為什么ParallelStream不會在遞回中使用所有commonPool的執行緒？

當我運行以下代碼時，8 個可用執行緒中只有 2 個運行，誰能解釋為什么會這樣？如何更改代碼以利用所有 8 個執行緒？

Tree.java：

package il.co.roy;

import java.util.HashSet;
import java.util.Objects;
import java.util.Set;

public class Tree<T>
{
    private final T data;
    private final Set<Tree<T>> subTrees;

    public Tree(T data, Set<Tree<T>> subTrees)
    {
        this.data = data;
        this.subTrees = subTrees;
    }

    public Tree(T data)
    {
        this(data, new HashSet<>());
    }

    public Tree()
    {
        this(null);
    }

    public T getData()
    {
        return data;
    }

    public Set<Tree<T>> getSubTrees()
    {
        return subTrees;
    }

    @Override
    public boolean equals(Object o)
    {
        if (this == o)
            return true;
        if (o == null || getClass() != o.getClass())
            return false;
        Tree<?> tree = (Tree<?>) o;
        return Objects.equals(data, tree.data) &&
                Objects.equals(subTrees, tree.subTrees);
    }

    @Override
    public int hashCode()
    {
        return Objects.hash(data, subTrees);
    }

    @Override
    public String toString()
    {
        return "Tree{"  
                "data="   data  
                ", subTrees="   subTrees  
                '}';
    }

    public void sendCommandAll()
    {
        if (data != null)
            System.out.println("["   Thread.currentThread().getName()   "] sending command to "   data);
        try
        {
            Thread.sleep(5000);
        } catch (InterruptedException e)
        {
            e.printStackTrace();
        }
        if (data != null)
            System.out.println("["   Thread.currentThread().getName()   "] tree with data "   data   " got "   true);
        subTrees.parallelStream()
//              .map(Tree::sendCommandAll)
                .forEach(Tree::sendCommandAll);
//              .reduce(true, (aBoolean, aBoolean2) -> aBoolean && aBoolean2);
    }
}

（我使用forEach或都沒有關系reduce）。

Main.java：

package il.co.roy;

import java.util.Set;
import java.util.concurrent.Executor;
import java.util.concurrent.Executors;
import java.util.stream.Collectors;
import java.util.stream.IntStream;

public class Main
{
    public static void main(String... args)
    {
        System.out.println("Processors: "   Runtime.getRuntime().availableProcessors());


        final Tree<Integer> root = new Tree<>(null,
                Set.of(new Tree<>(1,
                        IntStream.range(2, 7)
                                        .boxed()
                                        .map(Tree::new)
                                        .collect(Collectors.toSet()))));

        root.sendCommandAll();

//      IntStream.generate(() -> 1)
//              .parallel()
//              .forEach(i ->
//              {
//                  System.out.println(Thread.currentThread().getName());
//                  try
//                  {
//                      Thread.sleep(5000);
//                  } catch (InterruptedException e)
//                  {
//                      e.printStackTrace();
//                  }
//              });
    }
}

在該main方法中，我創建了一個具有以下結構的樹：\

root (data is `null`)
  |- 1
     |- 2
     |- 3
     |- 4
     |- 5
     |- 6

sendCommandAll函式處理每個子樹（并行）只有當它的父級完成處理時。但結果如下：

Processors: 8
[main] sending command to 1
[main] tree with data 1 got true
[main] sending command to 6
[ForkJoinPool.commonPool-worker-2] sending command to 5
[main] tree with data 6 got true
[ForkJoinPool.commonPool-worker-2] tree with data 5 got true
[ForkJoinPool.commonPool-worker-2] sending command to 4
[ForkJoinPool.commonPool-worker-2] tree with data 4 got true
[ForkJoinPool.commonPool-worker-2] sending command to 3
[ForkJoinPool.commonPool-worker-2] tree with data 3 got true
[ForkJoinPool.commonPool-worker-2] sending command to 2
[ForkJoinPool.commonPool-worker-2] tree with data 2 got true

(For the record, when I execute the commented code in Main.java, the JVM uses all 7 ( 1) threads available commonPool)

How can I improve my code?

uj5u.com熱心網友回復：

如本答案后半部分所述，處理HashMaps 或HashSets時的執行緒利用率取決于支持陣列中元素的分布，后者取決于哈希碼。特別是在元素數量較少的情況下，與（默認）容量相比，這可能會導致糟糕的作業拆分。

一個簡單的解決方法是使用new ArrayList<>(subTrees).parallelStream()代替subTrees.parallelStream().

但請注意，您的方法sleep在處理子節點之前執行當前節點的實際作業（在使用 a 模擬的示例中），這也降低了潛在的并行度。

您可以使用

public void sendCommandAll() {
    if(subTrees.isEmpty()) {
        actualSendCommand();
        return;
    }
    List<Tree<T>> tmp = new ArrayList<>(subTrees.size()   1);
    tmp.addAll(subTrees);
    tmp.add(this);
    tmp.parallelStream().forEach(t -> {
        if(t != this) t.sendCommandAll(); else t.actualSendCommand();
    });
}

private void actualSendCommand() {
    if (data != null)
        System.out.println("["   Thread.currentThread().getName()
                           "] sending command to "   data);
    try {
        Thread.sleep(5000);
    } catch (InterruptedException e) {
        e.printStackTrace();
    }
    if (data != null)
        System.out.println("["   Thread.currentThread().getName()
                           "] tree with data "   data   " got "   true);
}

這允許在處理子節點的同時處理當前節點。

標籤：java multithreading java-8 java-stream java-17

上一篇：time.sleep不適用于多執行緒

下一篇：如何同步慢計算并快取它？