Forkjoin线程池

　　之前学习了ThreadPoolExecutor线程池的用法，现在学习下Forkjoin线程池的用法。主要也就是ForkJoinPool，该类和ThreadPoolExecutor是兄弟类，都继承自AbstractExecutorService。

1.简介

1.什么是Forkjoin以及适用场景　　

　　虽然目前处理器核心数已经发展到最大，但是按任务并发处理并不能完全充分的利用处理器资源，因为一般的应用程序没有那么多的并发任务。基于这种现象，考虑将一个任务拆分成多个单元，每个单元分别得到执行，最后合并每个单元的结果。

　　Fork/join是Java7提供的一个用于并发执行任务的框架，是一个把多个大任务分成若干个小任务，最终汇总每个小任务结果得到大任务结果的框架。

　　说白了就是充分利用多处理器的优势，让处理器充分利用起来，并发处理任务。

　　如果需要查询CPU数量可以从任务管理器查看，也可以用java代码查看，如下：

System.out.println(Runtime.getRuntime().availableProcessors());

结果：

2.工作窃取算法

　　一个大任务拆分成许多小任务，为了减小线程间的竞争，把这些子任务分别放到不同的队列中，并且每个队列都有单独的线程来执行任务，线程和队列一一对应。但是会出现一种情况：A线程处理完了自己队列的任务，但是B线程的队列还有很多任务要处理。如果A过去处理B队列的任务会访问同一个队列，造成竞争，解决办法就是A从双端队列的尾部拿任务、B从双端队列的头部拿任务。

注意：每个线程都有自己的队列（这点和TjreadPoolExecutor不一样），当自己队列的任务完成以后会从其他线程的队列窃取一个任务执行，这样可以充分利用资源。

优点:利用了线程并行计算，减少了线程间的竞争。

缺点：

(1)如果双端队列只有一个任务会产生竞争

(2)窃取算法消耗了更多的系统资源，如会创建多个线程和多个双端队列。

3.涉及的主要类

1.ForkJoinTask

　　它提供任务执行中fork和join操作的机制。一般只需要继承其子类：

RecursiveTask：用于执行有返回结果的任务，这个泛型需要自己传。

RecursiveAction：用于执行没有返回结果的任务，其实这个的泛型只不过是Void

2.ForkJoinPool

　　ForkJoinTask任务需要在ForkJoinPool中执行。默认的线程池数量等价于CPU数量(我的机子是4)。

3.ForrkjoinWorkerThread

　　ForkJoinPool中执行任务的线程。

2.简单使用

比如一个简单的使用，实现Forkjoin计算0-100000000000的和

ForkJoinTask类：

　　当需要计算的数值大于step的时候继续拆分任务，否则拆分任务。

package forkjoin;

import java.util.concurrent.RecursiveTask;

/**
 * @author: 乔利强
 * @date: 2021/1/7 19:37
 * @description:
 */
public class SumTask extends RecursiveTask<Long> {

    private long start, end;

    private long step = 1000000;

    public SumTask(long start, long end) {
        this.start = start;
        this.end = end;
    }

    @Override
    protected Long compute() {
        long sum = 0;
        if (end - start <= step) {
            for (long i = start; i <= end; i++)
                sum += i;
        } else {
            long mid = start + (end - start) / 2;
            SumTask lt = new SumTask(start, mid);
            SumTask rt = new SumTask(mid + 1, end);
            lt.fork();
            rt.fork();
            long leftsum = lt.join();
            long rightsum = rt.join();
            sum = leftsum + rightsum;
        }
        
        return sum;
    }
}

测试类：测试单线程同步执行和Forkjoin框架执行

package forkjoin;

import java.util.concurrent.ExecutionException;
import java.util.concurrent.ForkJoinPool;
import java.util.concurrent.ForkJoinTask;

/**
 * @author: 乔利强
 * @date: 2021/1/7 19:32
 * @description:
 */
public class SumClient {

    public static void main(String[] args) throws ExecutionException, InterruptedException {
        // 单线程测试
        long startNum = 0;
        long endNum = 100000000000L;
        long sum = 0;
        long startTime = System.currentTimeMillis();
        for (; startNum <= endNum; startNum++) {
            sum += startNum;
        }
        long endTime = System.currentTimeMillis();
        System.out.println("sum: " + sum + "， 用时: " + (endTime - startTime) + "ms");

        // fork-join测试
        startTime = System.currentTimeMillis();
        // 1. 创建线程池
        ForkJoinPool fp = new ForkJoinPool();
        // 2.创建一个任务
        SumTask task = new SumTask(0, endNum);
        // 3. 任务交给线程
        ForkJoinTask<Long> result = fp.submit(task);
        // 4.得到结果
        Long fkResult = result.get();
        endTime = System.currentTimeMillis();
        System.out.println("sum: " + fkResult + "， 用时: " + (endTime - startTime) + "ms");
    }
}

结果：(我在自己的电脑测的时候计算太耗时了就中断了。我在另一个电脑测的时候单线程大概是40多s、forkjoin大概是20多s)

查看ForkJoinPool的默认构造，如下：可以创建的线程池的大小是CPU数量

    public ForkJoinPool() {
        this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
             defaultForkJoinWorkerThreadFactory, null, false);
    }

补充：并行流parallelStream底层也是用了ForkJoin，并行流适合没有线程安全问题、较单纯的数据处理任务

　　parallelStream提供了流的并行处理，它是Stream的另一重要特性，其底层使用Fork/Join框架实现。简单理解就是多线程异步任务的一种实现。

测试：

        List<Integer> numbers = Arrays.asList(1, 2, 3, 4, 5, 6, 7, 8, 9);
        numbers.parallelStream().forEach(num -> {
            try {
                Thread.sleep(1 * 100);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + ">>" + num);
        });

结果： (可以看出打印的顺序是乱序证明确实是异步并行执行；打印的线程名字是ForkJoinPool.commonPool-worker-1到3 也证明是采用的ForkJoinPool)

main>>6

ForkJoinPool.commonPool-worker-1>>8

ForkJoinPool.commonPool-worker-2>>3

ForkJoinPool.commonPool-worker-3>>7

main>>5

ForkJoinPool.commonPool-worker-2>>4

ForkJoinPool.commonPool-worker-1>>9

ForkJoinPool.commonPool-worker-3>>2

main>>1

补充：可以通过修改VM参数-Djava.util.concurrent.ForkJoinPool.common.parallelism=N设置worker的数量，测试如下：

        System.setProperty("java.util.concurrent.ForkJoinPool.common.parallelism", "6");

        List<Integer> numbers = Arrays.asList(1, 2, 3, 4, 5, 6, 7, 8, 9);
        numbers.parallelStream().forEach(num -> {
            try {
                Thread.sleep(1 * 100);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + ">>" + num);
        });

结果：

main>>6

ForkJoinPool.commonPool-worker-2>>3

ForkJoinPool.commonPool-worker-1>>8

ForkJoinPool.commonPool-worker-6>>5

ForkJoinPool.commonPool-worker-3>>7

ForkJoinPool.commonPool-worker-5>>9

ForkJoinPool.commonPool-worker-4>>2

main>>4

ForkJoinPool.commonPool-worker-2>>1

补充：ForkJoin不一定就肯定比单线程处理块，这还要看你单线程串行处理每个任务的耗时情况；ForkJoin本身创建线程以及Fork、Join任务都是需要花费时间的，比如：

package forkjoin;

import java.util.Arrays;
import java.util.List;

public class SumClient2 {

    public static void main(String[] args) {
        // 修改forkjoin默认的线程池大小(可以修改为比处理器数量多，但是没必要，一般和处理器数目相同就可以)
        System.setProperty("java.util.concurrent.ForkJoinPool.common.parallelism", "16");
        // 模拟处理用时
        int handleTime = 2;

        List<Integer> numbers = Arrays.asList(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
        long start = System.currentTimeMillis();
        numbers.parallelStream().forEach(num -> {
            try {
                Thread.sleep(1 * handleTime);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + ">>" + num);
        });
        long end = System.currentTimeMillis();
        System.out.println((end - start) + "ms ===");

        start = System.currentTimeMillis();
        numbers.stream().forEach(num -> {
            try {
                Thread.sleep(1 * handleTime);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + ">>" + num);
        });
        end = System.currentTimeMillis();
        System.out.println((end - start) + "ms ---");
    }
}

结果：可以看出使用ForkJoinPool反而时间多于单线程，当 handleTime 增大的时候使用ForkJoinPool会快于单线程。

main>>11
ForkJoinPool.commonPool-worker-1>>1
ForkJoinPool.commonPool-worker-8>>13
ForkJoinPool.commonPool-worker-9>>6
ForkJoinPool.commonPool-worker-4>>14
ForkJoinPool.commonPool-worker-10>>5
ForkJoinPool.commonPool-worker-6>>2
ForkJoinPool.commonPool-worker-2>>15
ForkJoinPool.commonPool-worker-13>>16
ForkJoinPool.commonPool-worker-11>>3
ForkJoinPool.commonPool-worker-15>>8
ForkJoinPool.commonPool-worker-14>>7
ForkJoinPool.commonPool-worker-3>>12
ForkJoinPool.commonPool-worker-12>>4
ForkJoinPool.commonPool-worker-5>>10
ForkJoinPool.commonPool-worker-7>>9
103ms ===
main>>1
main>>2
main>>3
main>>4
main>>5
main>>6
main>>7
main>>8
main>>9
main>>10
main>>11
main>>12
main>>13
main>>14
main>>15
main>>16
49ms ---

补充：并行流parallelStream底层也是用了ForkJoin，而且全局公用一个Pool，默认大小是CPU数量减去一

验证一：

import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.List;

public class PlainTest {

    public static void main(String[] args) {
        List<String> fruitsList = new ArrayList<String>(Arrays.asList("Apple", "Orange", "1", "3", "4", "5", "7"));
        List<String> fruitsUnmodifiableList = Collections.unmodifiableList(fruitsList);
        System.out.println(fruitsUnmodifiableList.getClass());
        System.out.println(Runtime.getRuntime().availableProcessors());

        fruitsList.parallelStream().forEach(c -> {
            System.out.println(Thread.currentThread().getName() + "\t" + c);
            try {
                Thread.sleep(10 * 100000);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
        });
    }
}

结果：

class java.util.Collections$UnmodifiableRandomAccessList
4
main 4
ForkJoinPool.commonPool-worker-2 7
ForkJoinPool.commonPool-worker-3 Apple
ForkJoinPool.commonPool-worker-1 Orange

查看是用的java.util.concurrent.ForkJoinPool#common，创建过程如下：java.util.concurrent.ForkJoinPool#makeCommonPool

    /**
     * Creates and returns the common pool, respecting user settings
     * specified via system properties.
     */
    private static ForkJoinPool makeCommonPool() {
        int parallelism = -1;
        ForkJoinWorkerThreadFactory factory = null;
        UncaughtExceptionHandler handler = null;
        try {  // ignore exceptions in accessing/parsing properties
            String pp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.parallelism");
            String fp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.threadFactory");
            String hp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
            if (pp != null)
                parallelism = Integer.parseInt(pp);
            if (fp != null)
                factory = ((ForkJoinWorkerThreadFactory)ClassLoader.
                           getSystemClassLoader().loadClass(fp).newInstance());
            if (hp != null)
                handler = ((UncaughtExceptionHandler)ClassLoader.
                           getSystemClassLoader().loadClass(hp).newInstance());
        } catch (Exception ignore) {
        }
        if (factory == null) {
            if (System.getSecurityManager() == null)
                factory = defaultForkJoinWorkerThreadFactory;
            else // use security-managed default
                factory = new InnocuousForkJoinWorkerThreadFactory();
        }
        if (parallelism < 0 && // default 1 less than #cores
            (parallelism = Runtime.getRuntime().availableProcessors() - 1) <= 0)
            parallelism = 1;
        if (parallelism > MAX_CAP)
            parallelism = MAX_CAP;
        return new ForkJoinPool(parallelism, factory, handler, LIFO_QUEUE,
                                "ForkJoinPool.commonPool-worker-");
    }

验证二：

1、默认cpu-1加当前线程

package qz;

import com.google.common.collect.Lists;
import lombok.extern.slf4j.Slf4j;

import java.util.List;
import java.util.stream.Collectors;
import java.util.stream.IntStream;

/**
 * PlainTest
 *
 * @author qlq
 */
@Slf4j
public class PlainTest {
    public static void main(String[] args) throws InterruptedException {
        log.info(Thread.currentThread().getName() + "\tcpu核数: " + Runtime.getRuntime().availableProcessors());

        Thread thread = new Thread(() -> {
            // 数字流
            List<Integer> collect = IntStream.rangeClosed(1, 100).boxed().collect(Collectors.toList());
            /**
             * cpu 核数 - 8
             * fork-join用的common 线程 会开 (cpu核数-1 = 7), 当前线程也会参与处理
             */

            // 10个线程打印, fork-join 线程池
            List<List<Integer>> partition = Lists.partition(collect, 10);
            partition.parallelStream().forEach(list -> {
                log.info(Thread.currentThread().getName() + "\t" + list);
            });

            log.info(Thread.currentThread().getName() + "\t end");
        });
        thread.setName("cus_thread");

        thread.start();

        Thread.sleep(2 * 1000);
        log.info(Thread.currentThread().getName() + "\t end");
    }
}

结果:

2024-07-29 16:43:25 [main] [qz.PlainTest]-[INFO] main    cpu核数: 8
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-3] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-3    [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-7] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-7    [31, 32, 33, 34, 35, 36, 37, 38, 39, 40]
2024-07-29 16:43:25 [cus_thread] [qz.PlainTest]-[INFO] cus_thread    [61, 62, 63, 64, 65, 66, 67, 68, 69, 70]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-4] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-4    [51, 52, 53, 54, 55, 56, 57, 58, 59, 60]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-2] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-2    [11, 12, 13, 14, 15, 16, 17, 18, 19, 20]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-6] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-6    [81, 82, 83, 84, 85, 86, 87, 88, 89, 90]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-1] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-1    [21, 22, 23, 24, 25, 26, 27, 28, 29, 30]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-5] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-5    [41, 42, 43, 44, 45, 46, 47, 48, 49, 50]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-7] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-7    [91, 92, 93, 94, 95, 96, 97, 98, 99, 100]
2024-07-29 16:43:25 [ForkJoinPool.commonPool-worker-3] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-3    [71, 72, 73, 74, 75, 76, 77, 78, 79, 80]
2024-07-29 16:43:25 [cus_thread] [qz.PlainTest]-[INFO] cus_thread     end
2024-07-29 16:43:27 [main] [qz.PlainTest]-[INFO] main     end

2、修改后变成固定数 + 当前线程

package qz;

import com.google.common.collect.Lists;
import lombok.extern.slf4j.Slf4j;

import java.util.List;
import java.util.stream.Collectors;
import java.util.stream.IntStream;

/**
 * PlainTest
 *
 * @author qlq
 */
@Slf4j
public class PlainTest {
    public static void main(String[] args) throws InterruptedException {
        log.info(Thread.currentThread().getName() + "\tcpu核数: " + Runtime.getRuntime().availableProcessors());

        // 修改默认的线程池大小，实际会变成4， 当前线程也会参与处理
        System.setProperty("java.util.concurrent.ForkJoinPool.common.parallelism", "4");

        Thread thread = new Thread(() -> {
            // 数字流
            List<Integer> collect = IntStream.rangeClosed(1, 100).boxed().collect(Collectors.toList());
            /**
             * cpu 核数 - 8
             * fork-join用的common 线程 会开 (cpu核数-1 = 7), 当前线程也会参与处理
             */

            // 10个线程打印, fork-join 线程池
            List<List<Integer>> partition = Lists.partition(collect, 10);
            partition.parallelStream().forEach(list -> {
                try {
                    Thread.sleep(1 * 100);
                } catch (InterruptedException e) {
                    // ignore
                }
                log.info(Thread.currentThread().getName() + "\t" + list);
            });

            log.info(Thread.currentThread().getName() + "\t end");
        });
        thread.setName("cus_thread");
        thread.start();

        Thread.sleep(2 * 1000);
        log.info(Thread.currentThread().getName() + "\t end");
    }
}
// 结果
2024-07-29 16:48:19 [main] [qz.PlainTest]-[INFO] main    cpu核数: 8
2024-07-29 16:48:19 [ForkJoinPool.commonPool-worker-2] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-2    [11, 12, 13, 14, 15, 16, 17, 18, 19, 20]
2024-07-29 16:48:19 [ForkJoinPool.commonPool-worker-3] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-3    [41, 42, 43, 44, 45, 46, 47, 48, 49, 50]
2024-07-29 16:48:19 [cus_thread] [qz.PlainTest]-[INFO] cus_thread    [61, 62, 63, 64, 65, 66, 67, 68, 69, 70]
2024-07-29 16:48:19 [ForkJoinPool.commonPool-worker-0] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-0    [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
2024-07-29 16:48:19 [ForkJoinPool.commonPool-worker-1] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-1    [21, 22, 23, 24, 25, 26, 27, 28, 29, 30]
2024-07-29 16:48:20 [ForkJoinPool.commonPool-worker-2] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-2    [31, 32, 33, 34, 35, 36, 37, 38, 39, 40]
2024-07-29 16:48:20 [ForkJoinPool.commonPool-worker-3] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-3    [81, 82, 83, 84, 85, 86, 87, 88, 89, 90]
2024-07-29 16:48:20 [cus_thread] [qz.PlainTest]-[INFO] cus_thread    [51, 52, 53, 54, 55, 56, 57, 58, 59, 60]
2024-07-29 16:48:20 [ForkJoinPool.commonPool-worker-1] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-1    [91, 92, 93, 94, 95, 96, 97, 98, 99, 100]
2024-07-29 16:48:20 [ForkJoinPool.commonPool-worker-0] [qz.PlainTest]-[INFO] ForkJoinPool.commonPool-worker-0    [71, 72, 73, 74, 75, 76, 77, 78, 79, 80]
2024-07-29 16:48:20 [cus_thread] [qz.PlainTest]-[INFO] cus_thread     end
2024-07-29 16:48:21 [main] [qz.PlainTest]-[INFO] main     end

补充：还有另一种用法，执行callable任务的线程池

package com.zd.bx;

import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ForkJoinPool;
import java.util.concurrent.Future;

public class PLainTest {

    public static void main(String[] args) throws ExecutionException, InterruptedException {
        ForkJoinPool fp = new ForkJoinPool(2);
        List<Callable<Integer>> callableArrayList = new ArrayList<>(2);
        callableArrayList.add(() -> {
            Thread.sleep(1 * 1000);
            System.out.println(Thread.currentThread().getName() + " ==1");
            return 1;
        });
        callableArrayList.add(() -> {
            Thread.sleep(1 * 1000);
            System.out.println(Thread.currentThread().getName() + " ==3");
            return 3;
        });
        // 内部创建ForkJoinTask， 然后执行所有的任务，然后调用forkJoinTask.get() 阻塞获取其结果
        List<Future<Integer>> futures = fp.invokeAll(callableArrayList);
        for (Future<Integer> integerFuture : futures) {
            System.out.println(integerFuture.get());
        }
        System.out.println(Thread.currentThread().getName() + " ==6");
        fp.shutdown();
    }
}

结果：

ForkJoinPool-1-worker-0 ==1
ForkJoinPool-1-worker-1 ==3
1
3
main ==6

posted @ 2021-01-07 23:21 QiaoZhi 阅读(4680) 评论(0) 编辑收藏举报

刷新页面返回顶部

登录后才能查看或发表评论，立即登录或者逛逛博客园首页

阅读排行：
· 分享4款.NET开源、免费、实用的商城系统
· 全程不用写代码，我用AI程序员写了一个飞机大战
· Obsidian + DeepSeek：免费 AI 助力你的知识管理，让你的笔记飞起来！
· MongoDB 8.0这个新功能碉堡了，比商业数据库还牛
· 白话解读 Dapr 1.15：你的「微服务管家」又秀新绝活了

历史上的今天：
2020-01-07 Markdown使用介绍
2020-01-07 基于github与hexo搭建个人博客
2019-01-07 原子类型的使用&Unsafe&CAS
2018-01-07 Maven 集成Tomcat7插件自动部署
2018-01-07 【Tomcat】tomcat热部署和热加载（转载）
2018-01-07 maven 编译出错 Failed to execute goal org.apache.maven.plugins:maven-clean-plugin:2.5:clean
2018-01-07 eclipse maven 打war包的几种方式

公告

昵称： QiaoZhi
园龄： 7年7个月
粉丝： 1000
关注： 9

+加关注

2025年3月

日

一

二

三

四

五

六

Qiao_Zhi

有远大抱负的人不可忽略眼前的工作!!!

Forkjoin线程池

1.简介

1.什么是Forkjoin以及适用场景

2.工作窃取算法

3.涉及的主要类

1.ForkJoinTask

2.ForkJoinPool

3.ForrkjoinWorkerThread

2.简单使用

公告

搜索

积分与排名

随笔分类 (1546)

相册 (4)

阅读排行榜

评论排行榜

推荐排行榜

最新评论