确保所有任务完成的同步对象

Synchronisation object to ensure all tasks are completed

我应该使用哪个 Java 同步对象来确保完成任意数量的任务?约束是:

  1. 每个任务都需要很长时间才能完成,并行执行任务是合适的。
  2. 太多任务无法放入内存(即我无法将每个任务的 Future 放入 Collection,然后对所有未来调用 get)。
  3. 我不知道会有多少任务(即我不能使用CountDownLatch)。
  4. ExecutorService可能是共享的所以我不能使用awaitTermination( long, TimeUnit )

例如,对于 Grand Central Dispatch,我可能会这样做:

let workQueue = dispatch_get_global_queue( QOS_CLASS_BACKGROUND, 0 )
let latch = dispatch_group_create()
let startTime = NSDate()
var itemsProcessed = 0
let countUpdateQueue = dispatch_queue_create( "countUpdateQueue", DISPATCH_QUEUE_SERIAL )
for item in fetchItems() // generator returns too many items to store in memory
{
    dispatch_group_enter( latch )
    dispatch_async( workQueue )
    {
        self.processItem( item ) // method takes a non-trivial amount of time to run
        dispatch_async( countUpdateQueue )
        {
            itemsProcessed++
        }
        dispatch_group_leave( latch )
    }
}
dispatch_group_wait( latch, DISPATCH_TIME_FOREVER )
let endTime = NSDate()
let totalTime = endTime.timeIntervalSinceDate( startTime )
print( "Processed \(itemsProcessed) items in \(totalTime) seconds." )

它产生如下所示的输出(对于 128 个项目):Processed 128 items in 1.846794962883 seconds.

我用 Phaser 尝试了类似的东西:

final Executor executor = new ThreadPoolExecutor( 64, 64, 1l, MINUTES, new LinkedBlockingQueue<Runnable>( 8 ), new CallerRunsPolicy() );
final Phaser latch = new Phaser( 0 );
final long startTime = currentTimeMillis();
final AtomicInteger itemsProcessed = new AtomicInteger( 0 );
for( final String item : fetchItems() ) // iterator returns too many items to store in memory
{
    latch.register();
    final Runnable task = new Runnable() {
        public void run() {
            processItem( item ); // method takes a non-trivial amount of time to run
            itemsProcessed.incrementAndGet();
            latch.arrive();
        }
    };
    executor.execute( task );
}
latch.awaitAdvance( 0 );
final long endTime = currentTimeMillis();
out.println( "Processed " + itemsProcessed.get() + " items in " + ( endTime - startTime ) / 1000.0 + " seconds." );

任务并不总是在最后一个 print 语句之前完成,我可能会得到如下所示的输出(对于 128 个项目):Processed 121 items in 5.296 seconds. Phaser 是否是正确的对象?文档表明它仅支持 65,535 个参与方,因此我需要对要处理的项目进行批处理或引入某种 Phaser 分层。

此示例中 Phaser 用法的问题在于 CallerRunsPolicy 允许任务在启动线程上执行。因此,当循环仍在进行时,到达方的数量可以等于注册方的数量,从而导致阶段增加。解决方案是用1方初始化Phaser然后,当循环完成时,到达并等待其他方到达。这确保在所有任务完成之前阶段不会增加到 1。

final Executor executor = new ThreadPoolExecutor( 64, 64, 1l, MINUTES, new LinkedBlockingQueue<Runnable>( 8 ), new CallerRunsPolicy() );
final Phaser latch = new Phaser( 1 );
final long startTime = currentTimeMillis();
final AtomicInteger itemsProcessed = new AtomicInteger( 0 );
for( final String item : fetchItems() ) // iterator returns too many items to store in memory
{
    latch.register();
    final Runnable task = new Runnable() {
        public void run() {
            processItem( item ); // method takes a non-trivial amount of time to run
            itemsProcessed.incrementAndGet();
            final int arrivalPhase = latch.arrive();
        }
    };
    executor.execute( task );
}
latch.arriveAndAwaitAdvance();
final long endTime = currentTimeMillis();
out.println( "Processed " + itemsProcessed.get() + " items in " + ( endTime - startTime ) / 1000.0 + " seconds." );

万一你在 java8 你可以使用 CompletableFuture

java.util.concurrent.CompletableFuture.allOf(CompletableFuture<?>... cfs)

这将等待传递数组中所有期货的结果。

"to ensure an arbitrarily large number of tasks are completed" - 最简单的方法是维护已完成任务的计数器,阻塞操作等待达到给定数量的任务。没有现成的class,但很容易制作:

class EventCounter {
   long counter=0;

   synchronized void up () {
     counter++;
     notifyAll();
   }
   synchronized void ensure (long count) {
     while (counter<count) wait();
   }
 }

"There are too many tasks to fit into memory" - 所以当运行任务数量过多时,必须暂停提交新任务的过程。最简单的方法就是把运行个任务的数量看成是一个资源,用一个信号量来统计:

Semaphore runningTasksSema=new Semaphore(maxNumberOfRunningTasks);
EventCounter  eventCounter =new EventCounter ();

for( final String item : fetchItems() ) {
    final Runnable task = new Runnable() {
       public void run() {
            processItem( item ); 
            runningTasksSema.release();
            eventCounter.up();
       }
    };
   runningTasksSema.aquire();
   executor.execute(task);
}

当一个线程想要确保完成一些给定数量的任务时,它会调用:

eventCounter.ensure(givenNumberOfFinishedTasks);

可以设计 runningTasksSema.aquire()eventCounter.ensure() 操作的异步(非阻塞)版本,但它们会更复杂。