Java 7 Phaser 有什么新特点
openkk
13年前
<p><strong>1 Overview</strong></p> <p>Java 7的并发包中推出了Phaser,其功能跟CyclicBarrier和CountDownLatch有些重叠,但是提供了更灵活的用法,例如支持动态调整注册任务的数量等。本文在Phaser自带的示例代码基础上进行一下简单的分析。</p> <p></p> <p><strong>2 Glossary</strong></p> <p><strong>2.1 Registration</strong></p> <p>Phaser支持通过register()和bulkRegister(int parties)方法来动态调整注册任务的数量,此外也支持通过其构造函数进行指定初始数量。在适当的时机,Phaser支持减少注册任务的数量,例如arriveAndDeregister()。单个Phaser实例允许的注册任务数的上限是65535。</p> <p><strong><br /> </strong></p> <p><strong>2.2 Arrival</strong></p> <p>正如Phaser类的名字所暗示,每个Phaser实例都会维护一个phase number,初始值为0。每当所有注册的任务都到达Phaser时,phase number累加,并在超过Integer.MAX_VALUE后清零。arrive()和arriveAndDeregister()方法用于记录到 达,arriveAndAwaitAdvance()方法用于记录到达,并且等待其它未到达的任务。</p> <p><strong><br /> </strong></p> <p><strong>2.3 Termination</strong></p> <p>Phaser支持终止。Phaser终止之后,调用register()和bulkRegister(int parties)方法没有任何效果,arriveAndAwaitAdvance()方法也会立即返回。触发终止的时机是在protected boolean onAdvance(int phase, int registeredParties)方法返回时,如果该方法返回true,那么Phaser会被终止。默认实现是在注册任务数为0时返回true(即return registeredParties == 0;)。此外,forceTermination()方法用于强制终止,isTerminated()方法用于判断是否已经终止。</p> <p><strong><br /> </strong></p> <p><strong>2.4 Tiering</strong></p> <p>Phaser支持层次结构,即通过构造函数Phaser(Phaser parent)和Phaser(Phaser parent, int parties)构造一个树形结构。这有助于减轻因在单个的Phaser上注册过多的任务而导致的竞争,从而提升吞吐量,代价是增加单个操作的开销。</p> <p></p> <p><strong>3 Sample Usage</strong></p> <p><strong>3.1 Sample 1</strong></p> <p>在有些场景下,我们希望控制多个线程的启动时机:例如在并发相关的单元测试中,有时需要控制线程的启动时机,以期获得最大程度的并发,通常我们会使用CountDownLatch,以下是使用Phaser的版本。</p> <pre class="brush:java; toolbar: true; auto-links: false;">import java.util.concurrent.Phaser; public class PhaserTest1 { public static void main(String args[]) { // final int count = 5; final Phaser phaser = new Phaser(count); for(int i = 0; i < count; i++) { System.out.println("starting thread, id: " + i); final Thread thread = new Thread(new Task(i, phaser)); thread.start(); } } public static class Task implements Runnable { // private final int id; private final Phaser phaser; public Task(int id, Phaser phaser) { this.id = id; this.phaser = phaser; } @Override public void run() { phaser.arriveAndAwaitAdvance(); System.out.println("in Task.run(), phase: " + phaser.getPhase() + ", id: " + this.id); } } }</pre> <p></p> <p>以上例子中,由于线程是在一个循环中start,因此start的时机有一定的间隔。本例中这些线程实际开始工作的时机是在所有的线程都调用了phaser.arriveAndAwaitAdvance()之后。</p> <p>此外,如果留心arriveAndAwaitAdvance()方法的签名,会发现它并没有抛出InterruptedException,实际上,即使 当前线程被中断,arriveAndAwaitAdvance()方法也不会返回,而是继续等待。如果在等待时希望可中断,或者可超时,那么需要使用以下 方法:</p> <pre class="brush:java; toolbar: true; auto-links: false;">awaitAdvance(arrive()) // 等效于arriveAndAwaitAdvance() awaitAdvanceInterruptibly(int phase) awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)</pre> <p></p> <p><strong>3.2 Sample 2</strong></p> <p>有些时候我们希望只有在某些外部条件满足时,才真正开始任务的执行,例如:</p> <pre class="brush:java; toolbar: true; auto-links: false;">import java.io.BufferedReader; import java.io.InputStreamReader; import java.util.concurrent.Phaser; public class PhaserTest2 { public static void main(String args[]) throws Exception { // final Phaser phaser = new Phaser(1); for(int i = 0; i < 5; i++) { phaser.register(); System.out.println("starting thread, id: " + i); final Thread thread = new Thread(new Task(i, phaser)); thread.start(); } // System.out.println("Press ENTER to continue"); BufferedReader reader = new BufferedReader(new InputStreamReader(System.in)); reader.readLine(); phaser.arriveAndDeregister(); } public static class Task implements Runnable { // private final int id; private final Phaser phaser; public Task(int id, Phaser phaser) { this.id = id; this.phaser = phaser; } @Override public void run() { phaser.arriveAndAwaitAdvance(); System.out.println("in Task.run(), phase: " + phaser.getPhase() + ", id: " + this.id); } } }</pre> <p></p> <p>以上例子中,只有当用户按下回车之后,任务才真正开始执行。需要注意的是,arriveAndDeregister()方法不会被阻塞,并且返回到达时的phase number(arrive方法也是如此)。</p> <p></p> <p><strong>3.3 Sample 3</strong></p> <p>CyclicBarrier支持barrier action, Phaser同样也支持。不同之处是Phaser的barrier action需要改写onAdvance方法来进行定制。</p> <pre class="brush:java; toolbar: true; auto-links: false;">import java.util.concurrent.Phaser; public class PhaserTest3 { public static void main(String args[]) throws Exception { // final int count = 5; final int phaseToTerminate = 3; final Phaser phaser = new Phaser(count) { @Override protected boolean onAdvance(int phase, int registeredParties) { System.out.println("====== " + phase + " ======"); return phase >= phaseToTerminate || registeredParties == 0; } }; // for(int i = 0; i < count; i++) { System.out.println("starting thread, id: " + i); final Thread thread = new Thread(new Task(i, phaser)); thread.start(); } } public static class Task implements Runnable { // private final int id; private final Phaser phaser; public Task(int id, Phaser phaser) { this.id = id; this.phaser = phaser; } @Override public void run() { do { try { Thread.sleep(500); } catch(InterruptedException e) { // NOP } System.out.println("in Task.run(), phase: " + phaser.getPhase() + ", id: " + this.id); phaser.arriveAndAwaitAdvance(); } while(!phaser.isTerminated()); } } }</pre> <p></p> 本例中的barrier action只是简单地打印了一条信息,此外在超过指定的迭代次数后终止了Phaser。 <p><strong><br /> </strong></p> <p><strong>3.4 Sample 4</strong></p> <p>在Smaple 3的例子中,主线程在其它工作线程结束之前已经终止。如果希望主线程等待这些工作线程结束,除了使用Thread.join()之外,也可以尝试以下的方式:</p> <pre class="brush:java; toolbar: true; auto-links: false;">import java.util.concurrent.Phaser; public class PhaserTest4 { public static void main(String args[]) throws Exception { // final int count = 5; final int phaseToTerminate = 3; final Phaser phaser = new Phaser(count) { @Override protected boolean onAdvance(int phase, int registeredParties) { System.out.println("====== " + phase + " ======"); return phase == phaseToTerminate || registeredParties == 0; } }; // for(int i = 0; i < count; i++) { System.out.println("starting thread, id: " + i); final Thread thread = new Thread(new Task(i, phaser)); thread.start(); } // phaser.register(); while (!phaser.isTerminated()) { phaser.arriveAndAwaitAdvance(); } System.out.println("done"); } public static class Task implements Runnable { // private final int id; private final Phaser phaser; public Task(int id, Phaser phaser) { this.id = id; this.phaser = phaser; } @Override public void run() { while(!phaser.isTerminated()) { try { Thread.sleep(500); } catch(InterruptedException e) { // NOP } System.out.println("in Task.run(), phase: " + phaser.getPhase() + ", id: " + this.id); phaser.arriveAndAwaitAdvance(); } } } }</pre> <p></p> 如果希望主线程在特定的phase结束之后终止,那么可以在主线程中调用下述方法: <pre class="brush:java; toolbar: true; auto-links: false;">public static void awaitPhase(Phaser phaser, int phase) { int p = phaser.register(); // assumes caller not already registered while (p < phase) { if (phaser.isTerminated()) { break; // ... deal with unexpected termination } else { p = phaser.arriveAndAwaitAdvance(); } } phaser.arriveAndDeregister(); }</pre> <p></p> 需要注意的是,awaitPhase方法中的if (phaser.isTerminated()) 分支里需要能够正确处理Phaser终止的情况。否则由于在Phaser终止之后, phaser.register()和arriveAndAwaitAdvance()方法均返回负值,那么上述方法可能陷入死循环。 <p><strong><br /> </strong></p> <p><strong>3.5 Sample 5</strong></p> <p>以下对Phaser进行分层的例子:</p> <pre class="brush:java; toolbar: true; auto-links: false;">import java.util.concurrent.Phaser; public class PhaserTest6 { // private static final int TASKS_PER_PHASER = 4; public static void main(String args[]) throws Exception { // final int phaseToTerminate = 3; final Phaser phaser = new Phaser() { @Override protected boolean onAdvance(int phase, int registeredParties) { System.out.println("====== " + phase + " ======"); return phase == phaseToTerminate || registeredParties == 0; } }; // final Task tasks[] = new Task[10]; build(tasks, 0, tasks.length, phaser); for (int i = 0; i < tasks.length; i++) { System.out.println("starting thread, id: " + i); final Thread thread = new Thread(tasks[i]); thread.start(); } } public static void build(Task[] tasks, int lo, int hi, Phaser ph) { if (hi - lo > TASKS_PER_PHASER) { for (int i = lo; i < hi; i += TASKS_PER_PHASER) { int j = Math.min(i + TASKS_PER_PHASER, hi); build(tasks, i, j, new Phaser(ph)); } } else { for (int i = lo; i < hi; ++i) tasks[i] = new Task(i, ph); } } public static class Task implements Runnable { // private final int id; private final Phaser phaser; public Task(int id, Phaser phaser) { this.id = id; this.phaser = phaser; this.phaser.register(); } @Override public void run() { while (!phaser.isTerminated()) { try { Thread.sleep(200); } catch (InterruptedException e) { // NOP } System.out.println("in Task.run(), phase: " + phaser.getPhase() + ", id: " + this.id); phaser.arriveAndAwaitAdvance(); } } } }</pre> <p></p> <p>需要注意的是,TASKS_PER_PHASER的值取决于具体的Task实现。对于Task执行时间很短的场景(也就是竞争相对激烈),可以考虑使用较小的TASKS_PER_PHASER值,例如4。反之可以适当增大TASKS_PER_PHASER。</p> <p>原文出处:<a href="/misc/goto?guid=4959500085143036191" rel="nofollow">http://whitesock.iteye.com/blog/1135457</a></p>