Java并发编程之深入理解线程池原理及实现

Java线程池在实际的应用开发中十分广泛。虽然Java1.5之后在JUC包中提供了内置线程池可以拿来就用，但是这之前仍有许多老的应用和系统是需要程序员自己开发的。因此，基于线程池的需求背景、技术要求了解线程池原理和实现，一方面可以更为深刻理解Java多线程开发，有助于解决业务系统中因为线程问题所产生的bug；另一方面也有利于对遗留代码的重构。

如果需要先行了解Java并发编程的基础知识，可参考以下随笔：

1. Java并发编程之线程创建和启动(Thread、Runnable、Callable和Future)

2. Java并发编程之线程生命周期、守护线程、优先级、关闭和join、sleep、yield、interrupt

3. Java并发编程之线程安全、线程通信

4. Java并发编程之ThreadGroup

线程池原理

所谓的线程池，跟JDBC连接池、代理池等一样，属于一种“池”的设计模式。在设计好的数据结构中存放一定数量的线程，并且根据任务需要自动调整线程数量的多少，直到峰值。具体说来，线程池需要满足若干条件：

1. 任务队列：用于缓存提交的任务

2. QueueSize：任务队列存放的Runnable任务实例的数量，需要有限制值防止内存溢出。

3. 线程数量管理：创建线程时初始的数量init；线程池自动扩充时最大的线程数量max；空闲时的活跃线程或核心线程数量core。三者满足init<=core<=max

4. 工作线程队列：用于存储工作线程，并统计工作线程的数量。

5. 任务拒绝策略：如果线程数量达到上限且任务队列已满，需要有拒绝策略通知任务提交者，这个在工程实践中非常重要。

6. 线程工厂：用于个性化定制线程，如设置守护线程、线程名称等。

7. Keepedalive时间：线程各个重要参数自动维护的时间间隔。

线程池原理图如下：

线程池实现

“模块设计，类图先行”。明确了线程池需要实现的功能之后，就可以画出线程池的草图了，核心接口及实现类如下：

ThreadPool接口

ThreadPool接口主要定义一个线程池的基本属性，如任务提交、初始容量、最大容量、核心容量等。实现代码如下：

public interface ThreadPool {
    
    //submit tasks to thread pool
    void execute(Runnable runnable);
    //close pool
    void shutdown();
    //get the initial size of pool
    int getInitSize();
    //get the max size of pool
    int getMaxSize();
    //get the core size of pool
    int getCoreSize();
    //get the cache tasks queue of pool
    int getQueueSize();
    //get the active thread volume of pool
    int getActiveCount();
    //check if pool has been shutdown
    boolean isShutdown();
}

RunnableQueue接口

这个接口的作用与BlockingQueue接口一样，用于存储提交的Runnable实现类任务。

public interface RunnableQueue {
    //提交任务到队列
    void offer(Runnable runnable);
    //从队列中获取任务
    Runnable take() throws InterruptedException;
    //返回队列中任务数
    int size();
}

ThreadFactory接口

定义了个性化创建线程的工厂方法

@FunctionalInterface
public interface ThreadFactory {
    
    Thread createThread(Runnable runnable);

}

DenyPolicy接口

定义了线程池的拒绝策略，即当任务队列达到上限时，采取何种措施拒绝。注意接口内定义了内部类作为外围接口的实现类（该类自动为public和static，像这种嵌套类的实现，可查询《Java编程思想》）。

@FunctionalInterface
public interface DenyPolicy {
    
    void reject(Runnable runnable, ThreadPool threadPool);
    //定义嵌套类作为拒绝策略的实现类
    //1.拒绝并丢弃任务
    class DiscardDenyPolicy implements DenyPolicy{

        @Override
        public void reject(Runnable runnable, ThreadPool threadPool) {
            
        }    
    }
    
    //2.拒绝并抛出自定义异常
    class AbortDenyPolicy implements DenyPolicy{

        @Override
        public void reject(Runnable runnable, ThreadPool threadPool) {
            throw new RunnableDenyException("The runnable " + runnable + " will abort.");
        }        
    }
    
    //3.拒绝, 使用提交者所在线程来完成线程任务.
    class RunnerDenyPolicy implements DenyPolicy{

        @Override
        public void reject(Runnable runnable, ThreadPool threadPool) {

            if(!threadPool.isShutdown()) {
                runnable.run();
            }
        }    
    }
}

其实实现了自定义异常类RunnableDenyException：

public class RunnableDenyException extends RuntimeException {
    
    private static final long serialVersionUID = 112311231231412L;

    public RunnableDenyException(String message) {
        super(message);
    }
}

InternalTask实现类

Runnable的实现类，会使用到RunnableQueue，它的作用其实是封装了一个任务实例，把Runnable任务的run方法封装到自己的Run方法实现中，并且提供了一个stop方法，用于在线程池销毁或数量维护时停止当前线程。

public class InternalTask implements Runnable {
    //组合一个RunnableQueue的引用
    private final RunnableQueue runnableQueue;
    //使用volatile关键字修饰开关变量
    private volatile boolean running = true;
    
    public InternalTask(RunnableQueue runnableQueue) {
        this.runnableQueue = runnableQueue;
    }
    @Override
    public void run() {
        // if current task match "both running and isInterrupt" are true
        // continue to take runnable from queue and run
        while(running && !Thread.currentThread().isInterrupted()) {
            try {
                Runnable task = runnableQueue.take();
                task.run();
            } catch (Exception e) {
                running = false;
                break;
            }
        }

    }
    //停止线程的开关方法
    public void stop() {
        this.running = false;
    }
}

到这里，一个基本线程池的骨架就搭好了，接下来主要是实现各接口，实现具体的方法。

1. 队列的实现类LinkedRunnableQueue

public class LinkedRunnableQueue implements RunnableQueue {
    //设置队列上限
    private final int limit;
    //设置拒绝策略的引用
    private final DenyPolicy denyPolicy;
    //使用LinkedList作为队列的具体实现类
    private final LinkedList<Runnable> runnableList = new LinkedList<>();
    //设置线程池的引用
    private final ThreadPool threadPool;
    //构造方法时赋初始值
    public LinkedRunnableQueue(int limit, DenyPolicy denyPolicy, ThreadPool threadPool) {
        this.limit = limit;
        this.denyPolicy = denyPolicy;
        this.threadPool = threadPool;
    }

    @Override
    public void offer(Runnable runnable) {
        //使用同步锁, 确保入队的线程安全
        synchronized (runnableList) {
            //当达到队列上限, 调用拒绝策略;否则加入队尾, 并唤醒阻塞中的线程.
            if(runnableList.size() >= limit) {
                denyPolicy.reject(runnable, threadPool);
            }else {
                runnableList.addLast(runnable);
                runnableList.notifyAll();
            }
        }
    }

    @Override
    public Runnable take() throws InterruptedException {
        synchronized (runnableList) {
            
            while(runnableList.isEmpty()) {
                try {
                    //如果队列中没有可执行任务, 线程挂起, 进入runnableList关联的monitor waitset中等待被唤醒
                    runnableList.wait();
                } catch (InterruptedException e) {
                    //如果被中断, 需要抛出异常
                    throw e;
                }
            }
            return runnableList.removeFirst();
        }
    }

    @Override
    public int size() {
        synchronized (runnableList) {
            //返回队列中的任务数量
            return runnableList.size();
        }
    }
}

2. 线程工厂的实现

public class DefaultThreadFactory implements ThreadFactory {
    //定义原子类的Integer作为线程组的计数
    private static final AtomicInteger GROUP_COUNTER = new AtomicInteger(1);
    //定义线程组对象
    private static final ThreadGroup group = new ThreadGroup("MyThreadPool-"+ GROUP_COUNTER.getAndDecrement());
    //定义生产的线程计数
    private static final AtomicInteger COUNTER = new AtomicInteger(0);

    @Override
    public Thread createThread(Runnable runnable) {
        return new Thread(group, runnable, "thread-pool-" + COUNTER.getAndDecrement());
    }
}

3. 线程池的实现

线程池的实现相对比较复杂， 运用了多种设计模式的思想，核心的要点包括：

1. 使用私有内部类的方式来复用Thread类，防止向外暴露Thread类的方法；

2. 核心组成部分主要是LinkedList实现的任务队列和ArrayDeque实现的工作线程队列，构成了主要的存储主体。

3. 核心的扩容机制需要RunnableQueue + InternalTask + ThreadFactory的结合， 简单说来就是通过判定任务数是否达到阈值，然后增加工作线程的数量。

public class BasicThreadPool implements ThreadPool {
    //为了不暴露Thread类的方法, 使用私有内部类WorkThread来继承Thread类
    private WorkThread workThread;
    //线程池的基本属性
    private final int initSize;
    private final int maxSize;
    private final int coreSize;
    private int activeCount;
    //线程工厂引用
    private final ThreadFactory threadFactory;
    //队列引用
    private final RunnableQueue runnableQueue;
    //线程池销毁标识
    private volatile boolean isShutdown = false;
    //工作线程的队列, 使用ArrayDeque实现
    private final Queue<ThreadTask> threadQueue = new ArrayDeque<>();
    //定义了一个默认的拒绝策略
    private final static DenyPolicy DEFAULT_DENY_POLICY = new DenyPolicy.DiscardDenyPolicy();
    //定义了一个默认的工厂对象
    private final static ThreadFactory DEFAULT_THREAD_FACTORY = new DefaultThreadFactory();
    
    private final long keepAliveTime;
    private final TimeUnit timeUnit;
    //默认的构造器, 只需要传入初始容量, 最大容量, 核心容量和队列上限
    public BasicThreadPool(int initSize, int maxSize, int coreSize, int queueSize) {
        this(initSize, maxSize, coreSize, queueSize, DEFAULT_THREAD_FACTORY, 
                DEFAULT_DENY_POLICY,10,TimeUnit.SECONDS);
    }
    //完整构造器
    public BasicThreadPool(int initSize, int maxSize, int coreSize, int queueSize, ThreadFactory threadFactory,
            DenyPolicy denyPolicy,long keepAliveTime, TimeUnit timeUnit) {
        this.workThread = new WorkThread();
        this.initSize = initSize;
        this.maxSize = maxSize;
        this.coreSize = coreSize;
        this.threadFactory = threadFactory;
        this.runnableQueue = new LinkedRunnableQueue(queueSize, denyPolicy, this);
        this.keepAliveTime = keepAliveTime;
        this.timeUnit = timeUnit;
        this.init();
    }
    //线程池的初始化方法, 在构造器中被调用, 用于启动工作线程
    private void init() {
        workThread.start();
        for(int i = 0; i < initSize; i++) {
            newThread();
        }
    }
    //封装了工作线程的启动方法: 
    //1. 使用InternalTask封装RunnableQueue对象
    //2. 通过工厂方法制造工作线程并启动
    //3. 工作线程入队, 工作线程队列计数器+1
    private void newThread() {
        InternalTask internalTask = new InternalTask(runnableQueue);
        Thread thread = this.threadFactory.createThread(internalTask);
        ThreadTask threadTask = new ThreadTask(thread, internalTask);
        threadQueue.offer(threadTask);
        this.activeCount++;
        thread.start();
    }
    //工作线程出队的方法
    private void removeThread() {
        ThreadTask threadTask = threadQueue.remove();
        threadTask.internalTask.stop();
        this.activeCount--;
    }
    //核心:通过内部类继承Thread方法, 设计了自动扩容的机制.
    //为了防止过快增加到Max容量, 使用continue来退出循环
    private class WorkThread extends Thread{
        @Override
        public void run() {
            while(!isShutdown && !isInterrupted()) {
                try {
                    timeUnit.sleep(keepAliveTime);
                } catch (InterruptedException e) {
                    isShutdown = true;
                    break;
                }
                synchronized (this) {
                    if(isShutdown) {
                        break;
                    }
                    if(runnableQueue.size() > 0 && activeCount < coreSize) {
                        for(int i = initSize; i<coreSize;i++) {
                            newThread();
                        }
                        continue;
                    }
                    if(runnableQueue.size() > 0 && activeCount < maxSize) {
                        for(int i = coreSize; i<maxSize;i++) {
                            newThread();
                        }
                    }
                    if(runnableQueue.size()==0 && activeCount > coreSize) {
                        for(int i = coreSize; i < activeCount; i++) {
                            removeThread();
                        }
                    }
                    
                }
            }
        }
    }

    @Override
    public void execute(Runnable runnable) {
        //如果线程池已经销毁, 将抛出异常
        if(this.isShutdown) {
            throw new IllegalStateException("the thread pool is destoried");
        }
        this.runnableQueue.offer(runnable);    
    }

    @Override
    public void shutdown() {
        synchronized(this) {
            //防止重复销毁
            if(isShutdown) {
                return;
            }
            //重置关闭标识
            isShutdown = true;
            //关闭任务工作线程
            threadQueue.forEach(threadTask -> {
                threadTask.internalTask.stop();
                threadTask.thread.interrupt();
            });
            //关闭线程池的工作线程
            this.workThread.interrupt();
        }
    }

    @Override
    public int getInitSize() {
        if(isShutdown) {
            throw new IllegalStateException("The thread pool is destroy");
        }
        return this.initSize;
    }

    @Override
    public int getMaxSize() {
        if(isShutdown) {
            throw new IllegalStateException("The thread pool is destroy");
        }
        return this.maxSize;
    }

    @Override
    public int getCoreSize() {
        if(isShutdown) {
            throw new IllegalStateException("The thread pool is destroy");
        }
        return this.coreSize;
    }

    @Override
    public int getQueueSize() {
        if(isShutdown) {
            throw new IllegalStateException("The thread pool is destroy");
        }
        return runnableQueue.size();
    }

    @Override
    public int getActiveCount() {
        synchronized(this) {
            return this.activeCount;
        }
    }

    @Override
    public boolean isShutdown() {        
        return this.isShutdown;
    }
}

线程池的测试

编写一个简单的测试类，同时启动20个任务，测试线程池的活动状态：

public class ThreadPoolTest {

    public static void main(String[] args) throws InterruptedException {
        
        final ThreadPool threadPool = new BasicThreadPool(2, 6, 4, 1000);
        
        for(int i = 0; i < 20; i++) {
            threadPool.execute(() -> {
                try {
                    TimeUnit.SECONDS.sleep(10);
                    System.out.println(Thread.currentThread().getName() + "is Running and done");
                } catch (InterruptedException e) {
                    e.printStackTrace();
                }
            });
        }
        while(true) {
            System.out.println("getActiveCount: " + threadPool.getActiveCount());
            System.out.println("getQueueSize: " + threadPool.getQueueSize());
            System.out.println("getCoreSize: " + threadPool.getCoreSize());
            System.out.println("getMaxSize: " + threadPool.getMaxSize());
            System.out.println("================================================");
            TimeUnit.SECONDS.sleep(5);            
        }
    }
}

输出结果如下

thread-pool--1is Running and done
thread-pool-0is Running and done
getActiveCount: 4
getQueueSize: 14
getCoreSize: 4
getMaxSize: 6
================================================
getActiveCount: 4
getQueueSize: 14
getCoreSize: 4
getMaxSize: 6
================================================
thread-pool--3is Running and done
thread-pool--2is Running and done
thread-pool--1is Running and done
thread-pool-0is Running and done
getActiveCount: 6
getQueueSize: 8
getCoreSize: 4
getMaxSize: 6