Java：ExecutorService在特定队列大小后会在提交时阻止

Question 1

我正在尝试编写一个解决方案，其中单个线程会产生可并行执行的I / O密集型任务。每个任务都有重要的内存数据。因此，我希望能够限制当前待处理的任务数。

如果我这样创建ThreadPoolExecutor：

    ThreadPoolExecutor executor = new ThreadPoolExecutor(numWorkerThreads, numWorkerThreads,
                                  0L, TimeUnit.MILLISECONDS,
                                  new LinkedBlockingQueue<Runnable>(maxQueue));

然后，当队列已满并且所有线程都已经繁忙时，executor.submit(callable)抛出该异常RejectedExecutionException。

executor.submit(callable)当队列已满并且所有线程都忙时，我该怎么做才能阻塞？

编辑：我试过了：

executor.setRejectedExecutionHandler(new ThreadPoolExecutor.CallerRunsPolicy());

它以某种微妙的方式达到了我想要达到的效果（基本上被拒绝的线程在调用线程中运行，因此这阻止了调用线程提交更多信息）。

编辑：（问了问题5年后）

对于阅读此问题及其答案的任何人，请勿将接受的答案作为一种正确的解决方案。请通读所有答案和评论。

Question 2

我也做过同样的事情。诀窍是创建一个BlockingQueue，其中offer（）方法实际上是put（）。（您可以使用所需的任何基本BlockingQueue隐式表示形式）。

public class LimitedQueue<E> extends LinkedBlockingQueue<E> 
{
    public LimitedQueue(int maxSize)
    {
        super(maxSize);
    }

    @Override
    public boolean offer(E e)
    {
        // turn offer() and add() into a blocking calls (unless interrupted)
        try {
            put(e);
            return true;
        } catch(InterruptedException ie) {
            Thread.currentThread().interrupt();
        }
        return false;
    }

}

请注意，这仅适用于线程池，corePoolSize==maxPoolSize因此请务必注意（请参阅注释）。

Question 3

这是我最终解决此问题的方法：

（注意：此解决方案确实会阻止提交Callable的线程，因此可以防止抛出RejectedExecutionException）

public class BoundedExecutor extends ThreadPoolExecutor{

    private final Semaphore semaphore;

    public BoundedExecutor(int bound) {
        super(bound, Integer.MAX_VALUE, 60L, TimeUnit.SECONDS, new SynchronousQueue<Runnable>());
        semaphore = new Semaphore(bound);
    }

    /**Submits task to execution pool, but blocks while number of running threads 
     * has reached the bound limit
     */
    public <T> Future<T> submitButBlockIfFull(final Callable<T> task) throws InterruptedException{

        semaphore.acquire();            
        return submit(task);                    
    }


    @Override
    protected void afterExecute(Runnable r, Throwable t) {
        super.afterExecute(r, t);

        semaphore.release();
    }
}

Question 4

当前接受的答案有一个潜在的重大问题-它更改了ThreadPoolExecutor.execute的行为，因此，如果您具有corePoolSize < maxPoolSize，则ThreadPoolExecutor逻辑将永远不会在核心之外添加其他工作线程。

从ThreadPoolExecutor .execute（Runnable）：

    if (isRunning(c) && workQueue.offer(command)) {
        int recheck = ctl.get();
        if (! isRunning(recheck) && remove(command))
            reject(command);
        else if (workerCountOf(recheck) == 0)
            addWorker(null, false);
    }
    else if (!addWorker(command, false))
        reject(command);

具体来说，最后一个“ else”块将永远不会被击中。

更好的替代方法是执行与OP已经执行的操作类似的操作-使用RejectedExecutionHandler来执行相同的put逻辑：

public void rejectedExecution(Runnable r, ThreadPoolExecutor executor) {
    try {
        if (!executor.isShutdown()) {
            executor.getQueue().put(r);
        }
    } catch (InterruptedException e) {
        Thread.currentThread().interrupt();
        throw new RejectedExecutionException("Executor was interrupted while the task was waiting to put on work queue", e);
    }
}

正如评论中所指出的（请参考此答案），使用这种方法需要注意一些事项：

如果为corePoolSize==0，则存在争用条件，其中池中的所有线程可能在可见任务之前死亡
使用包装队列任务的实现（不适用于ThreadPoolExecutor）将导致问题，除非处理程序也以相同的方式包装它。

牢记这些陷阱，此解决方案适用于大多数典型的ThreadPoolExecutor，并且可以正确处理where的情况corePoolSize < maxPoolSize。

Question 5

我知道这是一个老问题，但是存在一个类似的问题，即创建新任务的速度非常快，并且由于现有任务执行得不够快而导致OutOfMemoryError发生的次数过多。

在我的情况下Callables，提交并且我需要结果，因此我需要存储所有由Futures返回executor.submit()。我的解决方案是将尺寸Futures设为BlockingQueue最大。一旦该队列已满，直到完成一些任务（从队列中删除元素）后，才会生成更多任务。用伪代码：

final ExecutorService executor = Executors.newFixedThreadPool(numWorkerThreads);
final LinkedBlockingQueue<Future> futures = new LinkedBlockingQueue<>(maxQueueSize);
try {   
    Thread taskGenerator = new Thread() {
        @Override
        public void run() {
            while (reader.hasNext) {
                Callable task = generateTask(reader.next());
                Future future = executor.submit(task);
                try {
                    // if queue is full blocks until a task
                    // is completed and hence no future tasks are submitted.
                    futures.put(compoundFuture);
                } catch (InterruptedException ex) {
                    Thread.currentThread().interrupt();         
                }
            }
        executor.shutdown();
        }
    }
    taskGenerator.start();

    // read from queue as long as task are being generated
    // or while Queue has elements in it
    while (taskGenerator.isAlive()
                    || !futures.isEmpty()) {
        Future compoundFuture = futures.take();
        // do something
    }
} catch (InterruptedException ex) {
    Thread.currentThread().interrupt();     
} catch (ExecutionException ex) {
    throw new MyException(ex);
} finally {
    executor.shutdownNow();
}

Question 6

我有类似的问题，我通过使用beforeExecute/afterExecute钩子实现了ThreadPoolExecutor：

import java.util.concurrent.BlockingQueue;
import java.util.concurrent.ThreadPoolExecutor;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.ReentrantLock;

/**
 * Blocks current task execution if there is not enough resources for it.
 * Maximum task count usage controlled by maxTaskCount property.
 */
public class BlockingThreadPoolExecutor extends ThreadPoolExecutor {

    private final ReentrantLock taskLock = new ReentrantLock();
    private final Condition unpaused = taskLock.newCondition();
    private final int maxTaskCount;

    private volatile int currentTaskCount;

    public BlockingThreadPoolExecutor(int corePoolSize, int maximumPoolSize,
            long keepAliveTime, TimeUnit unit,
            BlockingQueue<Runnable> workQueue, int maxTaskCount) {
        super(corePoolSize, maximumPoolSize, keepAliveTime, unit, workQueue);
        this.maxTaskCount = maxTaskCount;
    }

    /**
     * Executes task if there is enough system resources for it. Otherwise
     * waits.
     */
    @Override
    protected void beforeExecute(Thread t, Runnable r) {
        super.beforeExecute(t, r);
        taskLock.lock();
        try {
            // Spin while we will not have enough capacity for this job
            while (maxTaskCount < currentTaskCount) {
                try {
                    unpaused.await();
                } catch (InterruptedException e) {
                    t.interrupt();
                }
            }
            currentTaskCount++;
        } finally {
            taskLock.unlock();
        }
    }

    /**
     * Signalling that one more task is welcome
     */
    @Override
    protected void afterExecute(Runnable r, Throwable t) {
        super.afterExecute(r, t);
        taskLock.lock();
        try {
            currentTaskCount--;
            unpaused.signalAll();
        } finally {
            taskLock.unlock();
        }
    }
}

这对您来说应该足够了。顺便说一句，最初的实现是基于任务大小的，因为一个任务可能比另一个任务大100倍，而提交两个巨大的任务将使整个工作陷入瘫痪，但是运行一个大而小的任务就可以了。如果您的I / O密集型任务的大小大致相同，则可以使用此类，否则，请告诉我，我将发布基于大小的实现。

PS：您需要检查ThreadPoolExecutorjavadoc。这是Doug Lea很好的用户指南，说明如何轻松定制它。

Question 7

我已经按照装饰器模式实现了一个解决方案，并使用信号量来控制已执行任务的数量。您可以将其与以下任何一个一起使用Executor：

指定正在进行的任务的最大数量
指定等待任务执行许可的最大超时（如果超时过去并且没有获得许可，RejectedExecutionException则抛出a）

import static java.util.concurrent.TimeUnit.MILLISECONDS;

import java.time.Duration;
import java.util.Objects;
import java.util.concurrent.Executor;
import java.util.concurrent.RejectedExecutionException;
import java.util.concurrent.Semaphore;

import javax.annotation.Nonnull;

public class BlockingOnFullQueueExecutorDecorator implements Executor {

    private static final class PermitReleasingDecorator implements Runnable {

        @Nonnull
        private final Runnable delegate;

        @Nonnull
        private final Semaphore semaphore;

        private PermitReleasingDecorator(@Nonnull final Runnable task, @Nonnull final Semaphore semaphoreToRelease) {
            this.delegate = task;
            this.semaphore = semaphoreToRelease;
        }

        @Override
        public void run() {
            try {
                this.delegate.run();
            }
            finally {
                // however execution goes, release permit for next task
                this.semaphore.release();
            }
        }

        @Override
        public final String toString() {
            return String.format("%s[delegate='%s']", getClass().getSimpleName(), this.delegate);
        }
    }

    @Nonnull
    private final Semaphore taskLimit;

    @Nonnull
    private final Duration timeout;

    @Nonnull
    private final Executor delegate;

    public BlockingOnFullQueueExecutorDecorator(@Nonnull final Executor executor, final int maximumTaskNumber, @Nonnull final Duration maximumTimeout) {
        this.delegate = Objects.requireNonNull(executor, "'executor' must not be null");
        if (maximumTaskNumber < 1) {
            throw new IllegalArgumentException(String.format("At least one task must be permitted, not '%d'", maximumTaskNumber));
        }
        this.timeout = Objects.requireNonNull(maximumTimeout, "'maximumTimeout' must not be null");
        if (this.timeout.isNegative()) {
            throw new IllegalArgumentException("'maximumTimeout' must not be negative");
        }
        this.taskLimit = new Semaphore(maximumTaskNumber);
    }

    @Override
    public final void execute(final Runnable command) {
        Objects.requireNonNull(command, "'command' must not be null");
        try {
            // attempt to acquire permit for task execution
            if (!this.taskLimit.tryAcquire(this.timeout.toMillis(), MILLISECONDS)) {
                throw new RejectedExecutionException(String.format("Executor '%s' busy", this.delegate));
            }
        }
        catch (final InterruptedException e) {
            // restore interrupt status
            Thread.currentThread().interrupt();
            throw new IllegalStateException(e);
        }

        this.delegate.execute(new PermitReleasingDecorator(command, this.taskLimit));
    }

    @Override
    public final String toString() {
        return String.format("%s[availablePermits='%s',timeout='%s',delegate='%s']", getClass().getSimpleName(), this.taskLimit.availablePermits(),
                this.timeout, this.delegate);
    }
}

Question 8

我认为这就像使用aArrayBlockingQueue而不是aa一样简单LinkedBlockingQueue。

别理我...那是完全错误的。不会产生您需要的效果的ThreadPoolExecutor电话。Queue#offerput

您可以代替扩展ThreadPoolExecutor并提供execute(Runnable)该调用的实现。putoffer

恐怕这似乎不是一个完全令人满意的答案。