Java 中的 ForkJoin 任务

Java 中的 fork / join 框架非常适合于可以分成更小部分并且并行求解的问题。fork / join 问题的基本步骤是:

  • 将问题分成多个部分
  • 将每个部件彼此平行地求解
  • 将每个子解决方案组合成一个整体解决方案

一个 ForkJoinTask 是定义这样的问题的接口。通常期望你将其一个抽象实现(通常是 RecursiveTask ) 子类化,而不是直接实现接口。

在这个例子中,我们将总结一个整数的集合,直到我们得到不超过 10 的批量大小。

import java.util.List;
import java.util.concurrent.RecursiveTask;

public class SummingTask extends RecursiveTask<Integer> {
    private static final int MAX_BATCH_SIZE = 10;

    private final List<Integer> numbers;
    private final int minInclusive, maxExclusive;

    public SummingTask(List<Integer> numbers) {
        this(numbers, 0, numbers.size());
    }

    // This constructor is only used internally as part of the dividing process
    private SummingTask(List<Integer> numbers, int minInclusive, int maxExclusive) {
        this.numbers = numbers;
        this.minInclusive = minInclusive;
        this.maxExclusive = maxExclusive;
    }

    @Override
    public Integer compute() {
        if (maxExclusive - minInclusive > MAX_BATCH_SIZE) {
            // This is too big for a single batch, so we shall divide into two tasks
            int mid = (minInclusive + maxExclusive) / 2;
            SummingTask leftTask = new SummingTask(numbers, minInclusive, mid);
            SummingTask rightTask = new SummingTask(numbers, mid, maxExclusive);

            // Submit the left hand task as a new task to the same ForkJoinPool
            leftTask.fork();

            // Run the right hand task on the same thread and get the result
            int rightResult = rightTask.compute();

            // Wait for the left hand task to complete and get its result
            int leftResult = leftTask.join();

            // And combine the result
            return leftResult + rightResult;
        } else {
            // This is fine for a single batch, so we will run it here and now
            int sum = 0;
            for (int i = minInclusive; i < maxExclusive; i++) {
                sum += numbers.get(i);
            }
            return sum;
        }
    }
}

现在可以将此任务的实例传递给 ForkJoinPool 的实例。

// Because I am not specifying the number of threads
// it will create a thread for each available processor
ForkJoinPool pool = new ForkJoinPool();

// Submit the task to the pool, and get what is effectively the Future
ForkJoinTask<Integer> task = pool.submit(new SummingTask(numbers));

// Wait for the result
int result = task.join();