Java编程方法论-Spring WebFlux篇 Reactor-Netty下TcpServer的功能实现 1

本书主要针对Netty服务器来说，因此读者应具有有关Netty的基本知识和应用技能。接下来，咱们将对Reactor-netty从设计到实现的细节一一探究，让你们真的从中学习到好的封装设计理念。本书在写时所参考的最新版本是Reactor-netty 0.7.8.Release这个版本，但如今已有0.8版本，并且0.7与0.8版本在源码细节有不小的变更，这点给你们提醒下。我会针对0.8版本进行全新的解读并在将来出版的书中进行展现。

TcpServer的功能实现

这里，咱们会首先解读Reactor Netty是如何针对Netty中Bootstrap的ChildHandler进行封装以及响应式拓展等一些细节探究。接着，咱们会涉及到HttpHandler的引入，以此来对接咱们上层web服务。

针对Bootstrap的ChildHandler的封装

由于这是咱们切入自定义逻辑的地方，因此，咱们首先来关注下与其相关的ChannelHandler，以及前文并未提到的，服务器究竟是如何启动以及如何经过响应式来作到优雅的关闭，首先咱们会接触关闭服务器的设定。

ChannelHandler引入与使用响应式优雅关闭服务器

咱们再回到reactor.ipc.netty.http.server.HttpServer#HttpServer这个构造器中，由上一章咱们知道请求是HTTP层面的(应用层)，必须依赖于TCP的链接实现，因此这里就要有一个TCPServer的实现，其实就是Channel上Pipeline的操做。

//reactor.ipc.netty.http.server.HttpServer#HttpServer
private HttpServer(HttpServer.Builder builder) {
    HttpServerOptions.Builder serverOptionsBuilder = HttpServerOptions.builder();
            ...
    this.options = serverOptionsBuilder.build();
    this.server = new TcpBridgeServer(this.options);
}
复制代码

这里的话在DiscardServer Demo中，TCPServer咱们主要针对childHandler的内容的封装，也就是以下内容:

b.group(bossGroup, workerGroup)
    ...
 .childHandler(new ChannelInitializer<SocketChannel>() { 
            @Override
            public void initChannel(SocketChannel ch) throws Exception {
                ch.pipeline().addLast(new DiscardServerHandler());
            }
        })
    ...
复制代码

那childHandler到底表明什么类型，咱们能够在io.netty.bootstrap.ServerBootstrap找到其相关定义:

//io.netty.bootstrap.ServerBootstrap
public class ServerBootstrap extends AbstractBootstrap<ServerBootstrap, ServerChannel> {

    private static final InternalLogger logger = InternalLoggerFactory.getInstance(ServerBootstrap.class);

    private final Map<ChannelOption<?>, Object> childOptions = new LinkedHashMap<ChannelOption<?>, Object>();
    private final Map<AttributeKey<?>, Object> childAttrs = new LinkedHashMap<AttributeKey<?>, Object>();
    private final ServerBootstrapConfig config = new ServerBootstrapConfig(this);
    private volatile EventLoopGroup childGroup;
    private volatile ChannelHandler childHandler;

    public ServerBootstrap() { }

    private ServerBootstrap(ServerBootstrap bootstrap) {
        super(bootstrap);
        childGroup = bootstrap.childGroup;
        childHandler = bootstrap.childHandler;
        synchronized (bootstrap.childOptions) {
            childOptions.putAll(bootstrap.childOptions);
        }
        synchronized (bootstrap.childAttrs) {
            childAttrs.putAll(bootstrap.childAttrs);
        }
    }
...
}
复制代码

由字段定义可知，childHandler表明的是ChannelHandler，顾名思义，是关于Channel的一个处理类，这里经过查看其定义可知它是用来拦截处理Channel中的I/O事件，并经过Channel下的ChannelPipeline将处理后的事件转发到其下一个处理程序中。那这里如何实现DiscardServer Demo中的b.childHandler(xxx)行为，经过DiscardServer Demo咱们能够知道，咱们最关注的实际上是ch.pipeline().addLast(new DiscardServerHandler());中的DiscardServerHandler实现，可是咱们发现，这个核心语句是包含在ChannelInitializer内，其继承了ChannelInboundHandlerAdapter，它的最顶层的父类接口就是ChannelHandler，也就对应了io.netty.bootstrap.ServerBootstrap在执行b.childHandler(xxx)方法时，其须要传入ChannelHandler类型的设定。这里就能够分拆成两步来作，一个是b.childHandler(xxx)行为包装，一个是此ChannelHandler的定义拓展实现。那么，为了API的通用性，咱们先来看Netty的客户端的创建的一个Demo(摘自本人RPC项目的一段代码):

private Channel createNewConChannel() {
    Bootstrap bootstrap = new Bootstrap();
    bootstrap.channel(NioSocketChannel.class)
                .group(new NioEventLoopGroup(1))
                .handler(new ChannelInitializer<Channel>() {
                    protected void initChannel(Channel ch) throws Exception {
                        ch.pipeline().addLast(new LoggingHandler(LogLevel.INFO))
                        .addLast(new RpcDecoder(10 * 1024 * 1024))
                        .addLast(new RpcEncoder())
                        .addLast(new RpcClientHandler())
                        ;
                    }
                });
    try {
        final ChannelFuture f =
         bootstrap.option(ChannelOption.CONNECT_TIMEOUT_MILLIS, 3000)
                  .option(ChannelOption.TCP_NODELAY, true)
                  .connect(ip, port).sync(); // <1>
        f.addListener((ChannelFutureListener) future -> {
            if (future.isSuccess()) {
                LOGGER.info("Connect success {} ", f);
            }
        });
        final Channel channel = f.channel();
        channel.closeFuture().addListener((ChannelFutureListener) future -> LOGGER.info("Channel Close {} {}", ip, port));
        return channel;
    } catch (InterruptedException e) {
        e.printStackTrace();
    }
    return null;
}
复制代码

将Netty的客户端与服务端的创建进行对比，咱们能够发现b.childHandler(xxx)与相应的启动(Server端的话是serverBootstrap.bind(port).sync();，客户端的话是上述Demo中<1>处的内容)均可以抽取出来做为一个接口来进行功能的聚合，而后和相应的Server(如TcpServer)或Client(如TcpClient)进行其特有的实现。在Reactor Netty内的话，就是定义一个reactor.ipc.netty.NettyConnector接口，除了作到上述的功能以外，为了适配响应式的理念，也进行了响应式的设计。即在netty客户端与服务端在启动时，能够保存其状态，以及提供结束的对外接口方法，这种在响应式中能够很优雅的实现。接下来，咱们来看此reactor.ipc.netty.NettyConnector的接口定义:

//reactor.ipc.netty.NettyConnector
public interface NettyConnector<INBOUND extends NettyInbound, OUTBOUND extends NettyOutbound> {

Mono<? extends NettyContext> newHandler(BiFunction<? super INBOUND, ? super OUTBOUND, ? extends Publisher<Void>> ioHandler);
...
default <T extends BiFunction<INBOUND, OUTBOUND, ? extends Publisher<Void>>>
BlockingNettyContext start(T handler) {
    return new BlockingNettyContext(newHandler(handler), getClass().getSimpleName());
}
}
...
default <T extends BiFunction<INBOUND, OUTBOUND, ? extends Publisher<Void>>>
void startAndAwait(T handler, @Nullable Consumer<BlockingNettyContext> onStart) {
    BlockingNettyContext facade = new BlockingNettyContext(newHandler(handler), getClass().getSimpleName());

    facade.installShutdownHook();

    if (onStart != null) {
        onStart.accept(facade);
    }

    facade.getContext()
            .onClose()
            .block();
}
复制代码

其中，newHandler能够是咱们上层web处理，里面包含了INBOUND, OUTBOUND，具体的话就是request，response，后面会专门来涉及到这点。接着就是提供了一个启动方法start，其内建立了一个BlockingNettyContext实例，而逻辑的核心就在其构造方法内，就是要将配置好的服务器启动，整个启动过程仍是放在newHandler(handler)中，其返回的Mono<? extends NettyContext>中的NettyContext类型元素是管理io.netty.channel.Channel上下文信息的一个对象，这个对象更多的是一些无状态的操做，并不会对此对象作什么样的改变，也是经过对此对象的一个Mono<? extends NettyContext>包装而后经过block产生订阅，来作到sync()的效果，经过，经过block产生订阅后返回的NettyContext对象，可使中断关闭服务器的操做也能够作到更优雅:

public class BlockingNettyContext {

	private static final Logger LOG = Loggers.getLogger(BlockingNettyContext.class);

	private final NettyContext context;
	private final String description;

	private Duration lifecycleTimeout;
	private Thread shutdownHook;

	public BlockingNettyContext(Mono<? extends NettyContext> contextAsync, String description) {
		this(contextAsync, description, Duration.ofSeconds(45));
	}

	public BlockingNettyContext(Mono<? extends NettyContext> contextAsync, String description, Duration lifecycleTimeout) {
		this.description = description;
		this.lifecycleTimeout = lifecycleTimeout;
		this.context = contextAsync
				.timeout(lifecycleTimeout, Mono.error(new TimeoutException(description + " couldn't be started within " + lifecycleTimeout.toMillis() + "ms")))
				.doOnNext(ctx -> LOG.info("Started {} on {}", description, ctx.address()))
				.block();
	}
    ...
    /** * Shut down the {@link NettyContext} and wait for its termination, up to the * {@link #setLifecycleTimeout(Duration) lifecycle timeout}. */
	public void shutdown() {
		if (context.isDisposed()) {
			return;
		}

		removeShutdownHook(); //only applies if not called from the hook's thread

		context.dispose();
		context.onClose()
		       .doOnError(e -> LOG.error("Stopped {} on {} with an error {}", description, context.address(), e))
		       .doOnTerminate(() -> LOG.info("Stopped {} on {}", description, context.address()))
		       .timeout(lifecycleTimeout, Mono.error(new TimeoutException(description + " couldn't be stopped within " + lifecycleTimeout.toMillis() + "ms")))
		       .block();
	}

	...
}
复制代码

这里，咱们来接触下在Reactor中并无深刻接触的blockXXX()操做，其实整个逻辑仍是比较简单的，这里拿reactor.core.publisher.Mono#block()来说，就是获取并返回这个下发的元素:

//reactor.core.publisher.Mono#block()
@Nullable
public T block() {
    BlockingMonoSubscriber<T> subscriber = new BlockingMonoSubscriber<>();
    onLastAssembly(this).subscribe(Operators.toCoreSubscriber(subscriber));
    return subscriber.blockingGet();
}

//reactor.core.publisher.BlockingMonoSubscriber
final class BlockingMonoSubscriber<T> extends BlockingSingleSubscriber<T> {

	@Override
	public void onNext(T t) {
		if (value == null) {
			value = t;
			countDown();
		}
	}

	@Override
	public void onError(Throwable t) {
		if (value == null) {
			error = t;
		}
		countDown();
	}
}
//reactor.core.publisher.BlockingSingleSubscriber
abstract class BlockingSingleSubscriber<T> extends CountDownLatch implements InnerConsumer<T>, Disposable {

T         value;
Throwable error;

Subscription s;

volatile boolean cancelled;

BlockingSingleSubscriber() {
super(1);
}
...
@Nullable
final T blockingGet() {
if (Schedulers.isInNonBlockingThread()) {
    throw new IllegalStateException("block()/blockFirst()/blockLast() are blocking, which is not supported in thread " + Thread.currentThread().getName());
}
if (getCount() != 0) {
    try {
        await();
    }
    catch (InterruptedException ex) {
        dispose();
        throw Exceptions.propagate(ex);
    }
}

Throwable e = error;
if (e != null) {
    RuntimeException re = Exceptions.propagate(e);
    //this is ok, as re is always a new non-singleton instance
    re.addSuppressed(new Exception("#block terminated with an error"));
    throw re;
}
return value;
}

...
@Override
public final void onComplete() {
    countDown();
}
}
复制代码

能够看到，此处使用的CountDownLatch的一个特性，在元素下发赋值以后，等待数值减1，这里恰好也就这一个限定(由super(1)定义)，解除所调用的blockingGet中的等待，获得所需的值，这里，为了保证block()的语义，其onComplete方法也调用了countDown();，即当上游为Mono<Void>时，作到匹配。