分析同步工具Semaphore和CyclicBarrier的实现原理

Java并发编程源码分析系列:

  • 分析Java线程池的创建
  • 分析Java线程池执行原理
  • 分析Java线程池Callable任务执行原理
  • 分析ReentrantLock的实现原理
  • 分析CountDownLatch的实现原理

前两篇分别通过ReentrantLock和CountDownLatch分析了AQS的独占功能和共享功能。除CountDownLatch之外,还有Semaphore和CyclicBarrier,前者类似CountDownLatch使用AQS实现,后者使用ReentrantLock实现,本文分析一下。

信号量Semaphore

直接用个小例子描述信号量的用法:

final Semaphore semaphore = new Semaphore(2);
for (int studentIndex = 0; studentIndex < 5; studentIndex++) {
    final int finalI = studentIndex;
    new Thread(() -> {
        try {
            semaphore.acquire(); //如果没有许可,阻塞
            System.out.println("student " + finalI + " read book");
            Thread.sleep(new Random().nextInt(10000) + 1000);
            System.out.println("student " + finalI + " finish read book");
            semaphore.release(); //释放许可
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }).start();
}

5名学生阅读2本书,因为书的数量是固定的,所以只有2名学生可以持有书阅读,其他人需要等别人看完再看。程序执行结果如下:

student 0 read book
student 1 read book
student 1 finish read book
student 2 read book
student 0 finish read book
student 3 read book
student 2 finish read book
student 4 read book
student 3 finish read book
student 4 finish read book

信号量控制同时访问某个特定资源的操作数量,或者执行某个指定操作的数量。

信号量有许可的概念,初始许可数量在构造Semaphore时确定。许可还有剩余时,通过acquire获取许可;操作完成后,通过release释放许可。如果没有许可,acquire将阻塞直到有许可。

Semaphore使用起来非常简单,下面来看源码。

Semaphore构造函数
public Semaphore(int permits) {
     sync = new NonfairSync(permits);
}
public Semaphore(int permits, boolean fair) {
    sync = fair ? new FairSync(permits) : new NonfairSync(permits);
}

和ReentrantLock类似,Semaphore也分为公平和非公平两种。permits是许可数量,最终传递给state,此时state的含义就是许可的数量。

获取许可
public void acquire() throws InterruptedException {
    sync.acquireSharedInterruptibly(1);
}

Semaphore的acquire调用的是acquireSharedInterruptibly,和CountDownLatch的await一样,差异的地方是tryAcquireShared方法是由子类实现。

protected int tryAcquireShared(int acquires) {
    for (;;) {
        if (hasQueuedPredecessors())
            return -1;
        int available = getState();
        int remaining = available - acquires;
        if (remaining < 0 ||
            compareAndSetState(available, remaining))
            return remaining;
    }
}

final int nonfairTryAcquireShared(int acquires) {
    for (;;) {
        int available = getState();
        int remaining = available - acquires;
        if (remaining < 0 ||
            compareAndSetState(available, remaining))
            return remaining;
    }
}

公平信号量和非公平信号量分别调用tryAcquireShared和nonfairTryAcquireShared,目的都是将state减一,以此尝试获取许可。不同之处是公平信号量多调用了hasQueuedPredecessors,判断是否有比自己先的线程在等待;非公平信号量就不管排队,先试了再说。

释放许可

对于release操作,还是和CountDownloadLatch一样调用releaseShared。到这里,应该对AQS的使用非常了解,直接来看tryReleaseShared。

protected final boolean tryReleaseShared(int releases) {
    for (;;) {
        int current = getState();
        int next = current + releases;
        if (next < current) // overflow
            throw new Error("Maximum permit count exceeded");
        if (compareAndSetState(current, next))
            return true;
    }
}

release是acquire的反操作,将许可加一。

总结下来,Semaphore和CountDownLatch都是依赖AQS实现,两者只有少少不同。

栅栏CyclicBarrier

final CyclicBarrier cyclicBarrier = new CyclicBarrier(5, () -> System.out.println("all threads through barrier"));

for (int i = 0; i < 5; i++) {
    final int finalI = i + 1;
    new Thread(() -> {
        System.out.println("thread " + finalI + " is started");
        Random random = new Random();
        try {
            Thread.sleep(random.nextInt(10000) + 1000);
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        System.out.println("thread " + finalI + " has been completed");

        try {
            cyclicBarrier.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        } catch (BrokenBarrierException e) {
            e.printStackTrace();
        }

        System.out.println("thread " + finalI + " to continue");
    }).start();
}

CyclicBarrier设立一个barrier,只有当所有线程到达barrier的位置,才能继续执行下去。打个简单比喻,只有当所有人到达指定地方集中,才能开车出发。

thread 1 is started
thread 4 is started
thread 3 is started
thread 5 is started
thread 2 is started
thread 2 has been completed
thread 3 has been completed
thread 5 has been completed
thread 4 has been completed
thread 1 has been completed
all threads through barrier
CyclicBarrier的构造函数
public CyclicBarrier(int parties, Runnable barrierAction) {
    if (parties <= 0) throw new IllegalArgumentException();
    this.parties = parties;
    this.count = parties;
    this.barrierCommand = barrierAction;
}
  • parties表示barrier开启需要到达的线程数量;
  • count表示等待到达barrier的线程数量;
  • barrierCommand表示开启barrier后执行的操作;

CyclicBarrier的构造函数设置了参数,没有什么特别,下面来看核心方法await。await分为限时和不限时两个版本,最终都调用了dowait。在开始研究dowait代码之前,先要了解几点。

线程什么时候通过Barrier?

当三种条件之一发生时,线程才能继续执行:

  • 当parties个线程到达barrier;
  • 当前线程被中断,抛出InterruptedException;
  • 超时,抛出BrokenBarrierException。
显式条件Condition

dowait的实现使用了ReentrantLock和Condition。ReentrantLock前文介绍过,它是synchronized的高级版本。Object的wait、notify、notifyAll大家都知道,它和synchronized打包实现“条件不满足时,线程等待;条件满足时,等待该条件的线程被唤醒”的功能。Condition是它的高级版本,和ReentrantLock一起可以更加灵活精细地实现这种功能。Condition不是本文的重点,后续会开新篇讲解,现在先当成wait和notify理解。

Generation对象
private Generation generation = new Generation();

private static class Generation {
    boolean broken = false;
}

private void nextGeneration() {
   // signal completion of last generation
   trip.signalAll();
   // set up next generation
   count = parties;
   generation = new Generation();
}

Generation是一代的意思,唯一记录了barrier是否broken。看CyclicBarrier的名字也知道,它是可重复使用的,每次使用CyclicBarrier,本次所有线程同属于一代,即同一个Generation。当parties个线程到达barrier时,需要调用nextGeneration更新换代。

private void breakBarrier() {
    generation.broken = true;
    count = parties;
    trip.signalAll();
}

barrier被broken后,调用breakBarrier方法,将generation.broken设置为true,并使用signalAll通知所有等待的线程。

dowait

现在正式看dowait的代码,代码有点长,但有了前面的铺垫,看起来就简单多了。

private int dowait(boolean timed, long nanos)
    throws InterruptedException, BrokenBarrierException,
           TimeoutException {
    final ReentrantLock lock = this.lock;
    lock.lock();
    try {
        final Generation g = generation;
        //1
        if (g.broken)
            throw new BrokenBarrierException();
        if (Thread.interrupted()) {
            breakBarrier();
            throw new InterruptedException();
        }
        //2
        int index = --count;
        if (index == 0) {  // tripped
            boolean ranAction = false;
            try {
                final Runnable command = barrierCommand;
                if (command != null)
                    command.run();
                ranAction = true;
                nextGeneration();
                return 0;
            } finally {
                if (!ranAction)
                    breakBarrier();
            }
        }
        //3
        for (;;) {
            try {
                if (!timed)
                    trip.await();
                else if (nanos > 0L)
                    nanos = trip.awaitNanos(nanos);
            } catch (InterruptedException ie) {
                if (g == generation && ! g.broken) {
                    breakBarrier();
                    throw ie;
                } else {
                    Thread.currentThread().interrupt();
                }
            }

            if (g.broken)
                throw new BrokenBarrierException();

            if (g != generation)
                return index;

            if (timed && nanos <= 0L) {
                breakBarrier();
                throw new TimeoutException();
            }
        }
    } finally {
        lock.unlock();
    }
}

dowait一开始就获取ReentrantLock并锁定,然后在标记1,判断Generation是否broken,再判断线程是否被中断。

标记2,将等待的线程数量减1,如果count正好为零,说明parties个线程到达barrier了。执行预定的Runnable任务后,更新换代,准备下一次使用。

标记3的for循环里,通过标记timed,根据是否限时,线程调用await或者awaitNanos进入等待。其他代码很好理解,就不多说了。

后记

理解AQS之后,研究Semaphore易如反掌,CyclicBarrier也没什么难的,只有一个新东西Condition,下一篇研究。

你可能感兴趣的:(分析同步工具Semaphore和CyclicBarrier的实现原理)