Thread Pool

Thread Pool

Thread Pool Overview

What is the thread pool
Why use a thread pool
Thread pool advantage
- First: reduce resource consumption. By reusing the thread has been created to reduce thread creation and destruction caused by consumption.
- Second: improving the response speed. When the mission arrives, the task may not need to wait until the thread creation can be implemented immediately.
- Third: to improve the manageability of threads. A thread is a scarce resource, if the unlimited creation, not only consumes system resources, but also reduce the stability of the system, using a thread pool can be unified distribution, tuning and monitoring. But to be a reasonable use of the thread pool must be well aware of its principle.

Create a thread pool thread and submit the task

Source parsing thread pool

Parameters know

corePoolSize: The basic size of the thread pool, when submitting a job to the thread pool, thread pool will create a thread to perform the task, even if the other free basic thread can perform new tasks will be to create a thread, until the number of tasks need to be performed is greater than the thread created when the pool is no longer a basic size. If you call a method prestartAllCoreThreads thread pool, thread pool is created in advance and start all basic threads.
runnableTaskQueue: task column that stores the task of blocking queue waiting to be executed. You can select the following blocking queue.

ArrayBlockingQueue: an array-based configuration bounded blocking queue This queue FIFO (first in first out) principle sort elements.
LinkedBlockingQueue: blocking queue based on a list structure, according to this queue FIFO (First In First Out) collating element, usually higher than a certain ArrayBlockingQueue. Static factory method Executors.newFixedThreadPool () uses this queue.
SynchronousQueue: a blocking queue element is not stored. Each insert operation must wait until another thread calls the removal operation, or insert operation has been in a blocked state, throughput is usually higher than LinkedBlockingQueue, static factory method Executors.newCachedThreadPool use this queue.
PriorityBlockingQueue: a prioritized infinitely blocking queue.

maximumPoolSize: the maximum size of the thread pool, thread pool allows you to create the maximum number of threads. If the queue is full, and the number of threads that have been created less than the maximum number of threads, the thread pool will re-create a new thread to perform the task. It is worth noting that if you use unbounded task queue this parameter to no effect.
ThreadFactory: used to set the factory to create threads, and help can be set very meaningful to each created by a thread factory out of the thread name, Debug and positioning issues.
RejectedExecutionHandler (saturation policy): When the queues and thread pools are full, indicating that the thread pool is saturated, it must adopt a strategy to deal with the new task submission. By default, this strategy is AbortPolicy, it said it could not throw an exception when dealing with new tasks.

CallerRunsPolicy: where only the caller thread to run the task.
DiscardOldestPolicy: discard queue recent task and execute the current task.
DiscardPolicy: no treatment, discarded.
Of course, you may be required to implement a custom policy RejectedExecutionHandler interfaces depending on the application scenario. Such as logging or persistence task can not handle.

keepAliveTime: thread activity holding time, after the idle worker thread pool, to keep alive the time. So if a lot of tasks, and each task execution time is relatively short, it can turn up this time, improve the utilization of the thread.
TimeUnit: retention time thread activity units, optionally units of one day (DAYS), h (HOURS), min (MINUTES), ms (MILLISECONDS), microsecond (MICROSECONDS, thousandth ms) and ns ( nANOSECONDS, thousandth microseconds).
Other properties in the class


    // 线程池的控制状态:用来表示线程池的运行状态（整型的高3位）和运行的worker数量（低29位）
    private final AtomicInteger ctl = new AtomicInteger(ctlOf(RUNNING, 0));
    // 29位的偏移量
    private static final int COUNT_BITS = Integer.SIZE - 3;
    // 最大容量（2^29 - 1）
    private static final int CAPACITY   = (1 << COUNT_BITS) - 1;

    // runState is stored in the high-order bits
    // 线程运行状态，总共有5个状态，需要3位来表示（所以偏移量的29 = 32 - 3）
   /**
    * RUNNING    :    接受新任务并且处理已经进入阻塞队列的任务
    * SHUTDOWN    ：    不接受新任务，但是处理已经进入阻塞队列的任务
    * STOP        :    不接受新任务，不处理已经进入阻塞队列的任务并且中断正在运行的任务
    * TIDYING    :    所有的任务都已经终止，workerCount为0， 线程转化为TIDYING状态并且调用terminated钩子函数
    * TERMINATED:    terminated钩子函数已经运行完成
    **/
    private static final int RUNNING    = -1 << COUNT_BITS;
    private static final int SHUTDOWN   =  0 << COUNT_BITS;
    private static final int STOP       =  1 << COUNT_BITS;
    private static final int TIDYING    =  2 << COUNT_BITS;
    private static final int TERMINATED =  3 << COUNT_BITS;
    // 阻塞队列
    private final BlockingQueue<Runnable> workQueue;
    // 可重入锁
    private final ReentrantLock mainLock = new ReentrantLock();
    // 存放工作线程集合
    private final HashSet<Worker> workers = new HashSet<Worker>();
    // 终止条件
    private final Condition termination = mainLock.newCondition();
    // 最大线程池容量
    private int largestPoolSize;
    // 已完成任务数量
    private long completedTaskCount;
    // 线程工厂
    private volatile ThreadFactory threadFactory;
    // 拒绝执行处理器
    private volatile RejectedExecutionHandler handler;
    // 线程等待运行时间
    private volatile long keepAliveTime;
    // 是否运行核心线程超时
    private volatile boolean allowCoreThreadTimeOut;
    // 核心池的大小
    private volatile int corePoolSize;
    // 最大线程池大小
    private volatile int maximumPoolSize;
    // 默认拒绝执行处理器
    private static final RejectedExecutionHandler defaultHandler =
        new AbortPolicy();

Construction method

    public ThreadPoolExecutor(int corePoolSize,
                              int maximumPoolSize,
                              long keepAliveTime,
                              TimeUnit unit,
                              BlockingQueue<Runnable> workQueue,
                              ThreadFactory threadFactory,
                              RejectedExecutionHandler handler) {
        if (corePoolSize < 0 ||                                                // 核心大小不能小于0
            maximumPoolSize <= 0 ||                                            // 线程池的初始最大容量不能小于0
            maximumPoolSize < corePoolSize ||                                // 初始最大容量不能小于核心大小
            keepAliveTime < 0)                                                // keepAliveTime不能小于0
            throw new IllegalArgumentException();                                
        if (workQueue == null || threadFactory == null || handler == null)
            throw new NullPointerException();
        // 初始化相应的域
        this.corePoolSize = corePoolSize;
        this.maximumPoolSize = maximumPoolSize;
        this.workQueue = workQueue;
        this.keepAliveTime = unit.toNanos(keepAliveTime);
        this.threadFactory = threadFactory;
        this.handler = handler;
    }

Submit jobs

/*
* 进行下面三步
*
* 1. 如果运行的线程小于corePoolSize,则尝试使用用户定义的Runnalbe对象创建一个新的线程
*     调用addWorker函数会原子性的检查runState和workCount，通过返回false来防止在不应
*     该添加线程时添加了线程
* 2. 如果一个任务能够成功入队列，在添加一个线城时仍需要进行双重检查（因为在前一次检查后
*     该线程死亡了），或者当进入到此方法时，线程池已经shutdown了，所以需要再次检查状态，
*    若有必要，当停止时还需要回滚入队列操作，或者当线程池没有线程时需要创建一个新线程
* 3. 如果无法入队列，那么需要增加一个新线程，如果此操作失败，那么就意味着线程池已经shut
*     down或者已经饱和了，所以拒绝任务
*/
public void execute(Runnable command) {
    if (command == null)
        throw new NullPointerException();
    // 获取线程池控制状态
    int c = ctl.get();
    if (workerCountOf(c) < corePoolSize) { // worker数量小于corePoolSize
        if (addWorker(command, true)) // 添加worker
            // 成功则返回
            return;
        // 不成功则再次获取线程池控制状态
        c = ctl.get();
    }
    // 线程池处于RUNNING状态，将用户自定义的Runnable对象添加进workQueue队列
    if (isRunning(c) && workQueue.offer(command)) { 
        // 再次检查，获取线程池控制状态
        int recheck = ctl.get();
        // 线程池不处于RUNNING状态，将自定义任务从workQueue队列中移除
        if (! isRunning(recheck) && remove(command)) 
            // 拒绝执行命令
            reject(command);
        else if (workerCountOf(recheck) == 0) // worker数量等于0
            // 添加worker
            addWorker(null, false);
    }
    else if (!addWorker(command, false)) // 添加worker失败
        // 拒绝执行命令
        reject(command);
}

addWorker

Increasing atomic workerCount.
The user will be given the task of packaging a worker, worker and add this collection workers.
Start worker corresponding thread, and the thread starts to run worker's run method.
Create a rollback action worker, the worker will soon be removed from the set of workers, and atomic reduce workerCount.

private boolean addWorker(Runnable firstTask, boolean core) {
    retry:
    for (;;) { // 外层无限循环
        // 获取线程池控制状态
        int c = ctl.get();
        // 获取状态
        int rs = runStateOf(c);

        // Check if queue empty only if necessary.
        if (rs >= SHUTDOWN &&            // 状态大于等于SHUTDOWN，初始的ctl为RUNNING，小于SHUTDOWN
            ! (rs == SHUTDOWN &&        // 状态为SHUTDOWN
               firstTask == null &&        // 第一个任务为null
               ! workQueue.isEmpty()))     // worker队列不为空
            // 返回
            return false;

        for (;;) {
            // worker数量
            int wc = workerCountOf(c);
            if (wc >= CAPACITY ||                                // worker数量大于等于最大容量
                wc >= (core ? corePoolSize : maximumPoolSize))    // worker数量大于等于核心线程池大小或者最大线程池大小
                return false;
            if (compareAndIncrementWorkerCount(c))                 // 比较并增加worker的数量
                // 跳出外层循环
                break retry;
            // 获取线程池控制状态
            c = ctl.get();  // Re-read ctl
            if (runStateOf(c) != rs) // 此次的状态与上次获取的状态不相同
                // 跳过剩余部分，继续循环
                continue retry;
            // else CAS failed due to workerCount change; retry inner loop
        }
    }

    // worker开始标识
    boolean workerStarted = false;
    // worker被添加标识
    boolean workerAdded = false;
    // 
    Worker w = null;
    try {
        // 初始化worker
        w = new Worker(firstTask);
        // 获取worker对应的线程
        final Thread t = w.thread;
        if (t != null) { // 线程不为null
            // 线程池锁
            final ReentrantLock mainLock = this.mainLock;
            // 获取锁
            mainLock.lock();
            try {
                // Recheck while holding lock.
                // Back out on ThreadFactory failure or if
                // shut down before lock acquired.
                // 线程池的运行状态
                int rs = runStateOf(ctl.get());

                if (rs < SHUTDOWN ||                                    // 小于SHUTDOWN
                    (rs == SHUTDOWN && firstTask == null)) {            // 等于SHUTDOWN并且firstTask为null
                    if (t.isAlive()) // precheck that t is startable    // 线程刚添加进来，还未启动就存活
                        // 抛出线程状态异常
                        throw new IllegalThreadStateException();
                    // 将worker添加到worker集合
                    workers.add(w);
                    // 获取worker集合的大小
                    int s = workers.size();
                    if (s > largestPoolSize) // 队列大小大于largestPoolSize
                        // 重新设置largestPoolSize
                        largestPoolSize = s;
                    // 设置worker已被添加标识
                    workerAdded = true;
                }
            } finally {
                // 释放锁
                mainLock.unlock();
            }
            if (workerAdded) { // worker被添加
                // 开始执行worker的run方法
                t.start();
                // 设置worker已开始标识
                workerStarted = true;
            }
        }
    } finally {
        if (! workerStarted) // worker没有开始
            // 添加worker失败
            addWorkerFailed(w);
    }
    return workerStarted;
}

Mission

runWorker function will actually perform a given task (i.e., the user calls the run method rewritable), and after the completion of a given task, the task will continue to take from blocking the queue until the queue is empty obstruction (i.e., completed tasks). While performing a given task, it calls the hook function, using the hook function can do some user-defined logic. In runWorker will call the hook function getTask function and processWorkerExit

final void runWorker(Worker w) {
    // 获取当前线程
    Thread wt = Thread.currentThread();
    // 获取w的firstTask
    Runnable task = w.firstTask;
    // 设置w的firstTask为null
    w.firstTask = null;
    // 释放锁（设置state为0，允许中断）
    w.unlock(); // allow interrupts
    boolean completedAbruptly = true;
    try {
        while (task != null || (task = getTask()) != null) { // 任务不为null或者阻塞队列还存在任务
            // 获取锁
            w.lock();
            // If pool is stopping, ensure thread is interrupted;
            // if not, ensure thread is not interrupted.  This
            // requires a recheck in second case to deal with
            // shutdownNow race while clearing interrupt
            if ((runStateAtLeast(ctl.get(), STOP) ||    // 线程池的运行状态至少应该高于STOP
                 (Thread.interrupted() &&                // 线程被中断
                  runStateAtLeast(ctl.get(), STOP))) &&    // 再次检查，线程池的运行状态至少应该高于STOP
                !wt.isInterrupted())                    // wt线程（当前线程）没有被中断
                wt.interrupt();                            // 中断wt线程（当前线程）
            try {
                // 在执行之前调用钩子函数
                beforeExecute(wt, task);
                Throwable thrown = null;
                try {
                    // 运行给定的任务
                    task.run();
                } catch (RuntimeException x) {
                    thrown = x; throw x;
                } catch (Error x) {
                    thrown = x; throw x;
                } catch (Throwable x) {
                    thrown = x; throw new Error(x);
                } finally {
                    // 执行完后调用钩子函数
                    afterExecute(task, thrown);
                }
            } finally {
                task = null;
                // 增加给worker完成的任务数量
                w.completedTasks++;
                // 释放锁
                w.unlock();
            }
        }
        completedAbruptly = false;
    } finally {
        // 处理完成后，调用钩子函数
        processWorkerExit(w, completedAbruptly);
    }
}

This function is used to obtain workerQueue Runnable object from blocking the queue, since the queue is blocked, it waits for a limited time (poll) and infinite wait (take). In this function and also responds to operation shutDown, shutdownNow function, upon detecting the thread pool is in the STOP state SHUTDOWN or is returned null, and the object does not return Runnalbe blocked queue.

    private Runnable getTask() {
        boolean timedOut = false; // Did the last poll() time out?

        for (;;) { // 无限循环，确保操作成功
            // 获取线程池控制状态
            int c = ctl.get();
            // 运行的状态
            int rs = runStateOf(c);

            // Check if queue empty only if necessary.
            if (rs >= SHUTDOWN && (rs >= STOP || workQueue.isEmpty())) { // 大于等于SHUTDOWN（表示调用了shutDown）并且（大于等于STOP（调用了shutDownNow）或者worker阻塞队列为空）
                // 减少worker的数量
                decrementWorkerCount();
                // 返回null，不执行任务
                return null;
            }
            // 获取worker数量
            int wc = workerCountOf(c);

            // Are workers subject to culling?
            boolean timed = allowCoreThreadTimeOut || wc > corePoolSize; // 是否允许coreThread超时或者workerCount大于核心大小

            if ((wc > maximumPoolSize || (timed && timedOut))     // worker数量大于maximumPoolSize
                && (wc > 1 || workQueue.isEmpty())) {            // workerCount大于1或者worker阻塞队列为空（在阻塞队列不为空时，需要保证至少有一个wc）
                if (compareAndDecrementWorkerCount(c))            // 比较并减少workerCount
                    // 返回null，不执行任务，该worker会退出
                    return null;
                // 跳过剩余部分，继续循环
                continue;
            }

            try {
                Runnable r = timed ?
                    workQueue.poll(keepAliveTime, TimeUnit.NANOSECONDS) :    // 等待指定时间
                    workQueue.take();                                        // 一直等待，直到有元素
                if (r != null)
                    return r;
                // 等待指定时间后，没有获取元素，则超时
                timedOut = true;
            } catch (InterruptedException retry) {
                // 抛出了被中断异常，重试，没有超时
                timedOut = false;
            }
        }
    }

processWorkerExit function is invoked at the worker to quit the hook function, which leads to the following main factors worker quit

Blocking queue is already empty, that there is no task to run.
ShutDown function call or shutDownNow

This function will be determined based on whether or not the idle thread is interrupted to reduce the value of workerCount, and the worker removed from the workers set and will try to terminate the thread pool.

    private void processWorkerExit(Worker w, boolean completedAbruptly) {
        if (completedAbruptly) // 如果被中断，则需要减少workCount    // If abrupt, then workerCount wasn't adjusted
            decrementWorkerCount();
        // 获取可重入锁
        final ReentrantLock mainLock = this.mainLock;
        // 获取锁
        mainLock.lock();
        try {
            // 将worker完成的任务添加到总的完成任务中
            completedTaskCount += w.completedTasks;
            // 从workers集合中移除该worker
            workers.remove(w);
        } finally {
            // 释放锁
            mainLock.unlock();
        }
        // 尝试终止
        tryTerminate();
        // 获取线程池控制状态
        int c = ctl.get();
        if (runStateLessThan(c, STOP)) { // 小于STOP的运行状态
            if (!completedAbruptly) {
                int min = allowCoreThreadTimeOut ? 0 : corePoolSize;
                if (min == 0 && ! workQueue.isEmpty()) // 允许核心超时并且workQueue阻塞队列不为空
                    min = 1;
                if (workerCountOf(c) >= min) // workerCount大于等于min
                    // 直接返回
                    return; // replacement not needed
            }
            // 添加worker
            addWorker(null, false);
        }
    }

Close the thread pool

    public void shutdown() {
        final ReentrantLock mainLock = this.mainLock;
        mainLock.lock();
        try {
            // 检查shutdown权限
            checkShutdownAccess();
            // 设置线程池控制状态为SHUTDOWN
            advanceRunState(SHUTDOWN);
            // 中断空闲worker
            interruptIdleWorkers();
            // 调用shutdown钩子函数
            onShutdown(); // hook for ScheduledThreadPoolExecutor
        } finally {
            mainLock.unlock();
        }
        // 尝试终止
        tryTerminate();
    }

    final void tryTerminate() {
        for (;;) { // 无限循环，确保操作成功
            // 获取线程池控制状态
            int c = ctl.get();
            if (isRunning(c) ||                                            // 线程池的运行状态为RUNNING
                runStateAtLeast(c, TIDYING) ||                            // 线程池的运行状态最小要大于TIDYING
                (runStateOf(c) == SHUTDOWN && ! workQueue.isEmpty()))    // 线程池的运行状态为SHUTDOWN并且workQueue队列不为null
                // 不能终止，直接返回
                return;
            if (workerCountOf(c) != 0) { // 线程池正在运行的worker数量不为0    // Eligible to terminate
                // 仅仅中断一个空闲的worker
                interruptIdleWorkers(ONLY_ONE);
                return;
            }
            // 获取线程池的锁
            final ReentrantLock mainLock = this.mainLock;
            // 获取锁
            mainLock.lock();
            try {
                if (ctl.compareAndSet(c, ctlOf(TIDYING, 0))) { // 比较并设置线程池控制状态为TIDYING
                    try {
                        // 终止，钩子函数
                        terminated();
                    } finally {
                        // 设置线程池控制状态为TERMINATED
                        ctl.set(ctlOf(TERMINATED, 0));
                        // 释放在termination条件上等待的所有线程
                        termination.signalAll();
                    }
                    return;
                }
            } finally {
                // 释放锁
                mainLock.unlock();
            }
            // else retry on failed CAS
        }
    }

    private void interruptIdleWorkers(boolean onlyOne) {
        // 线程池的锁
        final ReentrantLock mainLock = this.mainLock;
        // 获取锁
        mainLock.lock();
        try {
            for (Worker w : workers) { // 遍历workers队列
                // worker对应的线程
                Thread t = w.thread;
                if (!t.isInterrupted() && w.tryLock()) { // 线程未被中断并且成功获得锁
                    try {
                        // 中断线程
                        t.interrupt();
                    } catch (SecurityException ignore) {
                    } finally {
                        // 释放锁
                        w.unlock();
                    }
                }
                if (onlyOne) // 若只中断一个，则跳出循环
                    break;
            }
        } finally {
            // 释放锁
            mainLock.unlock();
        }
    }

The Java parsing thread pool