性能主题

Netty原理和使用

  Netty是一个高性能 事件驱动的异步的非堵塞的IO(NIO)框架,用于建立TCP等底层的连接,基于Netty可以建立高性能的Http服务器。支持HTTP、 WebSocket 、Protobuf、 Binary TCP |和UDP,Netty已经被很多高性能项目作为其Socket底层基础,如HornetQ Infinispan Vert.x
Play Framework Finangle和 Cassandra。其竞争对手是:Apache MINA和 Grizzly。

   传统堵塞的IO读取如下:

InputStream is = new FileInputStream("input.bin");
int byte = is.read(); // 当前线程等待结果到达直至错误

   而使用NIO如下:

while (true) {
 selector.select(); // 从多个通道请求事件
 Iterator it = selector.selectedKeys().iterator();
 while (it.hasNext()) {
  SelectorKey key = (SelectionKey) it.next();
  handleKey(key);
  it.remove();
 }

堵塞与非堵塞原理

  传统硬件的堵塞如下,从内存中读取数据,然后写到磁盘,而CPU一直等到磁盘写完成,磁盘的写操作是慢的,这段时间CPU被堵塞不能发挥效率。

  使用非堵塞的DMA如下图:CPU只是发出写操作这样的指令,做一些初始化工作,DMA具体执行,从内存中读取数据,然后写到磁盘,当完成写后发出一个中断事件给CPU。这段时间CPU是空闲的,可以做别的事情。这个原理称为Zero.copy零拷贝。

  Netty底层基于上述Java NIO的零拷贝原理实现:

比较

  • Tomcat是一个Web服务器,它是采取一个请求一个线程,当有1000客户端时,会耗费很多内存。通常一个线程将花费 256kb到1mb的stack空间。
  • Node.js是一个线程服务于所有请求,在错误处理上有限制
  • Netty是一个线程服务于很多请求,如下图,当从Java NIO获得一个Selector事件,将激活通道Channel。

演示

Netty的使用代码如下:

Channel channel = ...
ChannelFuture cf = channel.write(data);
cf.addListener(
  new ChannelFutureListener() {
   @Override
   public void operationComplete(ChannelFuture future) throws Exception {
     if(!future.isSuccess() {
        future.cause().printStacktrace();
        ...
     }
     ...
   }
});
...
cf.sync();

通过引入观察者监听,当有数据时,将自动激活监听者中的代码运行。

我们使用Netty建立一个服务器代码:

public class EchoServer {

    private final int port;

    public EchoServer(int port) {
        this.port = port;
    }

    public void run() throws Exception {
        // Configure the server.
        EventLoopGroup bossGroup = new NioEventLoopGroup();
        EventLoopGroup workerGroup = new NioEventLoopGroup();
        try {
            ServerBootstrap b = new ServerBootstrap();
            b.group(bossGroup, workerGroup).channel(NioServerSocketChannel.class).option(ChannelOption.SO_BACKLOG, 100)
                   .handler(new LoggingHandler(LogLevel.INFO)).childHandler(new ChannelInitializer<SocketChannel>() {
                       @Override
                       public void initChannel(SocketChannel ch) throws Exception {
                           ch.pipeline().addLast(
                           // new LoggingHandler(LogLevel.INFO),
                                   new EchoServerHandler());
                       }
                   });

            // Start the server.
            ChannelFuture f = b.bind(port).sync();

            // Wait until the server socket is closed.
            f.channel().closeFuture().sync();
        } finally {
            // Shut down all event loops to terminate all threads.
            bossGroup.shutdownGracefully();
            workerGroup.shutdownGracefully();
        }
    }

   
}

这段代码调用:在9999端口启动

new EchoServer(9999).run();

我们需要完成的代码是EchoServerHandler

public class EchoServerHandler extends ChannelInboundHandlerAdapter {

    private static final Logger logger = Logger.getLogger(EchoServerHandler.class.getName());

    @Override
    public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
        ctx.write(msg);
    }

    @Override
    public void channelReadComplete(ChannelHandlerContext ctx) throws Exception {
        ctx.flush();
    }

    @Override
    public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) {
        // Close the connection when an exception is raised.
        logger.log(Level.WARNING, "Unexpected exception from downstream.", cause);
        ctx.close();
    }
}

原理

   一个Netty服务器的原理如下:

  图中每次请求的读取是通过UpStream来实现,然后激活我们的服务逻辑如EchoServerHandler,而服务器向外写数据,也就是响应是通过DownStream实现的。每个通道Channel包含一对UpStream和DownStream,以及我们的handlers(EchoServerHandler),如下图,这些都是通过channel pipeline封装起来的,数据流在管道里流动,每个Socket对应一个ChannelPipeline。

 

   CHANNELPIPELINE是关键,它类似Unix的管道,有以下作用:

  • 为每个Channel 保留 ChannelHandlers ,如EchoServerHandler
  • 所有的事件都要通过它
  • 不断地修改:类似unix的SH管道: echo "Netty is shit...." | sed -e 's/is /is the /'
  • 一个Channel对应一个 ChannelPipeline
  • 包含协议编码解码 安全验证SSL/TLS和应用逻辑

 

客户端代码

  前面我们演示了服务器端代码,下面是客户端代码:

public class EchoClient {
    private final String host;
    private final int port;
    private final int firstMessageSize;

    public EchoClient(String host, int port, int firstMessageSize) {
        this.host = host;
        this.port = port;
        this.firstMessageSize = firstMessageSize;
    }

    public void run() throws Exception {
        // Configure the client.
        EventLoopGroup group = new NioEventLoopGroup();
        try {
            Bootstrap b = new Bootstrap();
           b.group(group).channel(NioSocketChannel.class).option(ChannelOption.TCP_NODELAY, true).handler(new ChannelInitializer<SocketChannel>() {
                @Override
                public void initChannel(SocketChannel ch) throws Exception {
                   ch.pipeline().addLast(
                   // new LoggingHandler(LogLevel.INFO),
                           new EchoClientHandler(firstMessageSize));
                }
            });

            // Start the client.
            ChannelFuture f = b.connect(host, port).sync();

            // Wait until the connection is closed.
            f.channel().closeFuture().sync();
        } finally {
            // Shut down the event loop to terminate all threads.
            group.shutdownGracefully();
        }
    }
}

客户端的应用逻辑EchoClientHandler

public class EchoClientHandler extends ChannelInboundHandlerAdapter {

    private static final Logger logger = Logger.getLogger(EchoClientHandler.class.getName());

    private final ByteBuf firstMessage;

    /**
     * Creates a client-side handler.
     */
    public EchoClientHandler(int firstMessageSize) {
        if (firstMessageSize <= 0) {
            throw new IllegalArgumentException("firstMessageSize: " + firstMessageSize);
        }
        firstMessage = Unpooled.buffer(firstMessageSize);
        for (int i = 0; i < firstMessage.capacity(); i++) {
            firstMessage.writeByte((byte) i);
        }
    }

    @Override
    public void channelActive(ChannelHandlerContext ctx) {
        ctx.writeAndFlush(firstMessage);
        System.out.print("active");
    }

    @Override
    public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
        ctx.write(msg);
        System.out.print("read");
    }

    @Override
    public void channelReadComplete(ChannelHandlerContext ctx) throws Exception {
        ctx.flush();
        System.out.print("readok");
    }

    @Override
    public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) {
        // Close the connection when an exception is raised.
        logger.log(Level.WARNING, "Unexpected exception from downstream.", cause);
        ctx.close();
    }

}

Node.js的事件轮询Event Loop原理解释

Spray + Akka高性能异步IO并发

Java NIO原理和使用

NodeJS入门

Vertx入门教程

Netty作者谈Java性能优化要点

Ratpack可快速开发异步响应式的Java Web应用