标签:first size 算法 caller support star exclusive 共享模式 rup
共享模式acquire实现流程
上文我们讲解了AbstractQueuedSynchronizer独占模式的acquire实现流程,本文趁热打铁继续看一下AbstractQueuedSynchronizer共享模式acquire的实现流程。连续两篇文章的学习,也可以对比独占模式acquire和共享模式acquire的区别,加深对于AbstractQueuedSynchronizer的理解。
先看一下共享模式acquire的实现,方法为acquireShared和acquireSharedInterruptibly,两者差别不大,区别就在于后者有中断处理,以acquireShared为例:
1 public final void acquireShared(int arg) { 2 if (tryAcquireShared(arg) < 0) 3 doAcquireShared(arg); 4 }
这里就能看出第一个差别来了:独占模式acquire的时候子类重写的方法tryAcquire返回的是boolean,即是否tryAcquire成功;共享模式acquire的时候,返回的是一个int型变量,判断是否<0。doAcquireShared方法的实现为:
1 private void doAcquireShared(int arg) { 2 final Node node = addWaiter(Node.SHARED); 3 boolean failed = true; 4 try { 5 boolean interrupted = false; 6 for (;;) { 7 final Node p = node.predecessor(); 8 if (p == head) { 9 int r = tryAcquireShared(arg); 10 if (r >= 0) { 11 setHeadAndPropagate(node, r); 12 p.next = null; // help GC 13 if (interrupted) 14 selfInterrupt(); 15 failed = false; 16 return; 17 } 18 } 19 if (shouldParkAfterFailedAcquire(p, node) && 20 parkAndCheckInterrupt()) 21 interrupted = true; 22 } 23 } finally { 24 if (failed) 25 cancelAcquire(node); 26 } 27 }
我们来分析一下这段代码做了什么:
确实,共享模式下的acquire和独占模式下的acquire大部分逻辑差不多,最大的差别在于tryAcquireShared成功之后,独占模式的acquire是直接将当前节点设置为head节点即可,共享模式会执行setHeadAndPropagate方法,顾名思义,即在设置head之后多执行了一步propagate操作。setHeadAndPropagate方法源码为:
1 private void setHeadAndPropagate(Node node, int propagate) { 2 Node h = head; // Record old head for check below 3 setHead(node); 4 /* 5 * Try to signal next queued node if: 6 * Propagation was indicated by caller, 7 * or was recorded (as h.waitStatus) by a previous operation 8 * (note: this uses sign-check of waitStatus because 9 * PROPAGATE status may transition to SIGNAL.) 10 * and 11 * The next node is waiting in shared mode, 12 * or we don‘t know, because it appears null 13 * 14 * The conservatism in both of these checks may cause 15 * unnecessary wake-ups, but only when there are multiple 16 * racing acquires/releases, so most need signals now or soon 17 * anyway. 18 */ 19 if (propagate > 0 || h == null || h.waitStatus < 0) { 20 Node s = node.next; 21 if (s == null || s.isShared()) 22 doReleaseShared(); 23 } 24 }
第3行的代码设置重设head,第2行的代码由于第3行的代码要重设head,因此先定义一个Node型变量h获得原head的地址,这两行代码很简单。
第19行~第23行的代码是独占锁和共享锁最不一样的一个地方,我们再看独占锁acquireQueued的代码:
1 final boolean acquireQueued(final Node node, int arg) { 2 boolean failed = true; 3 try { 4 boolean interrupted = false; 5 for (;;) { 6 final Node p = node.predecessor(); 7 if (p == head && tryAcquire(arg)) { 8 setHead(node); 9 p.next = null; // help GC 10 failed = false; 11 return interrupted; 12 } 13 if (shouldParkAfterFailedAcquire(p, node) && 14 parkAndCheckInterrupt()) 15 interrupted = true; 16 } 17 } finally { 18 if (failed) 19 cancelAcquire(node); 20 } 21 }
这意味着独占锁某个节点被唤醒之后,它只需要将这个节点设置成head就完事了,而共享锁不一样,某个节点被设置为head之后,如果它的后继节点是SHARED状态的,那么将继续通过doReleaseShared方法尝试往后唤醒节点,实现了共享状态的向后传播。
共享模式release实现流程
上面讲了共享模式下acquire是如何实现的,下面再看一下release的实现流程,方法为releaseShared:
1 public final boolean releaseShared(int arg) { 2 if (tryReleaseShared(arg)) { 3 doReleaseShared(); 4 return true; 5 } 6 return false; 7 }
tryReleaseShared方法是子类实现的,如果tryReleaseShared成功,那么执行doReleaseShared()方法:
1 private void doReleaseShared() { 2 /* 3 * Ensure that a release propagates, even if there are other 4 * in-progress acquires/releases. This proceeds in the usual 5 * way of trying to unparkSuccessor of head if it needs 6 * signal. But if it does not, status is set to PROPAGATE to 7 * ensure that upon release, propagation continues. 8 * Additionally, we must loop in case a new node is added 9 * while we are doing this. Also, unlike other uses of 10 * unparkSuccessor, we need to know if CAS to reset status 11 * fails, if so rechecking. 12 */ 13 for (;;) { 14 Node h = head; 15 if (h != null && h != tail) { 16 int ws = h.waitStatus; 17 if (ws == Node.SIGNAL) { 18 if (!compareAndSetWaitStatus(h, Node.SIGNAL, 0)) 19 continue; // loop to recheck cases 20 unparkSuccessor(h); 21 } 22 else if (ws == 0 && 23 !compareAndSetWaitStatus(h, 0, Node.PROPAGATE)) 24 continue; // loop on failed CAS 25 } 26 if (h == head) // loop if head changed 27 break; 28 } 29 }
主要是两层逻辑:
Condition的await()方法实现原理----构建等待队列
我们知道,Condition是用于实现通知/等待机制的,和Object的wait()/notify()一样,由于本文之前描述AbstractQueuedSynchronizer的共享模式的篇幅不是很长,加之Condition也是AbstractQueuedSynchronizer的一部分,因此将Condition也放在这里写了。
Condition分为await()和signal()两部分,前者用于等待、后者用于唤醒,首先看一下await()是如何实现的。Condition本身是一个接口,其在AbstractQueuedSynchronizer中的实现为ConditionObject:
1 public class ConditionObject implements Condition, java.io.Serializable { 2 private static final long serialVersionUID = 1173984872572414699L; 3 /** First node of condition queue. */ 4 private transient Node firstWaiter; 5 /** Last node of condition queue. */ 6 private transient Node lastWaiter; 7 8 ... 9 }
这里贴了一些字段定义,后面都是方法就不贴了,会对重点方法进行分析的。从字段定义我们可以看到,ConditionObject全局性地记录了第一个等待的节点与最后一个等待的节点。
像ReentrantLock每次要使用ConditionObject,直接new一个ConditionObject出来即可。我们关注一下await()方法的实现:
1 public final void await() throws InterruptedException { 2 if (Thread.interrupted()) 3 throw new InterruptedException(); 4 Node node = addConditionWaiter(); 5 int savedState = fullyRelease(node); 6 int interruptMode = 0; 7 while (!isOnSyncQueue(node)) { 8 LockSupport.park(this); 9 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 10 break; 11 } 12 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 13 interruptMode = REINTERRUPT; 14 if (node.nextWaiter != null) // clean up if cancelled 15 unlinkCancelledWaiters(); 16 if (interruptMode != 0) 17 reportInterruptAfterWait(interruptMode); 18 }
第2行~第3行的代码用于处理中断,第4行代码比较关键,添加Condition的等待者,看一下实现:
1 private Node addConditionWaiter() { 2 Node t = lastWaiter; 3 // If lastWaiter is cancelled, clean out. 4 if (t != null && t.waitStatus != Node.CONDITION) { 5 unlinkCancelledWaiters(); 6 t = lastWaiter; 7 } 8 Node node = new Node(Thread.currentThread(), Node.CONDITION); 9 if (t == null) 10 firstWaiter = node; 11 else 12 t.nextWaiter = node; 13 lastWaiter = node; 14 return node; 15 }
首先拿到队列(注意数据结构,Condition构建出来的也是一个队列)中最后一个等待者,紧接着第4行的的判断,判断最后一个等待者的waitStatus不是CONDITION的话,执行第5行的代码,解绑取消的等待者,因为通过第8行的代码,我们看到,new出来的Node的状态都是CONDITION的。
那么unlinkCancelledWaiters做了什么?里面的流程就不看了,就是一些指针遍历并判断状态的操作,总结一下就是:从头到尾遍历每一个Node,遇到Node的waitStatus不是CONDITION的就从队列中踢掉,该节点的前后节点相连。
接着第8行的代码前面说过了,new出来了一个Node,存储了当前线程,waitStatus是CONDITION,接着第9行~第13行的操作很好理解:
用一张图表示一下构建的数据结构就是:
对比学习,我们总结一下Condition构建出来的队列和AbstractQueuedSynchronizer构建出来的队列的差别,主要体现在2点上:
整个过程中,我们看到没有使用任何CAS操作,firstWaiter和lastWaiter也没有用volatile修饰,其实原因很简单:要await()必然要先lock(),既然lock()了就表示没有竞争,没有竞争自然也没必要使用volatile+CAS的机制去保证什么。
Condition的await()方法实现原理----线程等待
前面我们看了Condition构建等待队列的过程,接下来我们看一下等待的过程,await()方法的代码比较短,再贴一下:
1 public final void await() throws InterruptedException { 2 if (Thread.interrupted()) 3 throw new InterruptedException(); 4 Node node = addConditionWaiter(); 5 int savedState = fullyRelease(node); 6 int interruptMode = 0; 7 while (!isOnSyncQueue(node)) { 8 LockSupport.park(this); 9 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 10 break; 11 } 12 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 13 interruptMode = REINTERRUPT; 14 if (node.nextWaiter != null) // clean up if cancelled 15 unlinkCancelledWaiters(); 16 if (interruptMode != 0) 17 reportInterruptAfterWait(interruptMode); 18 }
构建完毕队列之后,执行第5行的fullyRelease方法,顾名思义:fullyRelease方法的作用是完全释放Node的状态。方法实现为:
1 final int fullyRelease(Node node) { 2 boolean failed = true; 3 try { 4 int savedState = getState(); 5 if (release(savedState)) { 6 failed = false; 7 return savedState; 8 } else { 9 throw new IllegalMonitorStateException(); 10 } 11 } finally { 12 if (failed) 13 node.waitStatus = Node.CANCELLED; 14 } 15 }
这里第4行获取state,第5行release的时候将整个state传过去,理由是某线程可能多次调用了lock()方法,比如调用了10次lock,那么此线程就将state加到了10,所以这里要将10传过去,将状态全部释放,这样后面的线程才能重新从state=0开始竞争锁,这也是方法被命名为fullyRelease的原因,因为要完全释放锁,释放锁之后,如果有竞争锁的线程,那么就唤醒第一个,这都是release方法的逻辑了,前面的文章详细讲解过。
接着看await()方法的第7行判断"while(!isOnSyncQueue(node))":
1 final boolean isOnSyncQueue(Node node) { 2 if (node.waitStatus == Node.CONDITION || node.prev == null) 3 return false; 4 if (node.next != null) // If has successor, it must be on queue 5 return true; 6 /* 7 * node.prev can be non-null, but not yet on queue because 8 * the CAS to place it on queue can fail. So we have to 9 * traverse from tail to make sure it actually made it. It 10 * will always be near the tail in calls to this method, and 11 * unless the CAS failed (which is unlikely), it will be 12 * there, so we hardly ever traverse much. 13 */ 14 return findNodeFromTail(node); 15 }
注意这里的判断是Node是否在AbstractQueuedSynchronizer构建的队列中而不是Node是否在Condition构建的队列中,如果Node不在AbstractQueuedSynchronizer构建的队列中,那么调用LockSupport的park方法阻塞。
至此调用await()方法的线程构建Condition等待队列--释放锁--等待的过程已经全部分析完毕。
Condition的signal()实现原理
上面的代码分析了构建Condition等待队列--释放锁--等待的过程,接着看一下signal()方法通知是如何实现的:
1 public final void signal() { 2 if (!isHeldExclusively()) 3 throw new IllegalMonitorStateException(); 4 Node first = firstWaiter; 5 if (first != null) 6 doSignal(first); 7 }
首先从第2行的代码我们看到,要能signal(),当前线程必须持有独占锁,否则抛出异常IllegalMonitorStateException。
那么真正操作的时候,获取第一个waiter,如果有waiter,调用doSignal方法:
1 private void doSignal(Node first) { 2 do { 3 if ( (firstWaiter = first.nextWaiter) == null) 4 lastWaiter = null; 5 first.nextWaiter = null; 6 } while (!transferForSignal(first) && 7 (first = firstWaiter) != null); 8 }
第3行~第5行的代码很好理解:
接着执行第6行和第7行的代码,这里重点就是第6行的transferForSignal方法:
1 final boolean transferForSignal(Node node) { 2 /* 3 * If cannot change waitStatus, the node has been cancelled. 4 */ 5 if (!compareAndSetWaitStatus(node, Node.CONDITION, 0)) 6 return false; 7 8 /* 9 * Splice onto queue and try to set waitStatus of predecessor to 10 * indicate that thread is (probably) waiting. If cancelled or 11 * attempt to set waitStatus fails, wake up to resync (in which 12 * case the waitStatus can be transiently and harmlessly wrong). 13 */ 14 Node p = enq(node); 15 int ws = p.waitStatus; 16 if (ws > 0 || !compareAndSetWaitStatus(p, ws, Node.SIGNAL)) 17 LockSupport.unpark(node.thread); 18 return true; 19 }
方法本意是将一个节点从Condition队列转换为AbstractQueuedSynchronizer队列,总结一下方法的实现:
最后上面的步骤全部成功,返回true,返回true唤醒等待节点成功。从唤醒的代码我们可以得出一个重要结论:某个await()的节点被唤醒之后并不意味着它后面的代码会立即执行,它会被加入到AbstractQueuedSynchronizer队列的尾部,只有前面等待的节点获取锁全部完毕才能轮到它。
代码分析到这里,我想类似的signalAll方法也没有必要再分析了,显然signalAll方法的作用就是将所有Condition队列中等待的节点逐一队列中从移除,由CONDITION状态变为SIGNAL状态并加入AbstractQueuedSynchronizer队列的尾部。
代码示例
可能大家看了我分析半天代码会有点迷糊,这里最后我贴一段我用于验证上面Condition结论的示例代码,首先建立一个Thread,我将之命名为ConditionThread:
1 /** 2 * @author 五月的仓颉http://www.cnblogs.com/xrq730/p/7067904.html 3 */ 4 public class ConditionThread implements Runnable { 5 6 private Lock lock; 7 8 private Condition condition; 9 10 public ConditionThread(Lock lock, Condition condition) { 11 this.lock = lock; 12 this.condition = condition; 13 } 14 15 @Override 16 public void run() { 17 18 if ("线程0".equals(JdkUtil.getThreadName())) { 19 thread0Process(); 20 } else if ("线程1".equals(JdkUtil.getThreadName())) { 21 thread1Process(); 22 } else if ("线程2".equals(JdkUtil.getThreadName())) { 23 thread2Process(); 24 } 25 26 } 27 28 private void thread0Process() { 29 try { 30 lock.lock(); 31 System.out.println("线程0休息5秒"); 32 JdkUtil.sleep(5000); 33 condition.signal(); 34 System.out.println("线程0唤醒等待线程"); 35 } finally { 36 lock.unlock(); 37 } 38 } 39 40 private void thread1Process() { 41 try { 42 lock.lock(); 43 System.out.println("线程1阻塞"); 44 condition.await(); 45 System.out.println("线程1被唤醒"); 46 } catch (InterruptedException e) { 47 48 } finally { 49 lock.unlock(); 50 } 51 } 52 53 private void thread2Process() { 54 try { 55 System.out.println("线程2想要获取锁"); 56 lock.lock(); 57 System.out.println("线程2获取锁成功"); 58 } finally { 59 lock.unlock(); 60 } 61 } 62 63 }
这个类里面的方法就不解释了,反正就三个方法片段,根据线程名判断,每个线层执行的是其中的一个代码片段。写一段测试代码:
1 /** 2 * @author 五月的仓颉http://www.cnblogs.com/xrq730/p/7067904.html 3 */ 4 @Test 5 public void testCondition() throws Exception { 6 Lock lock = new ReentrantLock(); 7 Condition condition = lock.newCondition(); 8 9 // 线程0的作用是signal 10 Runnable runnable0 = new ConditionThread(lock, condition); 11 Thread thread0 = new Thread(runnable0); 12 thread0.setName("线程0"); 13 // 线程1的作用是await 14 Runnable runnable1 = new ConditionThread(lock, condition); 15 Thread thread1 = new Thread(runnable1); 16 thread1.setName("线程1"); 17 // 线程2的作用是lock 18 Runnable runnable2 = new ConditionThread(lock, condition); 19 Thread thread2 = new Thread(runnable2); 20 thread2.setName("线程2"); 21 22 thread1.start(); 23 Thread.sleep(1000); 24 thread0.start(); 25 Thread.sleep(1000); 26 thread2.start(); 27 28 thread1.join(); 29 }
测试代码的意思是:
代码执行结果为:
1 线程1阻塞 2 线程0休息5秒 3 线程2想要获取锁 4 线程0唤醒等待线程 5 线程2获取锁成功 6 线程1被唤醒
符合我们的结论:signal()并不意味着被唤醒的线程立即执行。由于线程2先于线程0排队,因此看到第5行打印的内容,线程2先获取锁。
再谈AbstractQueuedSynchronizer:共享模式与基于Condition的等待/通知机制实现
标签:first size 算法 caller support star exclusive 共享模式 rup
原文地址:http://www.cnblogs.com/xrq730/p/7067904.html