConnection.disconnect(); httpConnection = null; } if (in_stream != null) { in_stream.close(); in_stream = null; } httpConnection = (HttpURLConnection) url.openConnection(); httpConnection.setConnectTimeout(connect_timeout); httpConnection.setReadTimeout(read_timeout); String sProperty = "bytes=" + start_pos + "-"; httpConnection.setRequestProperty("Range", sProperty); //httpConnection.setRequestProperty("Connection", "Keep-Alive"); int responseCode = httpConnection.getResponseCode(); if (responseCode < 200 || responseCode >= 300) { try { Thread.sleep(200); } catch (InterruptedException e) { e.printStackTrace(); } throw new IOException("HTTP responseCode="+responseCode); } in_stream = httpConnection.getInputStream(); cur_pos = start_pos; }}
二、IWriterCallBack接口
package instream;/* * 读/写通信接口.类似于C++的回调函数 * * 例: * class BuffRandAcceURL 内实现本接口的方法tryWriting() * class BuffRandAcceURL 内new Writer(this, ...)传值到Writer.icb * class Writer 内调用icb.tryWriting() */public interface IWriterCallBack { public int tryWriting() throws InterruptedException; public void updateBuffer(int i, int len); public void updateWriterCount(); public int getWriterCount(); public void terminateWriters();}
三、Writer类：下载线程，负责向buf[]写数据。
package instream;import java.net.URL;public final class Writer implements Runnable { private static boolean isalive = true; // 一个线程超时其它线程也能退出 private static byte[] buf; private static IWriterCallBack icb; private HttpReader hr; public Writer(IWriterCallBack cb, URL u, byte[] b, int i) { hr = new HttpReader(u); icb = cb; buf = b; Thread t = new Thread(this,"dt_"+i); t.setPriority(Thread.NORM_PRIORITY + 1); t.start(); } public void run() { int wbytes=0, wpos=0, rema = 0, retry = 0; int idxmask = BuffRandAcceURL.UNIT_COUNT - 1; boolean cont = true; int index = 0; //buf[]内"块"索引号 int startpos = 0; //index对应的文件位置(相对于文件首的偏移量) long time0 = 0; while (cont) { try { // 1.等待空闲块 if(retry == 0) { if ((startpos = icb.tryWriting()) == -1) break; index = (startpos >> BuffRandAcceURL.UNIT_LENGTH_BITS) & idxmask; wpos = startpos & BuffRandAcceURL.BUF_LENGTH_MASK; wbytes = 0; rema = BuffRandAcceURL.UNIT_LENGTH; time0 = System.currentTimeMillis(); } // 2.定位 hr.seek(startpos); // 3.下载"一块" int w; while (rema > 0 && isalive) { w = (rema < 2048) ? rema : 2048; //每次读几K合适? if ((w = hr.read(buf, wpos, w)) == -1) { cont = false; break; } rema -= w; wpos += w; startpos += w; // 能断点续传 wbytes += w; } // 下载速度足够快就结束本线程 long t = System.currentTimeMillis() - time0; if(icb.getWriterCount() > 1 && t < 500) cont = false; //4.通知"读"线程 retry = 0; icb.updateBuffer(index, wbytes); } catch (Exception e) { if(++retry == HttpReader.MAX_RETRY) { isalive = false; icb.terminateWriters(); break; } } } icb.updateWriterCount(); try { hr.close(); } catch (Exception e) {} hr = null; }}
四、BuffRandAcceURL类功能：创建下载线程；read方法从buf[]读数据。关键是如何简单有效防止死锁？
package instream;import java.net.URL;public final class BuffRandAcceURL implements IWriterCallBack { public static final int UNIT_LENGTH_BITS = 15; public static final int UNIT_LENGTH = 1 << UNIT_LENGTH_BITS; //2^16=32K public static final int BUF_LENGTH = UNIT_LENGTH << 4; public static final int UNIT_COUNT = BUF_LENGTH >> UNIT_LENGTH_BITS; //16块 public static final int BUF_LENGTH_MASK = (BUF_LENGTH - 1); private static final int MAX_WRITER = 5; private static long file_pointer; private static int read_pos; private static int fill_bytes; private static byte[] buf; //同时作写线程同步锁 private static int[] unit_bytes; //同时作读线程互斥锁 private static int alloc_pos; private static URL url; private static boolean isalive = true; private static int writer_count; private static long file_length; private static long frame_bytes; private static int free_unit = UNIT_COUNT; // "信号量"计数器 public BuffRandAcceURL(String sURL) throws Exception { this(sURL,MAX_WRITER); } public BuffRandAcceURL(String sURL, int download_threads) throws Exception { buf = new byte[BUF_LENGTH]; unit_bytes = new int[UNIT_COUNT]; url = new URL(sURL); // 创建"写"线程 writer_count = download_threads; for (int i = 0; i < download_threads; i++) { new Writer(this, url, buf, i + 1); Thread.sleep(200); } try_cache(); } /* * 缓冲 */ private void try_cache() throws InterruptedException { int cache_size = BUF_LENGTH; int bi = unit_bytes[read_pos >> UNIT_LENGTH_BITS]; if(bi != 0) cache_size -= UNIT_LENGTH - bi; while (fill_bytes < cache_size) { if (writer_count == 0 || isalive == false) return; synchronized (unit_bytes) { unit_bytes.wait(200); //wait(200)错过通知也可结束循环? } } } private int try_reading(int i, int len) throws Exception { int n = (i + 1) & (UNIT_COUNT - 1); int r = (unit_bytes[i] > 0) ? (unit_bytes[i] + unit_bytes[n]) : unit_bytes[i]; if (r < len) { if (writer_count == 0 || isalive == false) return r; try_cache(); } return len; } /* * 各个"写"线程互斥等待空闲块. 空闲块按由小到大的顺序分配. */ public int tryWriting() throws InterruptedException { int ret = -1; synchronized (buf) { while (free_unit == 0 && isalive) buf.wait(); if(alloc_pos >= file_length || isalive == false) return -1; ret = alloc_pos; alloc_pos += UNIT_LENGTH; free_unit--; } return ret; } /* * "写"线程向buf[]写完数据后调用,通知"读"线程 */ public void updateBuffer(int i, int len) { synchronized (unit_bytes) { unit_bytes[i] = len; fill_bytes += len; unit_bytes.notify(); } } /* * "写"线程准备退出时调用 */ public void updateWriterCount() { synchronized (unit_bytes) { writer_count--; unit_bytes.notify(); } } public int getWriterCount() { return writer_count; } /* * read方法内调用 */ public void notifyWriter() { synchronized (buf) { buf.notifyAll(); } } /* * 被某个"写"线程调用,用于终止其它"写"线程;isalive也影响"读"线程流程 */ public void terminateWriters() { synchronized (unit_bytes) { if (isalive) { isalive = false; System.out.println("\n读取文件超时。重试 " + HttpReader.MAX_RETRY + " 次后放弃，请您稍后再试。"); } unit_bytes.notify(); } notifyWriter(); } public int read(byte[] b, int off, int len) throws Exception { int i = read_pos >> UNIT_LENGTH_BITS; // 1.等待有足够内容可读 if(try_reading(i, len) < len || isalive == false) return -1; // 2.读取 int tail = BUF_LENGTH - read_pos; // write_pos != BUF_LENGTH if (tail < len) { System.arraycopy(buf, read_pos, b, off, tail); System.arraycopy(buf, 0, b, off + tail, len - tail); } else System.arraycopy(buf, read_pos, b, off, len); fill_bytes -= len; file_pointer += len; read_pos += len; read_pos &= BUF_LENGTH_MASK; unit_bytes[i] -= len; if (unit_bytes[i] < 0) { int ni = read_pos >> UNIT_LENGTH_BITS; unit_bytes[ni] += unit_bytes[i]; unit_bytes[i] = 0; free_unit++; notifyWriter(); } else if (unit_bytes[i] == 0) { free_unit++; // 空闲块"信号量"计数加1 notifyWriter(); // 3.通知 } // 如果下一块未填满,意味着文件读完,第1步已处理一次读空两块的情况 return len; }}
本文是JAVA开源项目jmp123源代码的一部分，单独成文旨在与朋友们交流文中提出的问题模型的解决方法。
【jmp123下载地址】