首页 > 代码库 > JDK的BIO, NIO, AIO
JDK的BIO, NIO, AIO
背景知识点我
1. BIO
JDK5之前, JDK的IO模式只有BIO(同步阻塞)
问题: 因为阻塞的存在, 需对每个请求开启一个线程. 过多的线程切换影响操作系统性能
解决: 使用线程池, 处理不过来的放入队列, 再处理不过来的会触发其他机制
问题: 超过线程池数量的请求需要等待
public class Client { final static String ADDRESS = "127.0.0.1"; final static int PORT = 8765; public static void main(String[] args) throws IOException { Socket socket = null; BufferedReader in = null; PrintWriter out = null; try { socket = new Socket(ADDRESS, PORT); in = new BufferedReader(new InputStreamReader(socket.getInputStream())); out = new PrintWriter(socket.getOutputStream(), true); // true自动flush //向服务器端发送数据 out.println("来自客户端的请求"); //从服务端接收数据 String response = in.readLine(); // 阻塞 System.out.println("Client获取数据: " + response); } catch (Exception e) { e.printStackTrace(); } finally { out.close(); in.close(); socket.close(); } } }
服务端1: 一个请求~一个线程
public class Server { final static int PROT = 8765; public static void main(String[] args) throws IOException { ServerSocket server = null; try { server = new ServerSocket(PROT); System.out.println("server start"); while(true){ Socket socket = server.accept(); //监听 阻塞 , socket底层会新建线程处理与客户端的三次握手 //建立线程处理获取的 socket new Thread(new ServerHandler(socket)).start(); } } catch (Exception e) { e.printStackTrace(); } finally { server.close(); } } } class ServerHandler implements Runnable { private Socket socket; public ServerHandler(Socket socket) { this.socket = socket; } @Override public void run() { BufferedReader in = null; PrintWriter out = null; try { in = new BufferedReader(new InputStreamReader(this.socket.getInputStream())); out = new PrintWriter(this.socket.getOutputStream(), true); String body = null; while (true) { body = in.readLine(); // 阻塞 if (body == null) break; System.out.println("Server获取的请求: " + body); out.println("来自服务器的响应"); } } catch (Exception e) { e.printStackTrace(); } finally { try { out.close(); in.close(); socket.close(); } catch (IOException e) { e.printStackTrace(); } } } }
服务端2: 用线程池处理请求
public class Server { final static int PORT = 8765; public static void main(String[] args) throws IOException { ServerSocket server = null; try { server = new ServerSocket(PORT); System.out.println("server start"); HandlerExecutorPool executorPool = new HandlerExecutorPool(50, 1000); while(true){ Socket socket = server.accept(); executorPool.execute(new ServerHandler(socket)); } } catch (Exception e) { e.printStackTrace(); } finally { server.close(); } } } class HandlerExecutorPool { private ExecutorService executor; public HandlerExecutorPool(int maxPoolSize, int queueSize){ this.executor = new ThreadPoolExecutor( // 带阻塞队列的线程池 Runtime.getRuntime().availableProcessors(), // 初始线程数 maxPoolSize, // 线程数上限 如果要处理请求的Runnable对象装满了队列, 则提高现有线程数 120L, // 如在120个时间颗粒内某线程是空闲的, 将被回收 TimeUnit.SECONDS, new ArrayBlockingQueue<Runnable>(queueSize) // 存放处理请求的Runnable对象 ); } public void execute(Runnable task){ this.executor.execute(task); } } class ServerHandler implements Runnable { private Socket socket; public ServerHandler(Socket socket) { this.socket = socket; } @Override public void run() { BufferedReader in = null; PrintWriter out = null; try { in = new BufferedReader(new InputStreamReader(this.socket.getInputStream())); out = new PrintWriter(this.socket.getOutputStream(), true); String body = null; while (true) { body = in.readLine(); if (body == null) break; System.out.println("Server获取的请求: " + body); // 阻塞 out.println("来自服务器的响应"); } } catch (Exception e) { e.printStackTrace(); } finally { try { out.close(); in.close(); socket.close(); } catch (IOException e) { e.printStackTrace(); } } } }
2.NIO1.0
JDK5以后引入了NIO1.0(多路复用机制)
伴随多路复用在程序中引入了如下概念:
Channel(通道):TCP连接的抽象,一个TCP连接对应多个Channel,这样减少TCP的连接次数。
通道与BIO中socket类似
通道与BIO中的流类似, 不过channel是双向的而流是单向的
channel有多种状态位, 能被selector识别
Buffer(缓冲区):
缓冲区是一块内存区域(数组), 在NIO中被包装成Buffer对象. Buffer提供方法用来访问该内存。
BIO中,数据存储在流中,而NIO中,数据存储在缓冲区中。
除了boolean的其他java七种基本类型都有相应的Buffer类. 最常使用的是ByteBuffer
Selector(多路复用器):负责轮询所有注册通道,根据通道状态执行相关操作。状态包括:Connect,Accept,Read,Write。
在"四种常用IO模型"里提过用select系统调用实现IO多路复用. 除select外Linux还提供了poll/epoll函数, 其中select/poll函数按顺序扫描文件句柄是否就绪,支持的文件句柄数有限; 而epoll使用基于事件驱动方式替代顺序扫描,性能更高, 对文件句柄数没有数量限制. JDK的Selector使用了epoll, 只需要一个线程轮询, 就可以接入大量的客户端.
public class Client { public static void main(String[] args) throws IOException { SocketChannel sc = null; ByteBuffer writeBuf = ByteBuffer.allocate(1024); ByteBuffer readBuf = ByteBuffer.allocate(1024); try { //创建通道 sc = SocketChannel.open(); //进行连接 sc.connect(new InetSocketAddress("127.0.0.1", 8765)); // 下面步骤可以用selector轮询代替 while(true){ //定义一个字节数组,然后使用系统录入功能: byte[] bytes1 = new byte[1024]; System.in.read(bytes1); //阻塞 //把数据放到缓冲区中 writeBuf.put(bytes1); //对缓冲区进行复位 writeBuf.flip(); //写出数据 sc.write(writeBuf); //清空缓冲区 writeBuf.clear(); // 接收服务端响应 sc.read(readBuf); readBuf.flip(); byte[] bytes2 = new byte[readBuf.remaining()]; readBuf.get(bytes2); readBuf.clear(); String body = new String(bytes2); System.out.println("Client获取数据: " + body); } } catch (IOException e) { e.printStackTrace(); } finally { sc.close(); } } }
通过改变Selector监听Channel的状态位, 控制与客户端读写的先后顺序
public class Server implements Runnable{ private Selector seletor; private ByteBuffer readBuf = ByteBuffer.allocate(1024); private ByteBuffer writeBuf = ByteBuffer.allocate(1024); public Server(int port){ try { //1 创建多路复用器selector this.seletor = Selector.open(); //2 创建ServerSocket通道 ServerSocketChannel ssc = ServerSocketChannel.open(); //3 设置通道是否阻塞, 决定了通道了read/write/accept/connect方法是否阻塞 ssc.configureBlocking(false); //4 设置通道地址 ssc.bind(new InetSocketAddress(port)); //5 将ServerSocket通道注册到selector上, 指定监听其accept事件 ssc.register(this.seletor, SelectionKey.OP_ACCEPT); System.out.println("Server start"); } catch (IOException e) { e.printStackTrace(); } } @Override public void run() { while(true){ try { // select阻塞, 监听相关事件 this.seletor.select(); // 解除阻塞, 返回选择key, key含有通道, 状态等信息 Iterator<SelectionKey> keysIter = this.seletor.selectedKeys().iterator(); // 进行遍历 while(keysIter.hasNext()){ SelectionKey key = keysIter.next(); keysIter.remove(); if (key.isValid()) { // 等待接收连接状态 if (key.isAcceptable()) { accept(key); } // 可读状态 if (key.isReadable()) { read(key); } if (key.isWritable()) { write(key); } } } } catch (IOException e) { e.printStackTrace(); } } } private void write(SelectionKey key) { try { // 获取通道 SocketChannel sc = (SocketChannel) key.channel(); // 写回给客户端数据 writeBuf.put("来自服务器的响应".getBytes()); writeBuf.flip(); sc.write(writeBuf); writeBuf.clear(); // 修改监听的状态位, 如果保持OP_WRITE会导致重复写 key.interestOps(SelectionKey.OP_READ); } catch (IOException e) { e.printStackTrace(); } } private void read(SelectionKey key) { try { // 获取通道 SocketChannel sc = (SocketChannel) key.channel(); // 读取数据, 读到buffer. 按程序运行顺序, 这里sc是否设置为阻塞效果都一样 int count = sc.read(this.readBuf); // readBuf写时会改变position的值 if (count == -1) { key.channel().close(); key.cancel(); //取消该通道在selector的注册, 之后不会被select轮询到 return; } // 有数据则进行读取. 读取前需要将position和limit进行复位 readBuf.flip(); // 根据缓冲区的数据长度创建相应大小的byte数组, 接收缓冲区的数据 byte[] bytes = new byte[this.readBuf.remaining()]; // 接收缓冲区数据 readBuf.get(bytes); readBuf.clear(); String body = new String(bytes).trim(); System.out.println("Server获取的请求: " + body); // 如果保持OP_READ会导致重复读 sc.register(this.seletor, SelectionKey.OP_WRITE); } catch (IOException e) { e.printStackTrace(); } } private void accept(SelectionKey key) { try { // 获取服务通道 ServerSocketChannel ssc = (ServerSocketChannel) key.channel(); // 获取客户端通道. SocketChannel sc = ssc.accept(); // 设置非阻塞模式 sc.configureBlocking(false); // 将客户端通道注册到多路复用器上,指定监听事件 sc.register(this.seletor, SelectionKey.OP_READ | SelectionKey.OP_WRITE); } catch (IOException e) { e.printStackTrace(); } } public static void main(String[] args) { new Thread(new Server(8765)).start();; } }
BIO客户端与NIO服务端通信需注意的:
BIO服务端, 一次IO有明确的结束点, 客户端再次read会返回-1
NIO服务端一次IO结束后, 没有关闭通道, 它可能把通道从读状态转为写状态. 于是selector不监听读了, 客户端再次read什么都没返回, 就会阻塞.
3.NIO2.0
JDK7引入了NIO2.0(即AIO)
NIO1.0中, IO过程没有阻塞, 阻塞被转移到了Selector轮询上. Selector管理所有的Channel, 因此能把总阻塞时间缩到最短.
NIO2.0中, 供我们调用的IO API都是非阻塞的, 背后复杂的实现过程(肯定有阻塞)被转移到了JDK底层和操作系统上. 我们的程序的IO调用可以做到立即返回.
同样有Channel和Buffer, 但没有Selector
public class Server { //线程池 private ExecutorService executorService; //异步通道线程组 private AsynchronousChannelGroup threadGroup; //服务器通道 public AsynchronousServerSocketChannel assc; public Server(int port){ try { //创建一个线程池 executorService = Executors.newCachedThreadPool(); //使用线程池创建异步通道线程组, 该线程组在底层支持着我们的异步操作 threadGroup = AsynchronousChannelGroup.withCachedThreadPool(executorService, 1); //使用 异步通道线程组 创建服务器通道 assc = AsynchronousServerSocketChannel.open(threadGroup); //给通道绑定端口 assc.bind(new InetSocketAddress(port)); System.out.println("server start"); // 下面的accept不会阻塞 , 一个accept只能接收一个连接请求 // accept第一个参数: 被绑定到IO操作的关联对象(子类), 第二个参数 CompletionHandler<AsynchronousSocketChannel, 关联对象(父类)>, 操作成功后执行的回调句柄 // 如果接受了一个新的连接, 其结果AsynchronousSocketChannel会被绑定与assc通道到相同的AsynchronousChannelGroup assc.accept(this, new ServerCompletionHandler()); // 这里为了避免程序结束, 异步通道线程组结束就不会执行回调了 Thread.sleep(Integer.MAX_VALUE); } catch (Exception e) { e.printStackTrace(); } } public static void main(String[] args) { new Server(8765); } }
//第一个参数: IO操作结果; 第二个参数: 被绑定到IO操作的关联对象 public class ServerCompletionHandler implements CompletionHandler<AsynchronousSocketChannel, Server> { // 以下两个重载参数与CompletionHander的模板参数一致, 回调时被传入IO结果和IO操作时设置的关联对象 @Override public void completed(AsynchronousSocketChannel asc, Server attachment) { // 完成当前连接时, 首先, 为下一个客户端能接入再次调用accept异步方法 attachment.assc.accept(attachment, this); // 其次, 执行下一步的读操作 read(asc); } @Override public void failed(Throwable exc, Server attachment) { exc.printStackTrace(); } private void read(final AsynchronousSocketChannel asc) { //读取数据 ByteBuffer buf = ByteBuffer.allocate(1024); // 第一个参数: 读操作的Buffer, 第二个参数: IO关联对象, 第三个参数:CompletionHandler<Integer, IO管理对象父类> asc.read(buf, buf, new CompletionHandler<Integer, ByteBuffer>() { @Override public void completed(Integer resultSize, ByteBuffer attachment) { //进行读取之后,重置标识位 attachment.flip(); //获得读取的字节数 System.out.println("Server端" + "收到客户端的数据长度为:" + resultSize); //获取读取的数据 String resultData = http://www.mamicode.com/new String(attachment.array()).trim(); System.out.println("Server端" + "收到客户端的数据信息为:" + resultData); String response = "From服务端To客户端: 于" + new Date() + "收到了请求数据"+ resultData; write(asc, response); } @Override public void failed(Throwable exc, ByteBuffer attachment) { exc.printStackTrace(); } }); } private void write(AsynchronousSocketChannel asc, String response) { try { ByteBuffer buf = ByteBuffer.allocate(1024); buf.put(response.getBytes()); buf.flip(); // 写操作, 异步 Future<Integer> future = asc.write(buf); // 阻塞等待结果 future.get(); } catch (InterruptedException e) { e.printStackTrace(); } catch (ExecutionException e) { e.printStackTrace(); } } }
public class Client { private AsynchronousSocketChannel asc ; public Client() throws Exception { asc = AsynchronousSocketChannel.open(); } public void connect() throws InterruptedException, ExecutionException{ // get()阻塞 asc.connect(new InetSocketAddress("127.0.0.1", 8765)).get(); } public void write(String request){ try { // get()阻塞 asc.write(ByteBuffer.wrap(request.getBytes())).get(); read(); } catch (Exception e) { e.printStackTrace(); } } private void read() throws IOException { ByteBuffer buf = ByteBuffer.allocate(1024); try { // get()阻塞 asc.read(buf).get(); buf.flip(); byte[] respByte = new byte[buf.remaining()]; buf.get(respByte); System.out.println(new String(respByte,"utf-8").trim()); // 关闭 asc.close(); } catch (InterruptedException e) { e.printStackTrace(); } catch (ExecutionException e) { e.printStackTrace(); } catch (UnsupportedEncodingException e) { e.printStackTrace(); } } public static void main(String[] args) throws Exception { Client c1 = new Client(); Client c2 = new Client(); c1.connect(); c2.connect(); c1.write("aa"); c2.write("bbb"); } }
JDK的BIO, NIO, AIO