Question

有没有办法创建Executor，它总是至少有5个线程，最多20个线程，以及任务的无限队列（意味着没有任务被拒绝）

我尝试了新ThreadPoolExecutor(5, 20, 60L, TimeUnit.SECONDS, queue) 我想到了队列的所有可能性：

new LinkedBlockingQueue() // never runs more than 5 threads
new LinkedBlockingQueue(1000000) // runs more than 5 threads, only when there is more than 1000000 tasks waiting
new ArrayBlockingQueue(1000000) // runs more than 5 threads, only when there is more than 1000000 tasks waiting
new SynchronousQueue() // no tasks can wait, after 20, they are rejected

并且没有按预期工作。

Answer 1

这样的事情对你有用吗？我只是把它掀起来所以请捅它。基本上，它实现了一个溢出线程池，用于提供底层ThreadPoolExecutor

我看到它有两个主要的缺点：

submit()上缺少返回的Future对象。但也许这对你来说不是问题。
只有在提交作业时，辅助队列才会清空ThreadPoolExecutor。必须有一个优雅的解决方案，但我还没有看到它。如果你知道StusMagicExecutor中会有一连串的任务，那么这可能不是问题。（“可能”是关键词。）一个选项可能是让您提交的任务在完成后StusMagicExecutor点击？

斯图的魔法执行官：

public class StusMagicExecutor extends ThreadPoolExecutor {
    private BlockingQueue<Runnable> secondaryQueue = new LinkedBlockingQueue<Runnable>();  //capacity is Integer.MAX_VALUE.

    public StusMagicExecutor() {
        super(5, 20, 60L, SECONDS, new SynchronousQueue<Runnable>(true), new RejectionHandler());  
    }
    public void queueRejectedTask(Runnable task) {
        try {
            secondaryQueue.put(task);
        } catch (InterruptedException e) {
            // do something
        }
    }
    public Future submit(Runnable newTask) {
        //drain secondary queue as rejection handler populates it
        Collection<Runnable> tasks = new ArrayList<Runnable>();
        secondaryQueue.drainTo(tasks);

        tasks.add(newTask);

        for (Runnable task : tasks)
             super.submit(task);

        return null; //does not return a future!
    }
}

class RejectionHandler implements RejectedExecutionHandler {
    public void rejectedExecution(Runnable runnable, ThreadPoolExecutor executor) {
        ((StusMagicExecutor)executor).queueRejectedTask(runnable);
    }
}

Answer 2

ThreadPoolExecutor的javadoc非常清楚，一旦创建corePoolSize个线程，只有队列满了才会创建新线程。因此，如果您将core设置为5并将max设置为20，那么您将无法获得所需的行为。

但是，如果将core和max都设置为20，那么只有当所有20个线程都忙时，才会将任务添加到队列中。当然，这会使你的“5线程最小”要求变得毫无意义，因为所有20个都将保持活着（直到它们闲置，无论如何）。

Answer 3

我认为这个问题是类的缺点，并且在给定构造函数参数组合的情况下会产生误导。这是从SwingWorker的内部ThreadPoolExecutor获取的解决方案，我将其作为顶级类。它没有最小值，但至少使用上限。我唯一不知道的是你从锁定执行中获得的性能。

public class BoundedThreadPoolExecutor extends ThreadPoolExecutor {
    private final ReentrantLock pauseLock = new ReentrantLock();
    private final Condition unpaused = pauseLock.newCondition();
    private boolean isPaused = false;
    private final ReentrantLock executeLock = new ReentrantLock();

    public BoundedThreadPoolExecutor(int maximumPoolSize,
            long keepAliveTime, TimeUnit unit, BlockingQueue<Runnable> workQueue) {
        super(0, maximumPoolSize, keepAliveTime, unit, workQueue);
    }

    public BoundedThreadPoolExecutor(int maximumPoolSize,
            long keepAliveTime, TimeUnit unit,
        BlockingQueue<Runnable> workQueue, ThreadFactory threadFactory) {
        super(0, maximumPoolSize, keepAliveTime, unit, workQueue,
                threadFactory);
    }

    public BoundedThreadPoolExecutor(int maximumPoolSize,
            long keepAliveTime, TimeUnit unit,
            BlockingQueue<Runnable> workQueue, RejectedExecutionHandler handler) {
        super(0, maximumPoolSize, keepAliveTime, unit, workQueue,
                handler);
    }

    public BoundedThreadPoolExecutor(int maximumPoolSize,
            long keepAliveTime, TimeUnit unit,
            BlockingQueue<Runnable> workQueue, ThreadFactory threadFactory,
            RejectedExecutionHandler handler) {
        super(0, maximumPoolSize, keepAliveTime, unit, workQueue,
                threadFactory, handler);
    }

    @Override
    public void execute(Runnable command) {
        executeLock.lock();
        try {
            pauseLock.lock();
            try {
                isPaused = true;
            } finally {
                pauseLock.unlock();
            }
            setCorePoolSize(getMaximumPoolSize());
            super.execute(command);
            setCorePoolSize(0);
            pauseLock.lock();
            try {
                isPaused = false;
                unpaused.signalAll();
            } finally {
                pauseLock.unlock();
            }
        } finally {
            executeLock.unlock();
        }
    }

    @Override
    protected void afterExecute(Runnable r, Throwable t) {
        super.afterExecute(r, t);
        pauseLock.lock();
        try {
            while (isPaused) {
                unpaused.await();
            }
        } catch (InterruptedException ignore) {

        } finally {
            pauseLock.unlock();
        }
    }
}

指定ThreadPoolExecutor问题

3 个答案: