本文共 12180 字，大约阅读时间需要 40 分钟。

Semaphore 一个计数信号量。从概念上讲，信号量维护了一个许可集。如有必要，在许可可用前会阻塞每一个 acquire()，然后再获取该许可。每个 release() 添加一个许可，从而可能释放一个正在阻塞的获取者。Semaphore 通常用于限制可以访问某些资源（物理或逻辑的）的线程数目。

通常，应该将用于控制资源访问的信号量初始化为公平的，以确保所有线程都可访问资源。为其他的种类的同步控制使用信号量时，非公平排序的吞吐量优势通常要比公平考虑更为重要。

举个例子：初始化一个许可证为2，启动10个线程，每个线程模拟执行3秒。

/** * Created by Tangwz on 2019/6/28. */public class TestSemaphore {
       public static void main(String[] args) {
           Thread.currentThread().interrupt();        Semaphore semaphore = new Semaphore(2);        for (int i = 0; i < 10; i++) {
               new Thread(new SemaphoreWorker(semaphore), "Thread" + i).start();        }    }    private static class SemaphoreWorker implements Runnable {
           private Semaphore semaphore;        SemaphoreWorker(Semaphore semaphore) {
               this.semaphore = semaphore;        }        @Override        public void run() {
               try {
                   log("waiting for a permit");                semaphore.acquire();                log("get a permit");                Thread.sleep(3000L);                // log("executed");            } catch (InterruptedException e) {
                   e.printStackTrace();            }            log("release a permit");            semaphore.release();        }        private void log(String message) {
               String name = Thread.currentThread().getName();            System.out.println(name + "  " + message);        }    }}

输出结果：每次只有两个线程可以执行，其他的线程都在等待。两个线程运行三秒，执行完释放许可证后，只有另外的两个线程能获取到许可证然后执行，如此依次进行。

看下面的打印结果，本次执行顺序为：2,6 -> 3,7 -> 1,5 -> 0,9 -> 4,8

Thread2 waiting for a permit

Thread6 waiting for a permit

Thread2 get a permit

Thread6 get a permit

Thread3 waiting for a permit

Thread7 waiting for a permit

Thread1 waiting for a permit

Thread5 waiting for a permit

Thread0 waiting for a permit

Thread9 waiting for a permit

Thread8 waiting for a permit

Thread4 waiting for a permit

Thread2 release a permit

Thread3 get a permit

Thread6 release a permit

Thread7 get a permit

Thread3 release a permit

Thread1 get a permit

Thread7 release a permit

Thread5 get a permit

Thread1 release a permit

Thread0 get a permit

Thread5 release a permit

Thread9 get a permit

Thread0 release a permit

Thread9 release a permit

Thread4 get a permit

Thread8 get a permit

Thread4 release a permit

Thread8 release a permit

初始化一个许可证数量，fair 参数可选，默认非公平

/** * Creates a {@code Semaphore} with the given number of * permits and the given fairness setting. * * @param permits the initial number of permits available. *        This value may be negative, in which case releases *        must occur before any acquires will be granted. * @param fair {@code true} if this semaphore will guarantee *        first-in first-out granting of permits under contention, *        else {@code false} */public Semaphore(int permits, boolean fair) {
       sync = fair ? new FairSync(permits) : new NonfairSync(permits);}

获取一个许可证，许可证集中如果还有，立即返回并减少一个许可证；获取不到则阻塞，等待其他线程释放一个许可证，或者被其他线程中断。

/** * Acquires a permit from this semaphore, blocking until one is * available, or the thread is {@linkplain Thread#interrupt interrupted}. */public void acquire() throws InterruptedException {
   	//注意此方法响应中断    sync.acquireSharedInterruptibly(1);}/** * Acquires in shared mode, aborting if interrupted.  Implemented * by first checking interrupt status, then invoking at least once * {@link #tryAcquireShared}, returning on success.  Otherwise the * thread is queued, possibly repeatedly blocking and unblocking, * invoking {@link #tryAcquireShared} until success or the thread * is interrupted. * @param arg the acquire argument. * This value is conveyed to {@link #tryAcquireShared} but is * otherwise uninterpreted and can represent anything * you like. * @throws InterruptedException if the current thread is interrupted */public final void acquireSharedInterruptibly(int arg)        throws InterruptedException {
       //进入前检查中断    if (Thread.interrupted())        throw new InterruptedException();    //尝试获取一个许可证    //两种获取方式，公平还是非公平    if (tryAcquireShared(arg) < 0)        doAcquireSharedInterruptibly(arg);}//非公平，返回值剩余可用许可证 remaining  >=0 代表获取许可证成功，remaining <0 代表失败final int nonfairTryAcquireShared(int acquires) {
       for (;;) {
           int available = getState();        int remaining = available - acquires;        if (remaining < 0 ||            compareAndSetState(available, remaining))            //两种情况会退出循环            //1.可用许可证 available 为 0 ，返回 remaining = -1，代表失败            //2.可用许可证 available > 0 , 返回 remaining >=0 ，并且CAS成功，代表成功            return remaining;    }}/** * Fair version *///公平protected int tryAcquireShared(int acquires) {
       for (;;) {
       	//比非公平获取多加了一个判断，判断同步队列中是否已经有节点在等待获取许可证        if (hasQueuedPredecessors())            return -1;        //没有节点在排队，下面和非公平获取一样        int available = getState();        int remaining = available - acquires;        if (remaining < 0 ||            compareAndSetState(available, remaining))            return remaining;    }}/** * Acquires in shared interruptible mode. * @param arg the acquire argument *///获取许可证失败，进入这里private void doAcquireSharedInterruptibly(int arg)    throws InterruptedException {
       //创建一个共享节点    final Node node = addWaiter(Node.SHARED);    boolean failed = true;    try {
           for (;;) {
               final Node p = node.predecessor();            //前驱为头节点，才可以尝试获取许可证            if (p == head) {
               	//尝试获取许可证                int r = tryAcquireShared(arg);                if (r >= 0) {
                   	//获取许可证成功，如果可用许可证 >0，或者头结点被设置为传递状态，需要唤醒后续节点                    setHeadAndPropagate(node, r);                    p.next = null; // help GC                    failed = false;                    return;                }            }            //前驱不为头节点，或者获取许可证失败，被阻塞，等待其他线程释放一个许可证            if (shouldParkAfterFailedAcquire(p, node) &&                parkAndCheckInterrupt())                //等待期间被中断，抛异常                throw new InterruptedException();        }    } finally {
           if (failed)        	//被中断会进入到这里，将此节点从同步队列中移除            cancelAcquire(node);    }}/** * Sets head of queue, and checks if successor may be waiting * in shared mode, if so propagating if either propagate > 0 or * PROPAGATE status was set. * * @param node the node * @param propagate the return value from a tryAcquireShared */private void setHeadAndPropagate(Node node, int propagate) {
       Node h = head; // Record old head for check below    setHead(node);    /*     * Try to signal next queued node if:     *   Propagation was indicated by caller,     *     or was recorded (as h.waitStatus either before     *     or after setHead) by a previous operation     *     (note: this uses sign-check of waitStatus because     *      PROPAGATE status may transition to SIGNAL.)     * and     *   The next node is waiting in shared mode,     *     or we don't know, because it appears null     *     * The conservatism in both of these checks may cause     * unnecessary wake-ups, but only when there are multiple     * racing acquires/releases, so most need signals now or soon     * anyway.     */    //propagate > 0 需要唤醒等待节点可以理解，为啥还有后面这么多判断？    //这些判断是否可以去掉呢？见后文分析    if (propagate > 0 || h == null || h.waitStatus < 0 ||        (h = head) == null || h.waitStatus < 0) {
           Node s = node.next;        //后驱节点类型是共享的才进行唤醒        if (s == null || s.isShared())        	//需要唤醒后驱节点            doReleaseShared();    }}

分析 setHeadAndPropagate() 前，先看看怎么释放许可证。

任何一个线程调用 release() 就会释放一个许可证到 Semaphore 中，该线程不一定非要先获取一个许可证。

/** * Releases a permit, returning it to the semaphore. * * 
Releases a permit, increasing the number of available permits by * one.  If any threads are trying to acquire a permit, then one is * selected and given the permit that was just released.  That thread * is (re)enabled for thread scheduling purposes. * * 
There is no requirement that a thread that releases a permit must * have acquired that permit by calling {@link #acquire}. * Correct usage of a semaphore is established by programming convention * in the application. */public void release() {
       sync.releaseShared(1);}/** * Releases in shared mode.  Implemented by unblocking one or more * threads if {@link #tryReleaseShared} returns true. * * @param arg the release argument.  This value is conveyed to *        {@link #tryReleaseShared} but is otherwise uninterpreted *        and can represent anything you like. * @return the value returned from {@link #tryReleaseShared} */public final boolean releaseShared(int arg) {
   	//不管是公平还是非公平，在 Semaphore 中调用的都是同一个方法    if (tryReleaseShared(arg)) {
       	//释放 arg 个许可证成功后，唤醒在同步队列中可能存在等待许可证的节点        doReleaseShared();        return true;    }    return false;}//releases 大于 0 的情况下，一定是放入成功了才会返回protected final boolean tryReleaseShared(int releases) {
       for (;;) {
           int current = getState();        int next = current + releases;        //保证释放的许可证数量大于 0         if (next < current) // overflow            throw new Error("Maximum permit count exceeded");        //放许可证到 Semaphore 中可能会产生竞争        if (compareAndSetState(current, next))            return true;    }}/** * Release action for shared mode -- signals successor and ensures * propagation. (Note: For exclusive mode, release just amounts * to calling unparkSuccessor of head if it needs signal.) *///为共享节点做的唤醒，确保传递信号private void doReleaseShared() {
       /*     * Ensure that a release propagates, even if there are other     * in-progress acquires/releases.  This proceeds in the usual     * way of trying to unparkSuccessor of head if it needs     * signal. But if it does not, status is set to PROPAGATE to     * ensure that upon release, propagation continues.     * Additionally, we must loop in case a new node is added     * while we are doing this. Also, unlike other uses of     * unparkSuccessor, we need to know if CAS to reset status     * fails, if so rechecking.     */    for (;;) {
           Node h = head;        //头节点是一定不会被阻塞的，头尾相等即没有节点被阻塞        //头结点为空的情况，只发生在同步队列为空，一直没有线程获取许可证失败而阻塞        if (h != null && h != tail) {
               int ws = h.waitStatus;            //如果同时有多个线程进行释放，这里会产生竞争            if (ws == Node.SIGNAL) {
               	//步骤1 竞争先会发生在这里，成功的线程会将头设置成0，然后唤醒后驱                if (!compareAndSetWaitStatus(h, Node.SIGNAL, 0))            		//1.1 失败的线程会继续进行循环                    continue;            // loop to recheck cases                unparkSuccessor(h);            }            //1.1失败的线程会读到在步骤1中已经修改为0的状态            else if (ws == 0 &&            		  //1.2 失败的线程会将头节点状态标记为传播                	  //1.3 在步骤1中失败的线程不止一个，这里也可能CAS失败                     !compareAndSetWaitStatus(h, 0, Node.PROPAGATE))                //1.4 CAS失败的线程继续运行                continue;                // loop on failed CAS        }        //若线程发现头结点变动，继续循环        if (h == head)                   // loop if head changed            break;    }}

有没有发现上面释放许可证的时候，设置头结点状态为0失败后，为啥还非要执行 compareAndSetWaitStatus(h, 0, Node.PROPAGATE) ，将头结点状态设置为 -3 共享式传递状态，为啥 setHeadAndPropagate() 中有那么多关于头结点的判断？

private void setHeadAndPropagate(Node node, int propagate) {
       Node h = head; // Record old head for check below    setHead(node);    if (propagate > 0 || h == null || h.waitStatus < 0 ||        (h = head) == null || h.waitStatus < 0) {
           Node s = node.next;        if (s == null || s.isShared())            doReleaseShared();    }}

判断和设置-3都不可少，考虑这样一种情况：初始化一个信号量为2，线程A和线程B先获取许可证再释放一个许可证，线程C再获取许可证然后进同步队列，线程D再获取许可证然后在线程C之后进同步队列。接着按时间顺序发生如下步骤：

步骤1：线程A设置头结点状态为 0 ，然后唤醒C后发现头节点引用没有变化然后结束； 步骤2：线程C恢复执行，可用许可证数量 propagate 为 0； 步骤3：线程B发现头状态不为-1不做修改，线程B也发现头节点引用没有变化然后结束； 步骤4：线程C这时才执行 setHead(node) ，发现 propagate = 0，h.waitStatus = 0，然后退出； 步骤5：线程C可能只获取不执行 release() ，那么线程D不会被唤醒，但是此时还有一个许可证。 设置了 Node.PROPAGATE 就不一样了，这时不管线程C执行 setHead(node) 在线程B修改头结点为-3之前还是之后都没有问题：在之前线程C发现头节点状态等于0后退出，线程B发现头节点引用变化，然后继续循环重新获取头结点C，将头结点C的状态改为-1，然后唤醒节点C之后的节点D；在之后线程C发现原头节点状态小于0，唤醒原头节点后驱。

总结下：

setHeadAndPropagate() 作用：设置自己为头节点，传递头节点的共享状态唤醒后驱。当可用许可证大于0唤醒后驱；当原头节点变动说明有其他线程释放了许可证，也唤醒后驱；当自己成为头节点后，之后不管头节点再怎么变化，只要变化说明有其他释放许可证或者有新节点加入，唤醒后驱。并发竞争程度高的话，可能产生无用唤醒。 doReleaseShared() 作用：唤醒后驱，并确保释放的共享状态能传递下去。头节点为-1，唤醒后驱；头节点为0，设置头节点为-3传递释放状态。设置-3失败的情况可能是设置头节点为-1时失败的线程比较多，再次循环发现头节点已经是-3然后退出；或者有新节点加入，再次循环发现头节点为-1唤醒后驱。

同步队列的分析请看

参考

转载地址：http://rmrai.baihongyu.com/