Netty源码—2.Reactor线程模型一

大纲

1.关于NioEventLoop的问题整理

2.理解Reactor线程模型主要分三部分

3.NioEventLoop的创建

4.NioEventLoop的启动

1.关于NioEventLoop的问题整理

一.默认下Netty服务端起多少线程及何时启动?

答:默认是2倍CPU核数个线程。在调用EventExcutor的execute(task)方法时,会判断当前线程是否为Netty的Reactor线程,也就是判断当前线程是否为NioEventLoop对应的线程实体。如果是,则说明Netty的Reactor线程已经启动了。如果不是,则说明是外部线程调用EventExcutor的execute()方法。于是会先调用startThread()方法判断当前线程是否已被启动,如果还没有被启动就启动当前线程作为Netty的Reactor线程。

二.Netty是如何解决JDK空轮询的?

答:Netty会判断如果当前阻塞的一个Select()操作并没有花那么长时间,那么就说明此时有可能触发了空轮询Bug。默认情况下如果这个现象达到512次,那么就重建一个Selector,并且把之前Selector上所有的key重新移交到新Selector上。通过以上这种处理方式来避免JDK空轮询Bug。

三.Netty是如何保证异步串行无锁化的?

答:异步串行无锁化有两个场景。

场景一:拿到客户端一个Channel,不需要对该Channel进行同步,直接就可以多线程并发读写。

场景二:ChannelHandler里的所有操作都是线程安全的,不需要进行同步。

Netty在所有外部线程去调用EventLoop或者Channel的方法时,会通过inEventLoop()方法来判断出当前线程是外部线程(非NioEventLoop的线程实体)。在这种情况下,会把所有操作都封装成一个Task放入MPSC队列,然后在NioEventLoop的执行逻辑也就是run()方法里,这些Task会被逐个执行。

2.理解Reactor线程模型主要分三部分

一.NioEventLoop的创建

二.NioEventLoop的启动

三.NioEventLoop的执行

3.NioEventLoop的创建

(1)创建入口

(2)确定NioEventLoop的个数

(3)NioEventLoopGroup的创建流程

(4)创建线程执行器ThreadPerTaskExecutor

(5)创建NioEventLoop

(6)创建线程选择器EventExecutorChooser

(7)NioEventLoopGroup的创建总结

(1)创建入口

ini 复制代码
EventLoopGroup bossGroup = new NioEventLoopGroup();
EventLoopGroup workerGroup = new NioEventLoopGroup();

(2)确定NioEventLoop的个数

由NioEventLoopGroup的构造方法来确定NioEventLoop的个数。如果NioEventLoopGroup没有传递构造参数,那么NioEventLoop线程的个数为CPU核数的2倍。如果NioEventLoopGroup传递了参数n,那么NioEventLoop线程的个数就是n。

(3)NioEventLoopGroup的创建流程

NioEventLoopGroup的构造方法会触发创建流程。

一.创建线程执行器ThreadPerTaskExecutor

每次调用ThreadPerTaskExecutor.execute()方法时都会创建一个线程。

二.创建NioEventLoop

NioEventLoop对应NioEventLoopGroup线程池里的线程,NioEventLoopGroup的构造方法会用一个for循环通过调用newChild()方法来创建NioEventLoop线程。

三.创建线程选择器EventExecutorChooser

线程选择器的作用是用于给每个新连接分配一个NioEventLoop线程,也就是从NioEventLoopGroup线程池中选择一个NioEventLoop线程来处理新连接。

scala 复制代码
//MultithreadEventLoopGroup implementations which is used for NIO Selector based Channels.
public class NioEventLoopGroup extends MultithreadEventLoopGroup {

    //Create a new instance using the default number of threads, 
    //the default ThreadFactory and the SelectorProvider which is returned by SelectorProvider#provider().
    public NioEventLoopGroup() {
        this(0);
    }
    
    //Create a new instance using the specified number of threads, 
    //ThreadFactory and the SelectorProvider which is returned by SelectorProvider#provider().
    public NioEventLoopGroup(int nThreads) {
        this(nThreads, (Executor) null);
    }
    
    public NioEventLoopGroup(int nThreads, Executor executor) {
        this(nThreads, executor, SelectorProvider.provider());
    }
    
    public NioEventLoopGroup(int nThreads, Executor executor, final SelectorProvider selectorProvider) {
        this(nThreads, executor, selectorProvider, DefaultSelectStrategyFactory.INSTANCE);
    }
    
    public NioEventLoopGroup(int nThreads, Executor executor, final SelectorProvider selectorProvider, final SelectStrategyFactory selectStrategyFactory) { 
        super(nThreads, executor, selectorProvider, selectStrategyFactory, RejectedExecutionHandlers.reject());
    }
    ...
}

//Abstract base class for EventLoopGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventLoopGroup extends MultithreadEventExecutorGroup implements EventLoopGroup {
    private static final InternalLogger logger = InternalLoggerFactory.getInstance(MultithreadEventLoopGroup.class);
    private static final int DEFAULT_EVENT_LOOP_THREADS;
    static {
        DEFAULT_EVENT_LOOP_THREADS = Math.max(1, SystemPropertyUtil.getInt("io.netty.eventLoopThreads", Runtime.getRuntime().availableProcessors() * 2));
        if (logger.isDebugEnabled()) logger.debug("-Dio.netty.eventLoopThreads: {}", DEFAULT_EVENT_LOOP_THREADS);
    }

    protected MultithreadEventLoopGroup(int nThreads, Executor executor, Object... args) {
        super(nThreads == 0 ? DEFAULT_EVENT_LOOP_THREADS : nThreads, executor, args);
    }
    ...
}

//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutor[] children;
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...
    //Create a new instance.
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, Object... args) {
        this(nThreads, executor, DefaultEventExecutorChooserFactory.INSTANCE, args);
    }
    
    //Create a new instance.
    //@param nThreads,the number of threads that will be used by this instance.
    //@param executor,the Executor to use, or null if the default should be used.
    //@param chooserFactory,the EventExecutorChooserFactory to use.
    //@param args,arguments which will passed to each #newChild(Executor, Object...) call
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, EventExecutorChooserFactory chooserFactory, Object... args) {
        if (nThreads <= 0) throw new IllegalArgumentException(String.format("nThreads: %d (expected: > 0)", nThreads));
        //1.创建ThreadPerTaskExecutor线程执行器
        if (executor == null) executor = new ThreadPerTaskExecutor(newDefaultThreadFactory());
        //2.创建NioEventLoop
        children = new EventExecutor[nThreads];
        for (int i = 0; i < nThreads; i ++) {
            ...
            //创建每一个NioEventLoop时,都会调用newChild()方法给每一个NioEventLoop配置一些核心参数
            //传入线程执行器executor去创建NioEventLoop
            children[i] = newChild(executor, args);
        }
        //3.创建线程选择器
        chooser = chooserFactory.newChooser(children);
        ...
    }
    ...
}

创建NioEventLoopGroup的脉络如下:

scss 复制代码
new NioEventLoopGroup() //线程组,线程个数默认为2 * CPU核数
  new ThreadPerTaskExecutor() //创建线程执行器,作用是负责创建NioEventLoop对应的线程
  for(...) { newChild() } //构造NioEventLoop,创建NioEventLoop线程组
  chooserFactory.newChooser() //线程选择器,用于给每个新连接分配一个NioEventLoop线程

(4)创建线程执行器ThreadPerTaskExecutor

ThreadPerTaskExecutor的作用是:每次调用它的execute()方法执行Runnable任务时,都会通过threadFactory.newThread()创建出一个线程,然后把要执行的Runnable任务传递进该线程进行执行。

其中成员变量threadFactory是在传参给ThreadPerTaskExecutor的构造方法时,由newDefaultThreadFactory()方法构建的,也就是一个DefaultThreadFactory对象。

所以线程执行器ThreadPerTaskExecutor在通过threadFactory.newThread()创建线程时,其实就是调用DefaultThreadFactory的newThread()方法。

而DefaultThreadFactory.newThread()方法创建出来的线程实体,是Netty经过优化之后的FastThreadLocalThread对象,这个线程实体在操作ThreadLocal时,要比JDK快。

ThreadPerTaskExecutor线程执行器总结:

一.每次执行ThreadPerTaskExecutor的execute()方法时,都会创建出一个FastThreadLocalThread的线程实体,所以Netty的线程实体都是由ThreadPerTaskExecutor创建的。

二.FastThreadLocalThread线程实体的命名规则是:nioEventLoop-自增的线程池编号-自增的线程数编号。

java 复制代码
//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutor[] children;
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...    
    //Create a new instance.
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, EventExecutorChooserFactory chooserFactory, Object... args) {
        if (nThreads <= 0) throw new IllegalArgumentException(String.format("nThreads: %d (expected: > 0)", nThreads));
        //1.创建ThreadPerTaskExecutor线程执行器
        if (executor == null) executor = new ThreadPerTaskExecutor(newDefaultThreadFactory());
        //2.创建NioEventLoop
        children = new EventExecutor[nThreads];
        for (int i = 0; i < nThreads; i ++) {
            ...
            //创建每一个NioEventLoop时,都会调用newChild()方法给每一个NioEventLoop配置一些核心参数
            //传入线程执行器executor去创建NioEventLoop
            children[i] = newChild(executor, args);
        }
        //3.创建线程选择器
        chooser = chooserFactory.newChooser(children);
        ...
    }
    
    protected ThreadFactory newDefaultThreadFactory() {
        //getClass()是获取该方法所属的对象类型,也就是NioEventLoopGroup类型
        //因为是通过NioEventLoopGroup的构造方法层层调用到这里的
        return new DefaultThreadFactory(getClass());
    }
    ...
}

public final class ThreadPerTaskExecutor implements Executor {
    private final ThreadFactory threadFactory;
    public ThreadPerTaskExecutor(ThreadFactory threadFactory) {
        if (threadFactory == null) throw new NullPointerException("threadFactory");
        this.threadFactory = threadFactory;
    }
    
    @Override
    public void execute(Runnable command) {
        //调用DefaultThreadFactory的newThread()方法执行Runnable任务
        threadFactory.newThread(command).start();
    }
}

//A ThreadFactory implementation with a simple naming rule.
public class DefaultThreadFactory implements ThreadFactory {
    private static final AtomicInteger poolId = new AtomicInteger();
    private final AtomicInteger nextId = new AtomicInteger();
    private final boolean daemon;
    private final int priority;
    protected final ThreadGroup threadGroup;
    ...
    public DefaultThreadFactory(Class<?> poolType) {
        this(poolType, false, Thread.NORM_PRIORITY);
    }
    
    public DefaultThreadFactory(Class<?> poolType, boolean daemon, int priority) {
        //toPoolName()方法会把NioEventLoopGroup的首字母变成小写
        this(toPoolName(poolType), daemon, priority);
    }
    
    public DefaultThreadFactory(String poolName, boolean daemon, int priority) {
        this(poolName, daemon, priority, 
        System.getSecurityManager() == null? Thread.currentThread().getThreadGroup() : System.getSecurityManager().getThreadGroup());
    }
    
    public DefaultThreadFactory(String poolName, boolean daemon, int priority, ThreadGroup threadGroup) {
        ...
        //prefix用来标记线程名字的前缀
        prefix = poolName + '-' + poolId.incrementAndGet() + '-';
        this.daemon = daemon;
        this.priority = priority;
        this.threadGroup = threadGroup;
    }
    
    @Override
    public Thread newThread(Runnable r) {
        Thread t = newThread(new DefaultRunnableDecorator(r), prefix + nextId.incrementAndGet());
        if (t.isDaemon()) {
            if (!daemon) t.setDaemon(false);
        } else {
            if (daemon) t.setDaemon(true);
        }
        if (t.getPriority() != priority) t.setPriority(priority);
        return t;
    }
    
    protected Thread newThread(Runnable r, String name) {
        return new FastThreadLocalThread(threadGroup, r, name);
    }
    ...
}

(5)创建NioEventLoop

说明一:

由MultithreadEventExecutorGroup的构造方法可知,Netty会使用for循环 + newChild()方法来创建nThreads个NioEventLoop,而且一个NioEventLoop对应一个线程实体FastThreadLocalThread。

scala 复制代码
//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutor[] children;
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...    
    //Create a new instance.
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, EventExecutorChooserFactory chooserFactory, Object... args) {
        if (nThreads <= 0) throw new IllegalArgumentException(String.format("nThreads: %d (expected: > 0)", nThreads));
        //1.创建ThreadPerTaskExecutor线程执行器
        if (executor == null) executor = new ThreadPerTaskExecutor(newDefaultThreadFactory());
        //2.创建NioEventLoop
        children = new EventExecutor[nThreads];
        for (int i = 0; i < nThreads; i ++) {
            ...
            //创建每一个NioEventLoop时,都会调用newChild()方法给每一个NioEventLoop配置一些核心参数
            //传入线程执行器executor去创建NioEventLoop
            children[i] = newChild(executor, args);
        }
        //3.创建线程选择器
        chooser = chooserFactory.newChooser(children);
        ...
    }
    
    //Create a new EventExecutor which will later then accessible via the #next() method.
    //This method will be called for each thread that will serve this MultithreadEventExecutorGroup.
    protected abstract EventExecutor newChild(Executor executor, Object... args) throws Exception;
    ...
}

说明二:

MultithreadEventExecutorGroup的newChild()抽象方法是由NioEventLoopGroup实现的,所以在执行NioEventLoopGroup的默认构造方法时也会执行其newChild()方法。

NioEventLoopGroup的newChild()方法需要传递一个executor参数,该参数就是执行NioEventLoopGroup构造方法开始时创建的线程执行器,之后newChild()方法会返回一个新创建的NioEventLoop对象。

scala 复制代码
//MultithreadEventLoopGroup implementations which is used for NIO Selector based Channels.
public class NioEventLoopGroup extends MultithreadEventLoopGroup {
    ...
    @Override
    protected EventLoop newChild(Executor executor, Object... args) throws Exception {
        //executor是执行NioEventLoopGroup构造方法开始时创建的线程执行器ThreadPerTaskExecutor
        //this指的是NioEventLoopGroup,表示新创建的NioEventLoop对象归属哪个NioEventLoopGroup
        return new NioEventLoop(this, executor, (SelectorProvider) args[0],
            ((SelectStrategyFactory) args[1]).newSelectStrategy(), (RejectedExecutionHandler) args[2]);
    }
    ...
}

说明三:

创建NioEventLoop对象时,NioEventLoop的构造方法会通过调用其openSelector()方法来创建一个Selector,所以一个Selector就和一个NioEventLoop绑定了,而一个Selector可以将多个连接绑定在一起来负责监听这些连接的读写事件。

在NioEventLoop的openSelector()方法中,Netty会通过反射对Selector底层的数据结构进行优化(Hash Set => 数组)。

scala 复制代码
//SingleThreadEventLoop implementation which register the Channel's to a Selector and so does the multi-plexing of these in the event loop.
public final class NioEventLoop extends SingleThreadEventLoop {
    //The NIO Selector.
    Selector selector;
    private final SelectorProvider provider;
    private final SelectStrategy selectStrategy;
    ...
    NioEventLoop(NioEventLoopGroup parent, Executor executor, SelectorProvider selectorProvider,
        SelectStrategy strategy, RejectedExecutionHandler rejectedExecutionHandler) {
        //调用其父类SingleThreadEventLoop的构造方法
        super(parent, executor, false, DEFAULT_MAX_PENDING_TASKS, rejectedExecutionHandler);
        if (selectorProvider == null) throw new NullPointerException("selectorProvider");
        if (strategy == null) throw new NullPointerException("selectStrategy");
        this.provider = selectorProvider;
        this.selector = openSelector();//创建一个Selector
        this.selectStrategy = strategy;
    }
    
    private Selector openSelector() {
        final Selector selector;
        try {
            selector = provider.openSelector();
            ...
        } catch(IOException e) {
            ...
        }
        ...
        return selector;
    }
    ...
}

说明四:

NioEventLoop的构造方法还会调用其父类的父类SingleThreadEventExecutor的构造方法。SingleThreadEventExecutor的构造方法里有两个关键的操作:一是把线程执行器保存起来,因为后面创建NioEventLoop对应的线程时要用到。二是创建一个MPSC任务队列,因为Netty中所有异步执行的本质都是通过该任务队列来协调完成的。

scala 复制代码
//Abstract base class for EventLoops that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventLoop extends SingleThreadEventExecutor implements EventLoop {
    private final Queue<Runnable> tailTasks;
    ...
    protected SingleThreadEventLoop(EventLoopGroup parent, Executor executor, 
        boolean addTaskWakesUp, int maxPendingTasks, RejectedExecutionHandler rejectedExecutionHandler) {
        //调用其父类SingleThreadEventExecutor的构造方法
        super(parent, executor, addTaskWakesUp, maxPendingTasks, rejectedExecutionHandler);
        //调用父类SingleThreadEventExecutor的newTaskQueue()方法
        tailTasks = newTaskQueue(maxPendingTasks);
    }
    ...
}

//Abstract base class for OrderedEventExecutor's that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventExecutor extends AbstractScheduledEventExecutor implements OrderedEventExecutor {
    private final boolean addTaskWakesUp;
    private final Executor executor;
    private final int maxPendingTasks;
    private final Queue<Runnable> taskQueue;
    private final RejectedExecutionHandler rejectedExecutionHandler;
    ...
    //Create a new instance
    protectedSingleThreadEventExecutor(EventExecutorGroup parent, ThreadFactory threadFactory,
          boolean addTaskWakesUp, int maxPendingTasks, RejectedExecutionHandler rejectedHandler) {
        this(parent, new ThreadPerTaskExecutor(threadFactory), addTaskWakesUp, maxPendingTasks, rejectedHandler);
    }
    
    //Create a new instance
    //@param parent,the EventExecutorGroup which is the parent of this instance and belongs to it
    //@param executor,the Executor which will be used for executing
    //@param addTaskWakesUp,true if and only if invocation of #addTask(Runnable) will wake up the executor thread
    //@param maxPendingTasks,the maximum number of pending tasks before new tasks will be rejected.
    //@param rejectedHandler,the RejectedExecutionHandler to use.
    protected SingleThreadEventExecutor(EventExecutorGroup parent, Executor executor,
          boolean addTaskWakesUp, int maxPendingTasks, RejectedExecutionHandler rejectedHandler) {
        super(parent);
        this.addTaskWakesUp = addTaskWakesUp;
        this.maxPendingTasks = Math.max(16, maxPendingTasks);
        //关键操作一:把线程执行器保存起来
        this.executor = ObjectUtil.checkNotNull(executor, "executor");
        //关键操作二:创建一个MPSC任务队列
        this.taskQueue = newTaskQueue(this.maxPendingTasks);
        this.rejectedExecutionHandler = ObjectUtil.checkNotNull(rejectedHandler, "rejectedHandler");
    }
    
    //Create a new Queue which will holds the tasks to execute.
    //NioEventLoop会重写这个newTaskQueue()方法
    protected Queue<Runnable> newTaskQueue(int maxPendingTasks) {
        return new LinkedBlockingQueue<Runnable>(maxPendingTasks);
    }
    ...
}

//SingleThreadEventLoop implementation which register the Channel's to a Selector and so does the multi-plexing of these in the event loop.
public final class NioEventLoop extends SingleThreadEventLoop {
    ...
    //创建一个MPSC任务队列
    @Override
    protected Queue<Runnable> newTaskQueue(int maxPendingTasks) {
        //This event loop never calls takeTask()
        return PlatformDependent.newMpscQueue(maxPendingTasks);
    }
    ...
}

MPSC队列也就是多生产者单消费者队列。单消费者是指某个NioEventLoop对应的线程(执行其run()方法的那个线程)。多生产者就是这个NioEventLoop对应的线程之外的线程,通常情况下就是我们的业务线程。比如,一些线程在调用writeAndFlush()方法时可以不用考虑线程安全而随意调用,那么这些线程就是多生产者。

MPSC队列是通过JCTools这个工具包来实现的,Netty的高性能很大程度上要归功于这个工具包。MPSC的全称是Muti Producer Single Consumer。Muti Producer对应的是外部线程,Single Consumer对应的是Netty的NioEventLoop线程。外部线程在执行Netty的一些任务时,如果判断不是由NioEventLoop对应的线程执行的,就会直接放入一个任务队列里,然后由一个NioEventLoop对应的线程去执行。

创建NioEventLoop总结:

NioEventLoopGroup的newChild()方法创建NioEventLoop时做了三项事情:一.创建一个Selector用于轮询注册到该NioEventLoop上的连接,二.创建一个MPSC任务队列,三.保存线程执行器到NioEventLoop。

(6)创建线程选择器EventExecutorChooser

说明一:

在传统的BIO编程中,一个新连接被创建后,通常需要给这个连接绑定一个Selector,之后这个连接的整个生命周期都由这个Selector管理。

说明二:

创建NioEventLoop时会创建一个Selector,所以一个Selector会对应一个NioEventLoop,一个NioEventLoop上会有一个Selector。线程选择器的作用就是为一个连接在NioEventLoopGroup中选择一个NioEventLoop,从而将该连接绑定到这个NioEventLoop的Selector上。

说明三:

根据MultithreadEventExecutorGroup的构造方法,会使用DefaultEventExecutorChooserFactory的newChooser()方法来创建线程选择器。创建好线程选择器EventExecutorChooser之后,便可以通过其next()方法获取一个NioEventLoop。

Netty通过判断NioEventLoopGroup中的NioEventLoop个数是否是2的幂来创建不同的线程选择器。但不管是哪一种选择器,最终效果都是从第一个NioEventLoop开始遍历到最后一个NioEventLoop,然后再从第一个NioEventLoop开始,如此循环。

java 复制代码
//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutor[] children;
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...
    //Create a new instance.
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, Object... args) {
        this(nThreads, executor, DefaultEventExecutorChooserFactory.INSTANCE, args);
    }
    
    //Create a new instance.
    //@param nThreads,the number of threads that will be used by this instance.
    //@param executor,the Executor to use, or null if the default should be used.
    //@param chooserFactory,the EventExecutorChooserFactory to use.
    //@param args,arguments which will passed to each #newChild(Executor, Object...) call
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, EventExecutorChooserFactory chooserFactory, Object... args) {
        if (nThreads <= 0) throw new IllegalArgumentException(String.format("nThreads: %d (expected: > 0)", nThreads));
        //1.创建ThreadPerTaskExecutor线程执行器
        if (executor == null) executor = new ThreadPerTaskExecutor(newDefaultThreadFactory());
        //2.创建NioEventLoop
        children = new EventExecutor[nThreads];
        for (int i = 0; i < nThreads; i ++) {
            ...
            //创建每一个NioEventLoop时,都会调用newChild()方法给每一个NioEventLoop配置一些核心参数
            //传入线程执行器executor去创建NioEventLoop
            children[i] = newChild(executor, args);
        }
        //3.创建线程选择器,chooserFactory就是传入的DefaultEventExecutorChooserFactory实例
        chooser = chooserFactory.newChooser(children);
        ...
    }
    ...
}

//Default implementation which uses simple round-robin to choose next EventExecutor.
@UnstableApi
public final class DefaultEventExecutorChooserFactory implements EventExecutorChooserFactory {
    public static final DefaultEventExecutorChooserFactory INSTANCE = new DefaultEventExecutorChooserFactory();
    private DefaultEventExecutorChooserFactory() { }
    
    @SuppressWarnings("unchecked")
    @Override
    public EventExecutorChooser newChooser(EventExecutor[] executors) {
        if (isPowerOfTwo(executors.length)) {
            //如果NioEventLoop个数是2的幂,则进行位与运算
            return new PowerOfTowEventExecutorChooser(executors);
        } else {
            //如果NioEventLoop个数不是2的幂,则进行普通取模运算
            return new GenericEventExecutorChooser(executors);
        }
    }

    private static boolean isPowerOfTwo(int val) {
        return (val & -val) == val;
    }

    private static final class PowerOfTowEventExecutorChooser implements EventExecutorChooser {
        private final AtomicInteger idx = new AtomicInteger();
        private final EventExecutor[] executors;
        PowerOfTowEventExecutorChooser(EventExecutor[] executors) {
            this.executors = executors;
        }
        @Override
        public EventExecutor next() {
            return executors[idx.getAndIncrement() & executors.length - 1];
        }
    }

    private static final class GenericEventExecutorChooser implements EventExecutorChooser {
        private final AtomicInteger idx = new AtomicInteger();
        private final EventExecutor[] executors;
        GenericEventExecutorChooser(EventExecutor[] executors) {
            this.executors = executors;
        }
        @Override
        public EventExecutor next() {
            return executors[Math.abs(idx.getAndIncrement() % executors.length)];
        }
    }
}

说明四:

创建NioEventLoopGroup的最后一个步骤就是创建线程选择器chooser,创建线程选择器的流程如下:

scss 复制代码
chooserFactory.newChooser() //创建线程选择器的入口,chooser的作用就是为新连接绑定一个NioEventLoop
  DefaultEventExecutorChooserFactory.isPowerOfTwo() //判断NioEventLoop个数是否为2的幂
    PowerOfTowEventExecutorChooser //优化
      index++ & (length - 1) //位与运算
    GenericEventExecutorChooser //普通
      abs(index++ % length) //取模运算

(7)NioEventLoopGroup的创建总结

默认情况下,NioEventLoopGroup会创建2倍CPU核数个NioEventLoop。一个NioEventLoop和一个Selector以及一个MPSC任务队列一一对应。

NioEventLoop线程的命名规则是nioEventLoopGroup-xx-yy,其中xx表示全局第xx个NioEventLoopGroup线程池,yy表示这个NioEventLoop在这个NioEventLoopGroup中是属于第yy个。

线程选择器chooser的作用是为一个连接选择一个NioEventLoop,可通过线程选择器的next()方法返回一个NioEventLoop。如果NioEventLoop的个数为2的幂,则next()方法会使用位与运算进行优化。

一个NioEventLoopGroup会有一个线程执行器executor、一个线程选择器chooser、一个数组children存放2倍CPU核数个NioEventLoop。

4.NioEventLoop的启动

(1)启动NioEventLoop的两大入口

(2)判断当前线程是否是NioEventLoop线程

(3)创建一个线程并启动

(4)NioEventLoop的启动总结

(1)启动NioEventLoop的两大入口

入口一:服务端启动,注册服务端Channel到Selector时

入口二:新连接接入,通过chooser绑定一个NioEventLoop时

下面先看入口一:

调用ServerBootstrap的bind()方法其实会调用AbstractBootstrap的doBind()方法,然后会调用AbstractBootstrap的initAndRegister()方法,接着执行config().group().register(channel)注册服务端Channel。最后会逐层深入调用到AbstractChannel.AbstractUnsafe的register()方法,来启动一个NioEventLoop将服务端Channel注册到Selector上。

scala 复制代码
//Bootstrap sub-class which allows easy bootstrap of ServerChannel
public class ServerBootstrap extends AbstractBootstrap<ServerBootstrap, ServerChannel> {
    ...
    ...
}

//AbstractBootstrap is a helper class that makes it easy to bootstrap a Channel. 
//It support method-chaining to provide an easy way to configure the AbstractBootstrap.
//When not used in a ServerBootstrap context, the #bind() methods are useful for connectionless transports such as datagram (UDP).
public abstract class AbstractBootstrap<B extends AbstractBootstrap<B, C>, C extends Channel> implements Cloneable {
    ...
    //Create a new Channel and bind it.
    public ChannelFuture bind(int inetPort) {
        //首先根据端口号创建一个InetSocketAddress对象,然后调用重载方法bind()
        return bind(new InetSocketAddress(inetPort));
    }
    
    //Create a new Channel and bind it.
    public ChannelFuture bind(SocketAddress localAddress) {
        //验证服务启动需要的必要参数
        validate();
        if (localAddress == null) throw new NullPointerException("localAddress");
        return doBind(ObjectUtil.checkNotNull(localAddress, "localAddress"));
    }
    
    private ChannelFuture doBind(final SocketAddress localAddress) {
        final ChannelFuture regFuture = initAndRegister();//1.初始化和注册Channel
        final Channel channel = regFuture.channel();
        ...
        doBind0(regFuture, channel, localAddress, promise);//2.绑定服务端端口
        ...
        return promise;
    }
    
    final ChannelFuture initAndRegister() {
        Channel channel = null;
        ...
        //1.创建服务端Channel
        channel = channelFactory.newChannel();
        //2.初始化服务端Channel
        init(channel);
        ...
        //3.注册服务端Channel,比如通过NioEventLoopGroup的register()方法进行注册
        ChannelFuture regFuture = config().group().register(channel);
        ...
        return regFuture;
    }
    ...
}

//Bootstrap sub-class which allows easy bootstrap of ServerChannel
public class ServerBootstrap extends AbstractBootstrap<ServerBootstrap, ServerChannel> {
    private final ServerBootstrapConfig config = new ServerBootstrapConfig(this);
    ...
    @Override
    public final ServerBootstrapConfig config() {
        return config;
    }
    ...
}

public abstract class AbstractBootstrapConfig<B extends AbstractBootstrap<B, C>, C extends Channel> {
    protected final B bootstrap;
    ...
    protected AbstractBootstrapConfig(B bootstrap) {
        this.bootstrap = ObjectUtil.checkNotNull(bootstrap, "bootstrap");
    }
    
    //Returns the configured EventLoopGroup or null if non is configured yet.
    public final EventLoopGroup group() {
        //比如返回一个NioEventLoopGroup对象
        return bootstrap.group();
    }
    ...
}

//MultithreadEventLoopGroup implementations which is used for NIO Selector based Channels.
public class NioEventLoopGroup extends MultithreadEventLoopGroup {
    ...
    ...
}

//Abstract base class for EventLoopGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventLoopGroup extends MultithreadEventExecutorGroup implements EventLoopGroup {
    ...
    @Override
    public ChannelFuture register(Channel channel) {
        //先通过next()方法获取一个NioEventLoop,然后通过NioEventLoop.register()方法注册服务端Channel
        return next().register(channel);
    }
    
    @Override
    public EventLoop next() {
        return (EventLoop) super.next();
    }
    ...
}

//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...
    @Override
    public EventExecutor next() {
        //通过线程选择器chooser选择一个NioEventLoop
        return chooser.next();
    }    
    ...
}

//SingleThreadEventLoop implementation which register the Channel's to a Selector and so does the multi-plexing of these in the event loop.
public final class NioEventLoop extends SingleThreadEventLoop {
    ...
    ...
}

//Abstract base class for EventLoops that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventLoop extends SingleThreadEventExecutor implements EventLoop {
    ...
    @Override
    public ChannelFuture register(Channel channel) {
        return register(new DefaultChannelPromise(channel, this));
    }

    @Override
    public ChannelFuture register(final ChannelPromise promise) {
        ObjectUtil.checkNotNull(promise, "promise");
        //调用AbstractUnsafe的register()方法
        promise.channel().unsafe().register(this, promise);
        return promise;
    }
    ...
}

//A skeletal Channel implementation.
public abstract class AbstractChannel extends DefaultAttributeMap implements Channel {
    private volatile EventLoop eventLoop;
    ...
    //Unsafe implementation which sub-classes must extend and use.
    protected abstract class AbstractUnsafe implements Unsafe {
        ...
        @Override
        public final void register(EventLoop eventLoop, final ChannelPromise promise) {
            ...
            //绑定事件循环器,即绑定一个NioEventLoop到该Channel上
            AbstractChannel.this.eventLoop = eventLoop;
            //注册Selector,并启动一个NioEventLoop
            if (eventLoop.inEventLoop()) {
                register0(promise);
            } else {
                ...
                //通过启动这个NioEventLoop线程来调用register0()方法将这个服务端Channel注册到Selector上
                //其实执行的是SingleThreadEventExecutor的execute()方法
                eventLoop.execute(new Runnable() {
                    @Override
                    public void run() {
                        register0(promise);
                    }
                });
                ...
            }
        }
    }
    ...
}

(2)判断当前线程是否是NioEventLoop线程

调用NioEventLoop的inEventLoop()方法可以判断当前线程是否是Netty的Reactor线程,也就是NioEventLoop对应的线程实体。NioEventLoop的线程实体被创建之后,会将该线程实体保存到NioEventLoop父类的成员变量thread中。

服务端启动、注册服务端Channel到Selector,执行到AbstractUnsafe.register()方法中的eventLoop.inEventLoop()代码时,会将main方法对应的主线程传递进来与this.thread进行比较。由于this.thread此时并未赋值,所以为空,因此inEventLoop()方法返回false,于是便会执行eventLoop.execute()代码创建一个线程并启动。

scala 复制代码
//SingleThreadEventLoop implementation which register the Channel's 
//to a Selector and so does the multi-plexing of these in the event loop.
public final class NioEventLoop extends SingleThreadEventLoop {
    ...
    ...
}

//Abstract base class for EventLoops that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventLoop extends SingleThreadEventExecutor implements EventLoop {
    ...
    ...
}

//Abstract base class for OrderedEventExecutor's that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventExecutor extends AbstractScheduledEventExecutor implements OrderedEventExecutor {
    ...
    ...
}

//Abstract base class for EventExecutors that want to support scheduling.
public abstract class AbstractScheduledEventExecutor extends AbstractEventExecutor {
    ...
    ...
}

//Abstract base class for {@link EventExecutor} implementations.
public abstract class AbstractEventExecutor extends AbstractExecutorService implements EventExecutor {
    ...
    @Override
    public boolean inEventLoop() {
        //注册服务端Channel时是通过主线程进行注册的,Thread.currentThread()对应的就是main线程
        //调用SingleThreadEventExecutor.inEventLoop()方法
        return inEventLoop(Thread.currentThread());
    }
    ...
}

//Abstract base class for OrderedEventExecutor's that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventExecutor extends AbstractScheduledEventExecutor implements OrderedEventExecutor {
    private volatile Thread thread;
    ...
    @Override
    public boolean inEventLoop(Thread thread) {
        return thread == this.thread;//此时线程还没创建,this.thread为null
    }
    ...
}

(3)创建一个线程并启动

AbstractUnsafe.register()方法准备将服务端Channel注册到Selector上时,首先在判断条件中执行eventLoop.inEventLoop()代码发现为false,于是便执行eventLoop.execute()代码创建一个线程并启动它去执行注册任务。而执行eventLoop.execute()代码其实就是调用SingleThreadEventExecutor的execute()方法。

java 复制代码
//Abstract base class for EventLoops that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventLoop extends SingleThreadEventExecutor implements EventLoop {
    ...
    @Override
    public ChannelFuture register(Channel channel) {
        return register(new DefaultChannelPromise(channel, this));
    }

    @Override
    public ChannelFuture register(final ChannelPromise promise) {
        ObjectUtil.checkNotNull(promise, "promise");
        //调用AbstractUnsafe的register()方法,并把NioEventLoop自己当作参数传入
        promise.channel().unsafe().register(this, promise);
        return promise;
    }
    ...
}

//A skeletal Channel implementation.
public abstract class AbstractChannel extends DefaultAttributeMap implements Channel {
    private volatile EventLoop eventLoop;
    ...
    //Unsafe implementation which sub-classes must extend and use.
    protected abstract class AbstractUnsafe implements Unsafe {
        ...
        @Override
        public final void register(EventLoop eventLoop, final ChannelPromise promise) {
            ...
            //绑定事件循环器,即绑定一个NioEventLoop到该Channel上
            AbstractChannel.this.eventLoop = eventLoop;
            //注册Selector,并启动一个NioEventLoop
            if (eventLoop.inEventLoop()) {
                register0(promise);
            } else {
                ...
                //通过启动这个NioEventLoop线程来调用register0()方法将这个服务端Channel注册到Selector上
                //其实执行的是SingleThreadEventExecutor的execute()方法
                eventLoop.execute(new Runnable() {
                    @Override
                    public void run() {
                        register0(promise);
                    }
                });
                ...
            }
        }
    }
    ...
}

//Abstract base class for OrderedEventExecutor's that execute all its submitted tasks in a single thread.
public abstract class SingleThreadEventExecutor extends AbstractScheduledEventExecutor implements OrderedEventExecutor {
    private volatile Thread thread;
    //创建NioEventLoop时会通过构造方法传入NioEventLoopGroup的线程执行器executor
    private final Executor executor;
    ...
    @Override
    public void execute(Runnable task) {
        if (task == null) throw new NullPointerException("task");
        boolean inEventLoop = inEventLoop();
        //判断当前线程是否是Netty的Reactor线程
        if (inEventLoop) {
            addTask(task);
        } else {
            startThread();
            addTask(task);
            if (isShutdown() && removeTask(task)) reject();
        }
        if (!addTaskWakesUp && wakesUpForTask(task)) wakeup(inEventLoop);
    }
    
    private void startThread() {
        //判断Reactor线程有没有被启动;如果没有被启动,则通过CAS调用doStartThread()方法启动线程
        if (STATE_UPDATER.get(this) == ST_NOT_STARTED) {
            if (STATE_UPDATER.compareAndSet(this, ST_NOT_STARTED, ST_STARTED)) {
                doStartThread();
            }
        }
    }

    private void doStartThread() {
        assert thread == null;
        //executor.execute()方法会创建出一个FastThreadLocalThread线程来执行Runnable任务
        //所以在Runnable的run()方法中,Thread.currentThread()指的是这个FastThreadLocalThread线程
        executor.execute(new Runnable() {
            @Override
            public void run() {
                //Thread.currentThread()指的是FastThreadLocalThread线程
                thread = Thread.currentThread();
                ...
                SingleThreadEventExecutor.this.run();//启动线程
                ...
            }
        });
    }
    
    //具体的run()方法由子类比如NioEventLoop来实现
    protected abstract void run();
    ...
}

SingleThreadEventExecutor的execute()方法的说明如下:

一.这个方法也可能会被用户代码使用,如ctx.executor().execute(task)。所以execute()方法里又调用inEventLoop()方法进行了一次外部线程判断,确保执行task任务时不会遇到线程问题。

二.如果当前线程不是Netty的Reactor线程,则调用startThread()方法启动一个Reactor线程。在startThread()方法中首先会判断当前NioEventLoop对应的Reactor线程实体有没有被启动。如果没有被启动,则通过设置CAS成功后调用doStartThread()方法启动线程。

三.执行doStartThread()方法时,会调用NioEventLoop的内部成员变量executor的execute()方法。executor就是线程执行器ThreadPerTaskExecutor,它的作用是每次执行Runnable任务时都会创建一个线程来执行。也就是executor.execute()方法会通过DefaultThreadFactory的newThread()方法,创建出一个FastThreadLocalThread线程来执行Runnable任务。

四.doStartThread()方法的Runnable任务会由一个FastThreadLocalThread线程来执行。在Runnable任务的run()方法里,会保存ThreadPerTaskExecutor创建出来的FastThreadLocalThread对象到SingleThreadEventExecutor的成员变量thread中,然后调用SingleThreadEventExecutor的run()方法。

java 复制代码
//Abstract base class for EventExecutorGroup implementations that handles their tasks with multiple threads at the same time.
public abstract class MultithreadEventExecutorGroup extends AbstractEventExecutorGroup {
    private final EventExecutor[] children;
    private final EventExecutorChooserFactory.EventExecutorChooser chooser;
    ...    
    //Create a new instance.
    protected MultithreadEventExecutorGroup(int nThreads, Executor executor, EventExecutorChooserFactory chooserFactory, Object... args) {
        if (nThreads <= 0) throw new IllegalArgumentException(String.format("nThreads: %d (expected: > 0)", nThreads));
        //1.创建ThreadPerTaskExecutor线程执行器
        if (executor == null) executor = new ThreadPerTaskExecutor(newDefaultThreadFactory());
        //2.创建NioEventLoop
        children = new EventExecutor[nThreads];
        for (int i = 0; i < nThreads; i ++) {
            ...
            //创建每一个NioEventLoop时,都会调用newChild()方法给每一个NioEventLoop配置一些核心参数
            //传入线程执行器executor去创建NioEventLoop
            children[i] = newChild(executor, args);
        }
        //3.创建线程选择器
        chooser = chooserFactory.newChooser(children);
        ...
    }
    
    protected ThreadFactory newDefaultThreadFactory() {
        //getClass()是获取该方法所属的对象类型,也就是NioEventLoopGroup类型
        //因为是通过NioEventLoopGroup的构造方法层层调用到这里的
        return new DefaultThreadFactory(getClass());
    }
    ...
}

public final class ThreadPerTaskExecutor implements Executor {
    private final ThreadFactory threadFactory;
    public ThreadPerTaskExecutor(ThreadFactory threadFactory) {
        if (threadFactory == null) throw new NullPointerException("threadFactory");
        this.threadFactory = threadFactory;
    }
    
    @Override
    public void execute(Runnable command) {
        //调用DefaultThreadFactory的newThread()方法执行Runnable任务
        threadFactory.newThread(command).start();
    }
}

//A ThreadFactory implementation with a simple naming rule.
public class DefaultThreadFactory implements ThreadFactory {
    private static final AtomicInteger poolId = new AtomicInteger();
    private final AtomicInteger nextId = new AtomicInteger();
    private final boolean daemon;
    private final int priority;
    protected final ThreadGroup threadGroup;
    ...
    public DefaultThreadFactory(Class<?> poolType) {
        this(poolType, false, Thread.NORM_PRIORITY);
    }
    
    public DefaultThreadFactory(Class<?> poolType, boolean daemon, int priority) {
        //toPoolName()方法会把NioEventLoopGroup的首字母变成小写
        this(toPoolName(poolType), daemon, priority);
    }
    
    public DefaultThreadFactory(String poolName, boolean daemon, int priority) {
        this(poolName, daemon, priority, 
           System.getSecurityManager() == null ? Thread.currentThread().getThreadGroup() : System.getSecurityManager().getThreadGroup());
    }
    
    public DefaultThreadFactory(String poolName, boolean daemon, int priority, ThreadGroup threadGroup) {
        ...
        //prefix用来标记线程名字的前缀
        prefix = poolName + '-' + poolId.incrementAndGet() + '-';
        this.daemon = daemon;
        this.priority = priority;
        this.threadGroup = threadGroup;
    }
    
    @Override
    public Thread newThread(Runnable r) {
        Thread t = newThread(new DefaultRunnableDecorator(r), prefix + nextId.incrementAndGet());
        if (t.isDaemon()) {
            if (!daemon) t.setDaemon(false);
        } else {
            if (daemon) t.setDaemon(true);
        }
        if (t.getPriority() != priority) t.setPriority(priority);
        return t;
    }
    
    protected Thread newThread(Runnable r, String name) {
        return new FastThreadLocalThread(threadGroup, r, name);
    }
    ...
}

NioEventLoop是如何与一个线程实体绑定的? NioEventLoop会通过线程执行器ThreadPerTaskExecutor创建一个FastThreadLocalThread,然后再将该FastThreadLocalThread线程保存到其成员变量中,从而实现与一个线程实体进行绑定。

(4)NioEventLoop的启动总结

一.在注册服务端Channel的过程中,主线程最终会调用AbstractUnsafe的register()方法。该方法首先会将一个NioEventLoop绑定到这个服务端Channel上,然后把实际注册Selector的逻辑封装成一个Runnable任务,接着调用NioEventLoop的execute()方法来执行这个Runnable任务。

二.NioEventLoop的execute()方法其实就是其父类SingleThreadEventExecutor的execute()方法,它会先判断当前调用execute()方法的线程是不是Netty的Reactor线程,如果不是就调用startThread()方法来创建一个Reactor线程。

三.startThread()方法会通过线程执行器ThreadPerTaskExecutor的execute()方法来创建一个线程。这个线程是一个FastThreadLocalThread线程,这个线程需要执行如下逻辑:把线程保存到NioEventLoop的成员变量thread中,然后调用NioEventLoop的run()方法来启动NioEventLoop。

NioEventLoop的启动流程如下:

scss 复制代码
bind() -> initAndRegister() -> config().group().register() -> eventloop.execute() //入口
  startThread() -> doStartThread() //创建线程
    ThreadPerTaskExecutor.execute() //线程执行器创建FastThreadLocalThread线程
      thread = Thread.currentThread() //保存FastThreadLocalThread线程到NioEventLoop的成员变量中
      NioEventLoop.run() //启动NioEventLoop

NioEventLoop的启动流程说明如下:

首先bind()方法会将具体绑定端口的操作封装成一个Runnable任务,然后调用NioEventLoop的execute()方法,接着Netty会判断调用execute()方法的线程是否是NIO线程,如果发现不是就会调用startThread()方法开始创建线程。

创建线程是通过线程执行器ThreadPerTaskExecutor来创建的。线程执行器的作用是每执行一个任务都会创建一个线程,而且创建出来的线程就是NioEventLoop底层的一个FastThreadLocalThread线程。

创建完FastThreadLocalThread线程后会执行一个Runnable任务,该Runnable任务首先会将这个线程保存到NioEventLoop对象。保存的目的是为了判断后续对NioEventLoop的相关执行线程是否为本身。如果不是就将封装好的一个任务放入TaskQueue中进行串行执行,实现线程安全。该Runnable任务然后会调用NioEventLoop的run()方法,从而启动NioEventLoop。NioEventLoop的run()方法是驱动Netty运转的核心方法。

相关推荐
东阳马生架构11 小时前
Netty源码—2.Reactor线程模型二
netty·reactor线程模型
东阳马生架构16 小时前
Netty基础—7.Netty实现消息推送服务二
websocket·netty·消息推送
东阳马生架构2 天前
Netty基础—6.Netty实现RPC服务一
netty·rpc服务
东阳马生架构2 天前
Netty基础—6.Netty实现RPC服务二
rpc·netty
东阳马生架构3 天前
Netty源码—1.服务端启动流程
netty
东阳马生架构4 天前
Netty基础—8.Netty实现私有协议栈
netty
东阳马生架构6 天前
Netty基础—4.NIO的使用简介一
java·网络·netty