转自：
今天看到一篇文章，是关于java中如何等待所有线程都执行结束，文章总结得很好，原文如下
看过之后在想java中有很大的灵活性，应该有更多的方式可以做这件事。
这个事情的场景是这样的：许多线程并行的计算一堆问题，然后每个计算存在一个队列，在主线程要等待所有计算结果完成后排序并展示出来。这样的问题其实很常见。
1. 使用join。这种方式其实并不是那么的优雅，将所有线程启动完之后还需要将所有线程都join，但是每次join都会阻塞，直到被join线程完成，很可能所有被阻塞线程已经完事了，主线程还在不断地join，貌似有点浪费，而且两个循环也不太好看。
-
1 public void testThreadSync1() { 2 3 final Vectorlist = new Vector (); 4 Thread[] threads = new Thread[TEST_THREAD_COUNT]; 5 try { 6 for (int i = 0; i < TEST_THREAD_COUNT; i++) { 7 final int num = i; 8 threads[i] = new Thread(new Runnable() { 9 public void run() { 10 try { 11 Thread.sleep(random.nextInt(100)); 12 } catch (InterruptedException e) { 13 e.printStackTrace(); 14 } 15 list.add(num); 16 System.out.print(num + " add.\t"); 17 } 18 }); 19 threads[i].start(); 20 } 21 for (int i = 0; i < threads.length; i++) { 22 threads[i].join(); 23 System.out.print(i + " end.\t"); 24 } 25 } catch (InterruptedException ie) { 26 ie.printStackTrace(); 27 } 28 printSortedResult(list); 29 }
-
1 9 add. 7 add. 3 add. 5 add. 4 add. 1 add. 0 add. 0 end. 1 end. 8 add. 2 add. 2 end. 3 end. 4 end. 5 end. 6 add. 6 end. 7 end. 8 end. 9 end. 2 before sort 3 9 7 3 5 4 1 0 8 2 6 4 after sort 5 0 1 2 3 4 5 6 7 8 9
2. 使用wait/notifyAll，这个方式其实跟上面是类似的，只是比较底层些吧(join实际上也是wait)。
-
1 @Test 2 public void testThreadSync2() throws IOException, InterruptedException { 3 final Object waitObject = new Object(); 4 final AtomicInteger count = new AtomicInteger(TEST_THREAD_COUNT); 5 final Vectorlist = new Vector (); 6 Thread[] threads = new Thread[TEST_THREAD_COUNT]; 7 for (int i = 0; i < TEST_THREAD_COUNT; i++) { 8 final int num = i; 9 threads[i] = new Thread(new Runnable() { 10 public void run() { 11 try { 12 Thread.sleep(random.nextInt(100)); 13 } catch (InterruptedException e) { 14 e.printStackTrace(); 15 } 16 list.add(num); 17 System.out.print(num + " add.\t"); 18 synchronized (waitObject) { 19 int cnt = count.decrementAndGet(); 20 if (cnt == 0) { 21 waitObject.notifyAll(); 22 } 23 } 24 } 25 }); 26 threads[i].start(); 27 } 28 synchronized (waitObject) { 29 while (count.get() != 0) { 30 waitObject.wait(); 31 } 32 } 33 printSortedResult(list); 34 }
3. 使用CountDownLatch，这其实是最优雅的写法了，每个线程完成后都去将计数器减一，最后完成时再来唤醒。
例1
-
1 @Test 2 public void testThreadSync3() { 3 final Vectorlist = new Vector (); 4 Thread[] threads = new Thread[TEST_THREAD_COUNT]; 5 final CountDownLatch latch = new CountDownLatch(TEST_THREAD_COUNT); 6 for (int i = 0; i < TEST_THREAD_COUNT; i++) { 7 final int num = i; 8 threads[i] = new Thread(new Runnable() { 9 public void run() { 10 try { 11 Thread.sleep(random.nextInt(100)); 12 } catch (InterruptedException e) { 13 e.printStackTrace(); 14 } 15 list.add(num); 16 System.out.print(num + " add.\t"); 17 latch.countDown(); 18 } 19 }); 20 threads[i].start(); 21 } 22 try { 23 latch.await(); 24 } catch (InterruptedException e) { 25 e.printStackTrace(); 26 } 27 printSortedResult(list); 28 }
例2
CountDownLatch 初始化设置count，即等待(await)count个线程或一个线程count次计数，通过工作线程来countDown计数减一，直到计数为0，await阻塞结束。
设置的count不可更改，如需要动态设置计数的线程数，可以使用CyclicBarrier.
下面的例子，所有的工作线程中准备就绪以后，并不是直接运行，而是等待主线程的信号后再执行具体的操作。
-
1 package com.example.multithread; 2 3 import java.util.concurrent.CountDownLatch; 4 5 class Driver 6 { 7 private static final int TOTAL_THREADS = 10; 8 private final CountDownLatch mStartSignal = new CountDownLatch(1); 9 private final CountDownLatch mDoneSignal = new CountDownLatch(TOTAL_THREADS); 10 11 void main() 12 { 13 for (int i = 0; i < TOTAL_THREADS; i++) 14 { 15 new Thread(new Worker(mStartSignal, mDoneSignal, i)).start(); 16 } 17 System.out.println("Main Thread Now:" + System.currentTimeMillis()); 18 doPrepareWork();// 准备工作 19 mStartSignal.countDown();// 计数减一为0，工作线程真正启动具体操作 20 doSomethingElse();//做点自己的事情 21 try 22 { 23 mDoneSignal.await();// 等待所有工作线程结束 24 } 25 catch (InterruptedException e) 26 { 27 // TODO Auto-generated catch block 28 e.printStackTrace(); 29 } 30 System.out.println("All workers have finished now."); 31 System.out.println("Main Thread Now:" + System.currentTimeMillis()); 32 } 33 34 void doPrepareWork() 35 { 36 System.out.println("Ready,GO!"); 37 } 38 39 void doSomethingElse() 40 { 41 for (int i = 0; i < 100000; i++) 42 { 43 ;// delay 44 } 45 System.out.println("Main Thread Do something else."); 46 } 47 } 48 49 class Worker implements Runnable 50 { 51 private final CountDownLatch mStartSignal; 52 private final CountDownLatch mDoneSignal; 53 private final int mThreadIndex; 54 55 Worker(final CountDownLatch startSignal, final CountDownLatch doneSignal, 56 final int threadIndex) 57 { 58 this.mDoneSignal = doneSignal; 59 this.mStartSignal = startSignal; 60 this.mThreadIndex = threadIndex; 61 } 62 63 @Override 64 public void run() 65 { 66 // TODO Auto-generated method stub 67 try 68 { 69 mStartSignal.await();// 阻塞，等待mStartSignal计数为0运行后面的代码 70 // 所有的工作线程都在等待同一个启动的命令 71 doWork();// 具体操作 72 System.out.println("Thread " + mThreadIndex + " Done Now:" 73 + System.currentTimeMillis()); 74 mDoneSignal.countDown();// 完成以后计数减一 75 } 76 catch (InterruptedException e) 77 { 78 // TODO Auto-generated catch block 79 e.printStackTrace(); 80 } 81 } 82 83 public void doWork() 84 { 85 for (int i = 0; i < 1000000; i++) 86 { 87 ;// 耗时操作 88 } 89 System.out.println("Thread " + mThreadIndex + ":do work"); 90 } 91 } 92 93 public class CountDownLatchTest 94 { 95 public static void main(String[] args) 96 { 97 // TODO Auto-generated method stub 98 new Driver().main(); 99 } 100 101 }通过Executor启动线程：
-
1 class CountDownLatchDriver2 2 { 3 private static final int TOTAL_THREADS = 10; 4 private final CountDownLatch mDoneSignal = new CountDownLatch(TOTAL_THREADS); 5 6 7 8 void main() 9 { 10 System.out.println("Main Thread Now:" + System.currentTimeMillis()); 11 doPrepareWork();// 准备工作 12 13 Executor executor = Executors.newFixedThreadPool(TOTAL_THREADS); 14 for (int i = 0; i < TOTAL_THREADS; i++) 15 { 16 // 通过内建的线程池维护创建的线程 17 executor.execute(new RunnableWorker(mDoneSignal, i)); 18 } 19 doSomethingElse();// 做点自己的事情 20 try 21 { 22 mDoneSignal.await();// 等待所有工作线程结束 23 } 24 catch (InterruptedException e) 25 { 26 // TODO Auto-generated catch block 27 e.printStackTrace(); 28 } 29 System.out.println("All workers have finished now."); 30 System.out.println("Main Thread Now:" + System.currentTimeMillis()); 31 } 32 33 void doPrepareWork() 34 { 35 System.out.println("Ready,GO!"); 36 } 37 38 void doSomethingElse() 39 { 40 for (int i = 0; i < 100000; i++) 41 { 42 ;// delay 43 } 44 System.out.println("Main Thread Do something else."); 45 } 46 } 47 48 class RunnableWorker implements Runnable 49 { 50 51 private final CountDownLatch mDoneSignal; 52 private final int mThreadIndex; 53 54 RunnableWorker(final CountDownLatch doneSignal, final int threadIndex) 55 { 56 this.mDoneSignal = doneSignal; 57 this.mThreadIndex = threadIndex; 58 } 59 60 @Override 61 public void run() 62 { 63 // TODO Auto-generated method stub 64 65 doWork();// 具体操作 66 System.out.println("Thread " + mThreadIndex + " Done Now:" 67 + System.currentTimeMillis()); 68 mDoneSignal.countDown();// 完成以后计数减一 69 // 计数为0时，主线程接触阻塞，继续执行其他任务 70 try 71 { 72 // 可以继续做点其他的事情，与主线程无关了 73 Thread.sleep(5000); 74 System.out.println("Thread " + mThreadIndex 75 + " Do something else after notifing main thread"); 76 77 } 78 catch (InterruptedException e) 79 { 80 // TODO Auto-generated catch block 81 e.printStackTrace(); 82 } 83 84 } 85 86 public void doWork() 87 { 88 for (int i = 0; i < 1000000; i++) 89 { 90 ;// 耗时操作 91 } 92 System.out.println("Thread " + mThreadIndex + ":do work"); 93 } 94 }输出：
1 Main Thread Now:1359959480786 2 Ready,GO! 3 Thread 0:do work 4 Thread 0 Done Now:1359959480808 5 Thread 1:do work 6 Thread 1 Done Now:1359959480811 7 Thread 2:do work 8 Thread 2 Done Now:1359959480813 9 Main Thread Do something else.10 Thread 3:do work11 Thread 3 Done Now:135995948082512 Thread 5:do work13 Thread 5 Done Now:135995948082714 Thread 7:do work15 Thread 7 Done Now:135995948082916 Thread 9:do work17 Thread 9 Done Now:135995948083118 Thread 4:do work19 Thread 4 Done Now:135995948083320 Thread 6:do work21 Thread 6 Done Now:135995948083522 Thread 8:do work23 Thread 8 Done Now:135995948083724 All workers have finished now.25 Main Thread Now:135995948083826 Thread 0 Do something else after notifing main thread27 Thread 1 Do something else after notifing main thread28 Thread 2 Do something else after notifing main thread29 Thread 3 Do something else after notifing main thread30 Thread 9 Do something else after notifing main thread31 Thread 7 Do something else after notifing main thread32 Thread 5 Do something else after notifing main thread33 Thread 4 Do something else after notifing main thread34 Thread 6 Do something else after notifing main thread35 Thread 8 Do something else after notifing main thread
-
4. 使用CyclicBarrier。这里其实类似上面，这个berrier只是在等待完成后自动调用传入CyclicBarrier的Runnable。
例1
-
1 @Test 2 public void testThreadSync4() throws IOException { 3 final Vectorlist = new Vector (); 4 Thread[] threads = new Thread[TEST_THREAD_COUNT]; 5 final CyclicBarrier barrier = new CyclicBarrier(TEST_THREAD_COUNT, 6 new Runnable() { 7 public void run() { 8 printSortedResult(list); 9 } 10 }); 11 for (int i = 0; i < TEST_THREAD_COUNT; i++) { 12 final int num = i; 13 threads[i] = new Thread(new Runnable() { 14 public void run() { 15 try { 16 Thread.sleep(random.nextInt(100)); 17 } catch (InterruptedException e) { 18 e.printStackTrace(); 19 } 20 list.add(num); 21 System.out.print(num + " add.\t"); 22 try { 23 barrier.await(); 24 } catch (InterruptedException e) { 25 e.printStackTrace(); 26 } catch (BrokenBarrierException e) { 27 e.printStackTrace(); 28 } 29 } 30 }); 31 threads[i].start(); 32 } 33 System.in.read(); 34 } 例2
-
1 class WalkTarget 2 { 3 private final int mCount = 5; 4 private final CyclicBarrier mBarrier; 5 ExecutorService mExecutor; 6 7 class BarrierAction implements Runnable 8 { 9 @Override 10 public void run() 11 { 12 // TODO Auto-generated method stub 13 System.out.println("所有线程都已经完成任务,计数达到预设值"); 14 //mBarrier.reset();//恢复到初始化状态 15 16 } 17 } 18 19 WalkTarget() 20 { 21 //初始化CyclicBarrier 22 mBarrier = new CyclicBarrier(mCount, new BarrierAction()); 23 mExecutor = Executors.newFixedThreadPool(mCount); 24 25 for (int i = 0; i < mCount; i++) 26 { 27 //启动工作线程 28 mExecutor.execute(new Walker(mBarrier, i)); 29 } 30 } 31 } 32 33 //工作线程 34 class Walker implements Runnable 35 { 36 private final CyclicBarrier mBarrier; 37 private final int mThreadIndex; 38 39 Walker(final CyclicBarrier barrier, final int threadIndex) 40 41 42 { 43 mBarrier = barrier; 44 mThreadIndex = threadIndex; 45 } 46 47 @Override 48 public void run() 49 { 50 // TODO Auto-generated method stub 51 System.out.println("Thread " + mThreadIndex + " is running..."); 52 // 执行任务 53 try 54 { 55 TimeUnit.MILLISECONDS.sleep(5000); 56 // do task 57 } 58 catch (InterruptedException e) 59 { 60 // TODO Auto-generated catch block 61 e.printStackTrace(); 62 } 63 64 // 完成任务以后，等待其他线程完成任务 65 try 66 { 67 mBarrier.await(); 68 } 69 catch (InterruptedException e) 70 { 71 // TODO Auto-generated catch block 72 e.printStackTrace(); 73 } 74 catch (BrokenBarrierException e) 75 { 76 // TODO Auto-generated catch block 77 e.printStackTrace(); 78 } 79 // 其他线程任务都完成以后，阻塞解除，可以继续接下来的任务 80 System.out.println("Thread " + mThreadIndex + " do something else"); 81 } 82 83 } 84 85 public class CountDownLatchTest 86 { 87 public static void main(String[] args) 88 { 89 // TODO Auto-generated method stub 90 //new CountDownLatchDriver2().main(); 91 new WalkTarget(); 92 } 93 94 }输出(注意，只有所有的线程barrier.await之后才能继续执行其他的操作）：
Thread 0 is running... Thread 2 is running... Thread 3 is running... Thread 1 is running... Thread 4 is running... 所有线程都已经完成任务,计数达到预设值 Thread 4 do something else Thread 0 do something else Thread 2 do something else Thread 3 do something else Thread 1 do something else
-
5、
CountDownLatch和CyclicBarrier简单比较：
|
| CountDownLatch | CyclicBarrier |
|---|---|---|
| 软件包 | java.util.concurrent | java.util.concurrent |
| 适用情景 | 主线程等待多个工作线程结束 | 多个线程之间互相等待，直到所有线程达到一个障碍点(Barrier point) |
| 主要方法 | CountDownLatch(int count) （主线程调用） 初始化计数 CountDownLatch.await （主线程调用） 阻塞，直到等待计数为0解除阻塞 CountDownLatch.countDown 计数减一（工作线程调用） | CyclicBarrier(int parties, Runnable barrierAction) //初始化参与者数量和障碍点执行Action，Action可选。由主线程初始化 CyclicBarrier.await() //由参与者调用 阻塞，直到所有线程达到屏障点 |
| 等待结束 | 各线程之间不再互相影响，可以继续做自己的事情。不再执行下一个目标工作。 | 在屏障点达到后，允许所有线程继续执行，达到下一个目标。可以重复使用CyclicBarrier |
| 异常 |
| 如果其中一个线程由于中断，错误，或超时导致永久离开屏障点，其他线程也将抛出异常。 |
| 其他 |
| 如果BarrierAction不依赖于任何Party中的所有线程，那么在任何party中的一个线程被释放的时候，可以直接运行这个Action。 If(barrier.await()==2) { //do action } |