2019-12-19

什么是死锁?如何避免

所谓死锁:是指两个或两个以上的进程在执行过程中,因争夺资源而造成的一种互相等待的现象,若无外力作用,它们都将无法推进下去。此时称系统处于死锁状态或系统产生了死锁,这些永远在互相等待的进程称为死锁进程。由于资源占用是互斥的,当某个进程提出申请资源后,使得有关进程在无外力协助下,永远分配不到必需的资源而无法继续运行,这就产生了一种特殊现象死锁。

虽然进程在运行过程中,可能发生死锁,但死锁的发生也必须具备一定的条件,死锁的发生必须具备以下四个必要条件。

1)互斥条件:指进程对所分配到的资源进行排它性使用,即在一段时间内某资源只由一个进程占用。如果此时还有其它进程请求资源,则请求者只能等待,直至占有资源的进程用毕释放。

2)请求和保持条件:指进程已经保持至少一个资源,但又提出了新的资源请求,而该资源已被其它进程占有,此时请求进程阻塞,但又对自己已获得的其它资源保持不放。

3)不剥夺条件:指进程已获得的资源,在未使用完之前,不能被剥夺,只能在使用完时由自己释放。

4)环路等待条件:指在发生死锁时,必然存在一个进程——资源的环形链,即进程集合{P0,P1,P2,···,Pn}中的P0正在等待一个P1占用的资源;P1正在等待P2占用的资源,……,Pn正在等待已被P0占用的资源。

在系统中已经出现死锁后,应该及时检测到死锁的发生,并采取适当的措施来解除死锁。目前处理死锁的方法可归结为以下四种:

1)预防死锁。

  这是一种较简单和直观的事先预防的方法。方法是通过设置某些限制条件,去破坏产生死锁的四个必要条件中的一个或者几个,来预防发生死锁。预防死锁是一种较易实现的方法,已被广泛使用。但是由于所施加的限制条件往往太严格,可能会导致系统资源利用率和系统吞吐量降低。

2)避免死锁。

  该方法同样是属于事先预防的策略,但它并不须事先采取各种限制措施去破坏产生死锁的的四个必要条件,而是在资源的动态分配过程中,用某种方法去防止系统进入不安全状态,从而避免发生死锁。

3)检测死锁。

  这种方法并不须事先采取任何限制性措施,也不必检查系统是否已经进入不安全区,此方法允许系统在运行过程中发生死锁。但可通过系统所设置的检测机构,及时地检测出死锁的发生,并精确地确定与死锁有关的进程和资源,然后采取适当措施,从系统中将已发生的死锁清除掉。

4)解除死锁。

  这是与检测死锁相配套的一种措施。当检测到系统中已发生死锁时,须将进程从死锁状态中解脱出来。常用的实施方法是撤销或挂起一些进程,以便回收一些资源,再将这些资源分配给已处于阻塞状态的进程,使之转为就绪状态,以继续运行。死锁的检测和解除措施,有可能使系统获得较好的资源利用率和吞吐量,但在实现上难度也最大。

线程和进程的差别是什么?

进程是指在系统中正在运行的一个应用程序;程序一旦运行就是进程,或者更专业化来说:进程是指程序执行时的一个实例。线程是进程的一个实体。

进程——资源分配的最小单位,

线程——程序执行的最小单位。

Java里面的Threadlocal是怎样实现的?

ThreadLocal 是线程本地数据存储类,通过ThreadLocal可以在特定的线程中存储数据和变量, 并且这些数据之后只能由该线程访问,其他线程是访问不了的, 保证各个线程里数据和变量的独立性; 即ThreadLocal使每个线程可以访问自己内部的副本变量。

ThreadLocal类提供的几个方法:    

public Tget() { }

get 方法是用来获取 ThreadLocal 在当前线程中保存的变量副本

public voidset(T value) { }

set 用来设置当前线程中变量的副本

public voidremove() { }

remove 用来移除当前线程中变量的副本

protected TinitialValue() { }

initialValue 是一个protected方法,一般是用来在使用时进行重写的,做初始化操作

ConcurrentHashMap的实现原理是?

ConcurrentHashMap是Java1.5中引用的一个线程安全的支持高并发的HashMap集合类。

1、线程不安全的HashMap

因为多线程环境下,使用Hashmap进行put操作会引起死循环,导致CPU利用率接近100%,所以在并发情况下不能使用HashMap。

2、效率低下的HashTable

HashTable容器使用synchronized来保证线程安全,但在线程竞争激烈的情况下HashTable的效率非常低下。

因为当一个线程访问HashTable的同步方法时,其他线程访问HashTable的同步方法时,可能会进入阻塞或轮询状态。

如线程1使用put进行添加元素,线程2不但不能使用put方法添加元素,并且也不能使用get方法来获取元素,所以竞争越激烈效率越低。

3、锁分段技术

HashTable容器在竞争激烈的并发环境下表现出效率低下的原因,是因为所有访问HashTable的线程都必须竞争同一把锁,

那假如容器里有多把锁,每一把锁用于锁容器其中一部分数据,那么当多线程访问容器里不同数据段的数据时,线程间就不会存在锁竞争,从而可以有效的提高并发访问效率,这就是ConcurrentHashMap所使用的锁分段技术。首先将数据分成一段一段的存储,然后给每一段数据配一把锁,当一个线程占用锁访问其中一个段数据的时候,其他段的数据也能被其他线程访问。有些方法需要跨段,比如size()和containsValue(),它们可能需要锁定整个表而而不仅仅是某个段,这需要按顺序锁定所有段,操作完毕后,又按顺序释放所有段的锁。这里“按顺序”是很重要的,否则极有可能出现死锁,在ConcurrentHashMap内部,段数组是final的,并且其成员变量实际上也是final的,但是,仅仅是将数组声明为final的并不保证数组成员也是final的,这需要实现上的保证。这可以确保不会出现死锁,因为获得锁的顺序是固定的。

sleep和wait区别

对于sleep()方法,我们首先要知道该方法是属于Thread类中的。而wait()方法,则是属于Object类中的。

sleep()方法导致了程序暂停执行指定的时间,让出cpu该其他线程,但是他的监控状态依然保持着,当指定的时间到了又会自动恢复运行状态。在调用sleep()方法的过程中,线程不会释放对象锁。而当调用wait()方法的时候,线程会放弃对象锁,进入等待此对象的等待锁定池,只有针对此对象调用notify()方法后本线程才进入对象锁定池准备,获取对象锁进入运行状态。

notify和notifyAll区别

wait后notify方法只唤醒一个等待(对象的)线程并使该线程开始执行。所以如果有多个线程等待一个对象,这个方法只会唤醒其中一个线程,选择哪个线程取决于操作系统对多线程管理的实现。notifyAll会唤醒所有等待(对象的)线程,尽管哪一个线程将会第一个处理取决于操作系统的实现

ThreadLocal的作用与实现

ThreadLocal<Boolean> mBooleanThreadLocal = new ThreadLocal<>();

这些数据之后只能由该线程访问,其他线程是访问不了的, 保证各个线程里数据和变量的独立性; 即ThreadLocal使每个线程可以访问自己内部的副本变量

两个线程如何串行执行

为了控制线程执行的顺序,如ThreadA->ThreadB->ThreadC->ThreadA循环执行三个线程,

我们需要确定唤醒、等待的顺序。这时我们可以同时使用 Obj.wait()、Obj.notify()与synchronized(Obj)来实现这个目标。

通常情况下,wait是线程在获取对象锁后,主动释放对象锁,同时本线程休眠,直到有其它线程调用对象的notify()唤醒该线程,才能继续获取对象锁,并继续执行。而notify()则是对等待对象锁的线程的唤醒操作。但值得注意的是notify()调用后,并不是马上就释放对象锁,而是在相应的synchronized(){}语句块执行结束。释放对象锁后,JVM会在执行wait()等待对象锁的线程中随机选取一线程,赋予其对象锁,唤醒线程,继续执行。

上下文切换是什么含义

每个任务运行前,CPU 都需要知道任务从哪里加载、又从哪里开始运行,这就涉及到 CPU 寄存器 和 程序计数器(PC):

CPU 寄存器是 CPU 内置的容量小、但速度极快的内存;程序计数器会存储 CPU 正在执行的指令位置,或者即将执行的指令位置。

这两个是 CPU 运行任何任务前都必须依赖的环境,因此叫做 CPU 上下文。

上下文切换

将前一个 CPU 的上下文(也就是 CPU 寄存器和程序计数器里边的内容)保存起来;

然后加载新任务的上下文到寄存器和程序计数器;

最后跳转到程序计数器所指的新位置,运行新任务。

被保存起来的上下文会存储到系统内核中,等待任务重新调度执行时再次加载进来。

CPU 的上下文切换分三种:进程上下文切换、线程上下文切换、中断上下文切换。

可以运行时kill掉一个线程吗?

如果一个线程由于等待某些事件的发生而被阻塞,又该怎样停止该线程呢?这种情况经常会发生,比如当一个线程由于需要等候键盘输入而被阻塞,或者调用Thread.join()方法,或者Thread.sleep()方法,在网络中调用ServerSocket.accept()方法,或者调用了DatagramSocket.receive()方法时,都有可能导致线程阻塞,使线程处于处于不可运行状态时,即使主程序中将该线程的共享变量设置为true,但该线程此时根本无法检查循环标志,当然也就无法立即中断。这里我们给出的建议是,不要使用stop()方法,而是使用Thread提供的interrupt()方法,因为该方法虽然不会中断一个正在运行的线程,但是它可以使一个被阻塞的线程抛出一个中断异常,从而使线程提前结束阻塞状态,退出堵塞代码。

public class TestPersonProxy extends Thread {

    volatile boolean stop = false;

    public void run() {

        while (!stop) {

            System.out.println(getName() + " is running");

            try {

                sleep(1000);

            } catch (InterruptedException e) {

                System.out.println("week up from blcok...");

                stop = true; // 在异常处理代码中修改共享变量的状态

            }

        }

        System.out.println(getName() + " is exiting...");

    }

}

class InterruptThreadDemo3 {

    public static void main(String[] args) throws InterruptedException {

        TestPersonProxy m1 = new TestPersonProxy();

        System.out.println("Starting thread...");

        m1.start();

        Thread.sleep(3000);

        System.out.println("Interrupt thread...: " + m1.getName());

        m1.stop = true; // 设置共享变量为true

        m1.interrupt(); // 阻塞时退出阻塞状态

        Thread.sleep(3000); // 主线程休眠3秒以便观察线程m1的中断情况

        System.out.println("Stopping application...");

    }

}

什么是条件锁、读写锁、自旋锁、可重入锁?

自旋锁可以使线程在没有取得锁的时候,不被挂起,而转去执行一个空循环,(即所谓的自旋,就是自己执行空循环),若在若干个空循环后,线程如果可以获得锁,则继续执行。若线程依然不能获得锁,才会被挂起。

使用自旋锁后,线程被挂起的几率相对减少,线程执行的连贯性相对加强。因此,对于那些锁竞争不是很激烈,锁占用时间很短的并发线程,具有一定的积极意义,但对于锁竞争激烈,单线程锁占用很长时间的并发程序,自旋锁在自旋等待后,往往毅然无法获得对应的锁,不仅仅白白浪费了CPU时间,最终还是免不了被挂起的操作 ,反而浪费了系统的资源。

在JDK1.6中,Java虚拟机提供-XX:+UseSpinning参数来开启自旋锁,使用-XX:PreBlockSpin参数来设置自旋锁等待的次数。

在JDK1.7开始,自旋锁的参数被取消,虚拟机不再支持由用户配置自旋锁,自旋锁总是会执行,自旋锁次数也由虚拟机自动调整。

可能引起的问题:

1.过多占据CPU时间:如果锁的当前持有者长时间不释放该锁,那么等待者将长时间的占据cpu时间片,导致CPU资源的浪费,因此可以设定一个时间,当锁持有者超过这个时间不释放锁时,等待者会放弃CPU时间片阻塞;

2.死锁问题:试想一下,有一个线程连续两次试图获得自旋锁(比如在递归程序中),第一次这个线程获得了该锁,当第二次试图加锁的时候,检测到锁已被占用(其实是被自己占用),那么这时,线程会一直等待自己释放该锁,而不能继续执行,这样就引起了死锁。因此递归程序使用自旋锁应该遵循以下原则:递归程序决不能在持有自旋锁时调用它自己,也决不能在递归调用时试图获得相同的自旋锁。

2、阻塞锁

让线程进入阻塞状态进行等待,当获得相应的信号(唤醒,时间) 时,才可以进入线程的准备就绪状态,准备就绪状态的所有线程,通过竞争,进入运行状态。。

JAVA中,能够进入\退出、阻塞状态或包含阻塞锁的方法有 ,synchronized 关键字(其中的重量锁),ReentrantLock,Object.wait()\notify()

3、可重入锁

可重入锁,也叫做递归锁,指的是同一线程 外层函数获得锁之后 ,内层递归函数仍然有获取该锁的代码,但不受影响。

在JAVA环境下 ReentrantLock 和synchronized 都是 可重入锁

线程池ThreadPoolExecutor的实现原理?

线程池有多重要

线程是一个程序员一定会涉及到的一个概念,但是线程的创建和切换都是代价比较大的。所以,我们有没有一个好的方案能做到线程的复用呢?这就涉及到一个概念——线程池。合理的使用线程池能够带来3个很明显的好处:

1.降低资源消耗:通过重用已经创建的线程来降低线程创建和销毁的消耗

2.提高响应速度:任务到达时不需要等待线程创建就可以立即执行。

3.提高线程的可管理性:线程池可以统一管理、分配、调优和监控。

java多线程池的支持——ThreadPoolExecutor

java的线程池支持主要通过ThreadPoolExecutor来实现,我们使用的ExecutorService的各种线程池策略都是基于ThreadPoolExecutor实现的,所以ThreadPoolExecutor十分重要。要弄明白各种线程池策略,必须先弄明白ThreadPoolExecutor。

实现原理

首先看一个线程池的流程图

step1.调用ThreadPoolExecutor的execute提交线程,首先检查CorePool,如果CorePool内的线程小于CorePoolSize,新创建线程执行任务。

step2.如果当前CorePool内的线程大于等于CorePoolSize,那么将线程加入到BlockingQueue。

step3.如果不能加入BlockingQueue,在小于MaxPoolSize的情况下创建线程执行任务。

step4.如果线程数大于等于MaxPoolSize,那么执行拒绝策略

线程池的创建

线程池的创建可以通过ThreadPoolExecutor的构造方法实现:

public ThreadPoolExecutor(int corePoolSize,

                              int maximumPoolSize,

                              long keepAliveTime,

                              TimeUnit unit,

                              BlockingQueue<Runnable> workQueue,

                              ThreadFactory threadFactory,

                              RejectedExecutionHandler handler) {

        if (corePoolSize < 0 ||

            maximumPoolSize <= 0 ||

            maximumPoolSize < corePoolSize ||

            keepAliveTime < 0)

            throw new IllegalArgumentException();

        if (workQueue == null || threadFactory == null || handler == null)

            throw new NullPointerException();

        this.corePoolSize = corePoolSize;

        this.maximumPoolSize = maximumPoolSize;

        this.workQueue = workQueue;

        this.keepAliveTime = unit.toNanos(keepAliveTime);

        this.threadFactory = threadFactory;

        this.handler = handler;

    具体解释一下上述参数:

corePoolSize 核心线程池大小

maximumPoolSize 线程池最大容量大小

keepAliveTime 线程池空闲时,线程存活的时间

TimeUnit 时间单位

ThreadFactory 线程工厂

BlockingQueue任务队列

RejectedExecutionHandler 线程拒绝策略

线程的提交

ThreadPoolExecutor的构造方法如上所示,但是只是做一些参数的初始化,ThreadPoolExecutor被初始化好之后便可以提交线程任务,线程的提交方法主要是execute和submit。这里主要说execute,submit会在后续的博文中分析

public void execute(Runnable command) {

        if (command == null)

            throw new NullPointerException();

        /*

         * Proceed in 3 steps:

         *

         * 1. If fewer than corePoolSize threads are running, try to

         * start a new thread with the given command as its first

         * task.  The call to addWorker atomically checks runState and

         * workerCount, and so prevents false alarms that would add

         * threads when it shouldn't, by returning false.

         * 如果当前的线程数小于核心线程池的大小,根据现有的线程作为第一个Worker运行的线程,

         * 新建一个Worker,addWorker自动的检查当前线程池的状态和Worker的数量,

         * 防止线程池在不能添加线程的状态下添加线程

         *

         * 2. If a task can be successfully queued, then we still need

         * to double-check whether we should have added a thread

         * (because existing ones died since last checking) or that

         * the pool shut down since entry into this method. So we

         * recheck state and if necessary roll back the enqueuing if

         * stopped, or start a new thread if there are none.

         *  如果线程入队成功,然后还是要进行double-check的,因为线程池在入队之后状态是可能会发生变化的

         *

         * 3. If we cannot queue task, then we try to add a new

         * thread.  If it fails, we know we are shut down or saturated

         * and so reject the task.

         *

         * 如果task不能入队(队列满了),这时候尝试增加一个新线程,如果增加失败那么当前的线程池状态变化了或者线程池已经满了

         * 然后拒绝task

         */

        int c = ctl.get();

        //当前的Worker的数量小于核心线程池大小时,新建一个Worker。

        if (workerCountOf(c) < corePoolSize) {

            if (addWorker(command, true))

                return;

            c = ctl.get();

        }

        if (isRunning(c) && workQueue.offer(command)) {

            int recheck = ctl.get();

            if (! isRunning(recheck) && remove(command))//recheck防止线程池状态的突变,如果突变,那么将reject线程,防止workQueue中增加新线程

                reject(command);

            else if (workerCountOf(recheck) == 0)//上下两个操作都有addWorker的操作,但是如果在workQueue.offer的时候Worker变为0,

                                                //那么将没有Worker执行新的task,所以增加一个Worker.

                addWorker(null, false);

        }

        //如果workQueue满了,那么这时候可能还没到线程池的maxnum,所以尝试增加一个Worker

        else if (!addWorker(command, false))

            reject(command);//如果Worker数量到达上限,那么就拒绝此线程

    }

这里需要明确几个概念:

Worker和Task的区别,Worker是当前线程池中的线程,而task虽然是runnable,但是并没有真正执行,只是被Worker调用了run方法,后面会看到这部分的实现。

maximumPoolSize和corePoolSize的区别:这个概念很重要,maximumPoolSize为线程池最大容量,也就是说线程池最多能起多少Worker。corePoolSize是核心线程池的大小,当corePoolSize满了时,同时workQueue full(ArrayBolckQueue是可能满的) 那么此时允许新建Worker去处理workQueue中的Task,但是不能超过maximumPoolSize。超过corePoolSize之外的线程会在空闲超时后终止。

核心方法:addWorker

Worker的增加和Task的获取以及终止都是在此方法中实现的,也就是这一个方法里面包含了很多东西。在addWorker方法中提到了Status的概念,Status是线程池的核心概念,这里我们先看一段关于status的注释:

/**

     * 首先ctl是一个原子量,同时它里面包含了两个field,一个是workerCount,另一个是runState

     * workerCount表示当前有效的线程数,也就是Worker的数量

     * runState表示当前线程池的状态

     * The main pool control state, ctl, is an atomic integer packing

     * two conceptual fields

     *   workerCount, indicating the effective number of threads

     *   runState,    indicating whether running, shutting down etc

     *

     * 两者是怎么结合的呢?首先workerCount是占据着一个atomic integer的后29位的,而状态占据了前3位

     * 所以,workerCount上限是(2^29)-1。

     * In order to pack them into one int, we limit workerCount to

     * (2^29)-1 (about 500 million) threads rather than (2^31)-1 (2

     * billion) otherwise representable. If this is ever an issue in

     * the future, the variable can be changed to be an AtomicLong,

     * and the shift/mask constants below adjusted. But until the need

     * arises, this code is a bit faster and simpler using an int.

     *

     * The workerCount is the number of workers that have been

     * permitted to start and not permitted to stop.  The value may be

     * transiently different from the actual number of live threads,

     * for example when a ThreadFactory fails to create a thread when

     * asked, and when exiting threads are still performing

     * bookkeeping before terminating. The user-visible pool size is

     * reported as the current size of the workers set.

     *

     * runState是整个线程池的运行生命周期,有如下取值:

     *  1. RUNNING:可以新加线程,同时可以处理queue中的线程。

     *  2. SHUTDOWN:不增加新线程,但是处理queue中的线程。

     *  3.STOP 不增加新线程,同时不处理queue中的线程。

     *  4.TIDYING 所有的线程都终止了(queue中),同时workerCount为0,那么此时进入TIDYING

     *  5.terminated()方法结束,变为TERMINATED

     * The runState provides the main lifecyle control, taking on values:

     *

     *   RUNNING:  Accept new tasks and process queued tasks

     *   SHUTDOWN: Don't accept new tasks, but process queued tasks

     *   STOP:     Don't accept new tasks, don't process queued tasks,

     *             and interrupt in-progress tasks

     *   TIDYING:  All tasks have terminated, workerCount is zero,

     *             the thread transitioning to state TIDYING

     *             will run the terminated() hook method

     *   TERMINATED: terminated() has completed

     *

     * The numerical order among these values matters, to allow

     * ordered comparisons. The runState monotonically increases over

     * time, but need not hit each state. The transitions are:

     * 状态的转化主要是:

     * RUNNING -> SHUTDOWN(调用shutdown())

     *    On invocation of shutdown(), perhaps implicitly in finalize()

     * (RUNNING or SHUTDOWN) -> STOP(调用shutdownNow())

     *    On invocation of shutdownNow()

     * SHUTDOWN -> TIDYING(queue和pool均empty)

     *    When both queue and pool are empty

     * STOP -> TIDYING(pool empty,此时queue已经为empty)

     *    When pool is empty

     * TIDYING -> TERMINATED(调用terminated())

     *    When the terminated() hook method has completed

     *

     * Threads waiting in awaitTermination() will return when the

     * state reaches TERMINATED.

     *

     * Detecting the transition from SHUTDOWN to TIDYING is less

     * straightforward than you'd like because the queue may become

     * empty after non-empty and vice versa during SHUTDOWN state, but

     * we can only terminate if, after seeing that it is empty, we see

     * that workerCount is 0 (which sometimes entails a recheck -- see

     * below).

     */

下面是状态的代码:

//利用ctl来保证当前线程池的状态和当前的线程的数量。ps:低29位为线程池容量,高3位为线程状态。

    private final AtomicInteger ctl = new AtomicInteger(ctlOf(RUNNING, 0));

    //设定偏移量

    private static final int COUNT_BITS = Integer.SIZE - 3;

    //确定最大的容量2^29-1

    private static final int CAPACITY   = (1 << COUNT_BITS) - 1;

    //几个状态,用Integer的高三位表示

    // runState is stored in the high-order bits

    //111

    private static final int RUNNING    = -1 << COUNT_BITS;

    //000

    private static final int SHUTDOWN   =  0 << COUNT_BITS;

    //001

    private static final int STOP       =  1 << COUNT_BITS;

    //010

    private static final int TIDYING    =  2 << COUNT_BITS;

    //011

    private static final int TERMINATED =  3 << COUNT_BITS;

    //获取线程池状态,取前三位

    // Packing and unpacking ctl

    private static int runStateOf(int c)     { return c & ~CAPACITY; }

    //获取当前正在工作的worker,主要是取后面29位

    private static int workerCountOf(int c)  { return c & CAPACITY; }

    //获取ctl

    private static int ctlOf(int rs, int wc) { return rs | wc; }

接下来贴上addWorker方法看看:

    /**

     * Checks if a new worker can be added with respect to current

     * pool state and the given bound (either core or maximum). If so,

     * the worker count is adjusted accordingly, and, if possible, a

     * new worker is created and started running firstTask as its

     * first task. This method returns false if the pool is stopped or

     * eligible to shut down. It also returns false if the thread

     * factory fails to create a thread when asked, which requires a

     * backout of workerCount, and a recheck for termination, in case

     * the existence of this worker was holding up termination.

     *

     * @param firstTask the task the new thread should run first (or

     * null if none). Workers are created with an initial first task

     * (in method execute()) to bypass queuing when there are fewer

     * than corePoolSize threads (in which case we always start one),

     * or when the queue is full (in which case we must bypass queue).

     * Initially idle threads are usually created via

     * prestartCoreThread or to replace other dying workers.

     *

     * @param core if true use corePoolSize as bound, else

     * maximumPoolSize. (A boolean indicator is used here rather than a

     * value to ensure reads of fresh values after checking other pool

     * state).

     * @return true if successful

     */

    private boolean addWorker(Runnable firstTask, boolean core) {

        retry:

        for (;;) {

            int c = ctl.get();

            int rs = runStateOf(c);

            // Check if queue empty only if necessary.

            /**

             * rs!=Shutdown || fistTask!=null || workCount.isEmpty

             * 如果当前的线程池的状态>SHUTDOWN 那么拒绝Worker的add 如果=SHUTDOWN

             * 那么此时不能新加入不为null的Task,如果在WorkCount为empty的时候不能加入任何类型的Worker,

             * 如果不为empty可以加入task为null的Worker,增加消费的Worker

             */

            if (rs >= SHUTDOWN &&

                ! (rs == SHUTDOWN &&

                   firstTask == null &&

                   ! workQueue.isEmpty()))

                return false;

            for (;;) {

                int wc = workerCountOf(c);

                if (wc >= CAPACITY ||

                    wc >= (core ? corePoolSize : maximumPoolSize))

                    return false;

                if (compareAndIncrementWorkerCount(c))

                    break retry;

                c = ctl.get();  // Re-read ctl

                if (runStateOf(c) != rs)

                    continue retry;

                // else CAS failed due to workerCount change; retry inner loop

            }

        }

        Worker w = new Worker(firstTask);

        Thread t = w.thread;

        final ReentrantLock mainLock = this.mainLock;

        mainLock.lock();

        try {

            // Recheck while holding lock.

            // Back out on ThreadFactory failure or if

            // shut down before lock acquired.

            int c = ctl.get();

            int rs = runStateOf(c);

            /**

             * rs!=SHUTDOWN ||firstTask!=null

             *

             * 同样检测当rs>SHUTDOWN时直接拒绝减小Wc,同时Terminate,如果为SHUTDOWN同时firstTask不为null的时候也要Terminate

             */

            if (t == null ||

                (rs >= SHUTDOWN &&

                 ! (rs == SHUTDOWN &&

                    firstTask == null))) {

                decrementWorkerCount();

                tryTerminate();

                return false;

            }

            workers.add(w);

            int s = workers.size();

            if (s > largestPoolSize)

                largestPoolSize = s;

        } finally {

            mainLock.unlock();

        }

        t.start();

        // It is possible (but unlikely) for a thread to have been

        // added to workers, but not yet started, during transition to

        // STOP, which could result in a rare missed interrupt,

        // because Thread.interrupt is not guaranteed to have any effect

        // on a non-yet-started Thread (see Thread#interrupt).

        //Stop或线程Interrupt的时候要中止所有的运行的Worker

        if (runStateOf(ctl.get()) == STOP && ! t.isInterrupted())

            t.interrupt();

        return true;

    }

addWorker中首先进行了一次线程池状态的检测

            int c = ctl.get();

            int rs = runStateOf(c);

            // Check if queue empty only if necessary.

            //判断当前线程池的状态是不是已经shutdown,如果shutdown了拒绝线程加入

            //(rs!=SHUTDOWN || first!=null || workQueue.isEmpty())

            //如果rs不为SHUTDOWN,此时状态是STOP、TIDYING或TERMINATED,所以此时要拒绝请求

            //如果此时状态为SHUTDOWN,而传入一个不为null的线程,那么需要拒绝

            //如果状态为SHUTDOWN,同时队列中已经没任务了,那么拒绝掉

            if (rs >= SHUTDOWN &&

                ! (rs == SHUTDOWN &&

                   firstTask == null &&

                   ! workQueue.isEmpty()))

                return false;

其实是比较难懂的,主要在线程池状态判断条件这里:

如果是runing,那么跳过if。

如果rs>=SHUTDOWN,同时不等于SHUTDOWN,即为SHUTDOWN以上的状态,那么不接受新线程。

如果rs>=SHUTDOWN,同时等于SHUTDOWN,同时first!=null,那么拒绝新线程,如果first==null,那么可能是新增加线程消耗Queue中的线程。但是同时还要检测workQueue是否isEmpty(),如果为Empty,那么队列已空,不需要增加消耗线程,如果队列没有空那么运行增加first=null的Worker。

从这里是可以看出一些策略的

首先,在rs>SHUTDOWN时,拒绝一切线程的增加,因为STOP是会终止所有的线程,同时移除Queue中所有的待执行的线程的,所以也不需要增加first=null的Worker了

其次,在SHUTDOWN状态时,是不能增加first!=null的Worker的,同时即使first=null,但是此时Queue为Empty也是不允许增加Worker的,SHUTDOWN下增加的Worker主要用于消耗Queue中的任务。

SHUTDOWN状态时,是不允许向workQueue中增加线程的,isRunning(c) && workQueue.offer(command) 每次在offer之前都要做状态检测,也就是线程池状态变为>=SHUTDOWN时不允许新线程进入线程池了。

            for (;;) {

                int wc = workerCountOf(c);

                //如果当前的数量超过了CAPACITY,或者超过了corePoolSize和maximumPoolSize(试core而定)

                if (wc >= CAPACITY ||

                    wc >= (core ? corePoolSize : maximumPoolSize))

                    return false;

                //CAS尝试增加线程数,如果失败,证明有竞争,那么重新到retry。

                if (compareAndIncrementWorkerCount(c))

                    break retry;

                c = ctl.get();  // Re-read ctl

                //判断当前线程池的运行状态

                if (runStateOf(c) != rs)

                    continue retry;

                // else CAS failed due to workerCount change; retry inner loop

            }

这段代码做了一个兼容,主要是没有到corePoolSize 或maximumPoolSize上限时,那么允许添加线程,CAS增加Worker的数量后,跳出循环。

接下来实例化Worker,实例化Worker其实是很关键的,后面会说。

因为workers是HashSet线程不安全的,那么此时需要加锁,所以mainLock.lock(); 之后重新检查线程池的状态,如果状态不正确,那么减小Worker的数量,为什么tryTerminate()目前不大清楚。如果状态正常,那么添加Worker到workers。最后:

  if (runStateOf(ctl.get()) == STOP && ! t.isInterrupted())

            t.interrupt();

注释说的很清楚,为了能及时的中断此Worker,因为线程存在未Start的情况,此时是不能响应中断的,如果此时status变为STOP,则不能中断线程。此处用作中断线程之用。

接下来我们看Worker的方法:

/**

         * Creates with given first task and thread from ThreadFactory.

         * @param firstTask the first task (null if none)

         */

        Worker(Runnable firstTask) {

            this.firstTask = firstTask;

            this.thread = getThreadFactory().newThread(this);

        }

这里可以看出Worker是对firstTask的包装,并且Worker本身就是Runnable的,看上去真心很流氓的感觉~~~

通过ThreadFactory为Worker自己构建一个线程。

因为Worker是Runnable类型的,所以是有run方法的,上面也看到了会调用t.start() 其实就是执行了run方法:

        /** Delegates main run loop to outer runWorker  */

        public void run() {

            runWorker(this);

        }

调用了runWorker:

/**

     * Main worker run loop.  Repeatedly gets tasks from queue and

     * executes them, while coping with a number of issues:

     * 1 Worker可能还是执行一个初始化的task——firstTask。

     *    但是有时也不需要这个初始化的task(可以为null),只要pool在运行,就会

     *   通过getTask从队列中获取Task,如果返回null,那么worker退出。

     *   另一种就是external抛出异常导致worker退出。

     * 1. We may start out with an initial task, in which case we

     * don't need to get the first one. Otherwise, as long as pool is

     * running, we get tasks from getTask. If it returns null then the

     * worker exits due to changed pool state or configuration

     * parameters.  Other exits result from exception throws in

     * external code, in which case completedAbruptly holds, which

     * usually leads processWorkerExit to replace this thread.

     *

     *

     * 2 在运行任何task之前,都需要对worker加锁来防止other pool中断worker。

     *   clearInterruptsForTaskRun保证除了线程池stop,那么现场都没有中断标志

     * 2. Before running any task, the lock is acquired to prevent

     * other pool interrupts while the task is executing, and

     * clearInterruptsForTaskRun called to ensure that unless pool is

     * stopping, this thread does not have its interrupt set.

     *

     * 3. Each task run is preceded by a call to beforeExecute, which

     * might throw an exception, in which case we cause thread to die

     * (breaking loop with completedAbruptly true) without processing

     * the task.

     *

     * 4. Assuming beforeExecute completes normally, we run the task,

     * gathering any of its thrown exceptions to send to

     * afterExecute. We separately handle RuntimeException, Error

     * (both of which the specs guarantee that we trap) and arbitrary

     * Throwables.  Because we cannot rethrow Throwables within

     * Runnable.run, we wrap them within Errors on the way out (to the

     * thread's UncaughtExceptionHandler).  Any thrown exception also

     * conservatively causes thread to die.

     *

     * 5. After task.run completes, we call afterExecute, which may

     * also throw an exception, which will also cause thread to

     * die. According to JLS Sec 14.20, this exception is the one that

     * will be in effect even if task.run throws.

     *

     * The net effect of the exception mechanics is that afterExecute

     * and the thread's UncaughtExceptionHandler have as accurate

     * information as we can provide about any problems encountered by

     * user code.

     *

     * @param w the worker

     */

    final void runWorker(Worker w) {

        Runnable task = w.firstTask;

        w.firstTask = null;

        //标识线程是不是异常终止的

        boolean completedAbruptly = true;

        try {

            //task不为null情况是初始化worker时,如果task为null,则去队列中取线程--->getTask()

            while (task != null || (task = getTask()) != null) {

                w.lock();

                //获取woker的锁,防止线程被其他线程中断

                clearInterruptsForTaskRun();//清楚所有中断标记

                try {

                    beforeExecute(w.thread, task);//线程开始执行之前执行此方法,可以实现Worker未执行退出,本类中未实现

                    Throwable thrown = null;

                    try {

                        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 {

                        afterExecute(task, thrown);//线程执行后执行,可以实现标识Worker异常中断的功能,本类中未实现

                    }

                } finally {

                    task = null;//运行过的task标null

                    w.completedTasks++;

                    w.unlock();

                }

            }

            completedAbruptly = false;

        } finally {

            //处理worker退出的逻辑

            processWorkerExit(w, completedAbruptly);

        }

    }

    从上面代码可以看出,execute的Task是被“包装 ”了一层,线程启动时是内部调用了Task的run方法。

接下来所有的核心集中在getTask()方法上:

/**

     * Performs blocking or timed wait for a task, depending on

     * current configuration settings, or returns null if this worker

     * must exit because of any of:

     * 1. There are more than maximumPoolSize workers (due to

     *    a call to setMaximumPoolSize).

     * 2. The pool is stopped.

     * 3. The pool is shutdown and the queue is empty.

     * 4. This worker timed out waiting for a task, and timed-out

     *    workers are subject to termination (that is,

     *    {@code allowCoreThreadTimeOut || workerCount > corePoolSize})

     *    both before and after the timed wait.

     *

     * @return task, or null if the worker must exit, in which case

     *         workerCount is decremented

     *         

     *         

     *  队列中获取线程

     */

    private Runnable getTask() {

        boolean timedOut = false; // Did the last poll() time out?

        retry:

        for (;;) {

            int c = ctl.get();

            int rs = runStateOf(c);

            // Check if queue empty only if necessary.

            //当前状态为>stop时,不处理workQueue中的任务,同时减小worker的数量所以返回null,如果为shutdown 同时workQueue已经empty了,同样减小worker数量并返回null

            if (rs >= SHUTDOWN && (rs >= STOP || workQueue.isEmpty())) {

                decrementWorkerCount();

                return null;

            }

            boolean timed;      // Are workers subject to culling?

            for (;;) {

                int wc = workerCountOf(c);

                timed = allowCoreThreadTimeOut || wc > corePoolSize;

                if (wc <= maximumPoolSize && ! (timedOut && timed))

                    break;

                if (compareAndDecrementWorkerCount(c))

                    return null;

                c = ctl.get();  // Re-read ctl

                if (runStateOf(c) != rs)

                    continue retry;

                // else CAS failed due to workerCount change; retry inner loop

            }

            try {

                Runnable r = timed ?

                    workQueue.poll(keepAliveTime, TimeUnit.NANOSECONDS) :

                    workQueue.take();

                if (r != null)

                    return r;

                timedOut = true;

            } catch (InterruptedException retry) {

                timedOut = false;

            }

        }

    }

    这段代码十分关键,首先看几个局部变量:

boolean timedOut = false;

主要是判断后面的poll是否要超时

boolean timed;

主要是标识着当前Worker超时是否要退出。wc > corePoolSize时需要减小空闲的Worker数,那么timed为true,但是wc <= corePoolSize时,不能减小核心线程数timed为false。

timedOut初始为false,如果timed为true那么使用poll取线程。如果正常返回,那么返回取到的task。如果超时,证明worker空闲,同时worker超过了corePoolSize,需要删除。返回r=null。则 timedOut = true。此时循环到wc <= maximumPoolSize && ! (timedOut && timed)时,减小worker数,并返回null,导致worker退出。如果线程数<= corePoolSize,那么此时调用 workQueue.take(),没有线程获取到时将一直阻塞,知道获取到线程或者中断,关于中断后面Shutdown的时候会说。

至此线程执行过程就分析完了~~~~

关于终止线程池

我个人认为,如果想了解明白线程池,那么就一定要理解好各个状态之间的转换,想理解转换,线程池的终止机制是很好的一个途径。对于关闭线程池主要有两个方法shutdown()和shutdownNow():

首先从shutdown()方法开始:

    /**

     * Initiates an orderly shutdown in which previously submitted

     * tasks are executed, but no new tasks will be accepted.

     * Invocation has no additional effect if already shut down.

     *

     * <p>This method does not wait for previously submitted tasks to

     * complete execution.  Use {@link #awaitTermination awaitTermination}

     * to do that.

     *

     * @throws SecurityException {@inheritDoc}

     */

    public void shutdown() {

        final ReentrantLock mainLock = this.mainLock;

        mainLock.lock();

        try {

            //判断是否可以操作目标线程

            checkShutdownAccess();

            //设置线程池状态为SHUTDOWN,此处之后,线程池中不会增加新Task

            advanceRunState(SHUTDOWN);

            //中断所有的空闲线程

            interruptIdleWorkers();

            onShutdown(); // hook for ScheduledThreadPoolExecutor

        } finally {

            mainLock.unlock();

        }

        //转到Terminate

        tryTerminate();

    }        

shutdown做了几件事:

1. 检查是否能操作目标线程

2. 将线程池状态转为SHUTDOWN

3. 中断所有空闲线程

这里就引发了一个问题,什么是空闲线程?

这需要接着看看interruptIdleWorkers是怎么回事。

private void interruptIdleWorkers(boolean onlyOne) {

        final ReentrantLock mainLock = this.mainLock;

        mainLock.lock();

        //这里的意图很简单,遍历workers 对所有worker做中断处理。

        // w.tryLock()对Worker加锁,这保证了正在运行执行Task的Worker不会被中断,那么能中断哪些线程呢?

        try {

            for (Worker w : workers) {

                Thread t = w.thread;

                if (!t.isInterrupted() && w.tryLock()) {

                    try {

                        t.interrupt();

                    } catch (SecurityException ignore) {

                    } finally {

                        w.unlock();

                    }

                }

                if (onlyOne)

                    break;

            }

        } finally {

            mainLock.unlock();

        }

    }

    这里主要是为了中断worker,但是中断之前需要先获取锁,这就意味着正在运行的Worker不能中断。但是上面的代码有w.tryLock(),那么获取不到锁就不会中断,shutdown的Interrupt只是对所有的空闲Worker(正在从workQueue中取Task,此时Worker没有加锁)发送中断信号。

            while (task != null || (task = getTask()) != null) {

                w.lock();

                //获取woker的锁,防止线程被其他线程中断

                clearInterruptsForTaskRun();//清楚所有中断标记

                try {

                    beforeExecute(w.thread, task);//线程开始执行之前执行此方法,可以实现Worker未执行退出,本类中未实现

                    Throwable thrown = null;

                    try {

                        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 {

                        afterExecute(task, thrown);//线程执行后执行,可以实现标识Worker异常中断的功能,本类中未实现

                    }

                } finally {

                    task = null;//运行过的task标null

                    w.completedTasks++;

                    w.unlock();

                }

            }

在runWorker中,每一个Worker getTask成功之后都要获取Worker的锁之后运行,也就是说运行中的Worker不会中断。因为核心线程一般在空闲的时候会一直阻塞在获取Task上,也只有中断才可能导致其退出。这些阻塞着的Worker就是空闲的线程(当然,非核心线程,并且阻塞的也是空闲线程)。在getTask方法中

    private Runnable getTask() {

        boolean timedOut = false; // Did the last poll() time out?

        retry:

        for (;;) {

            int c = ctl.get();

            int rs = runStateOf(c);

            // Check if queue empty only if necessary.

            //当前状态为>stop时,不处理workQueue中的任务,同时减小worker的数量所以返回null,如果为shutdown 同时workQueue已经empty了,同样减小worker数量并返回null

            if (rs >= SHUTDOWN && (rs >= STOP || workQueue.isEmpty())) {

                decrementWorkerCount();

                return null;

            }

            boolean timed;      // Are workers subject to culling?

            for (;;) {

                //allowCoreThreadTimeOu是判断CoreThread是否会超时的,true为会超时,false不会超时。默认为false

                int wc = workerCountOf(c);

                timed = allowCoreThreadTimeOut || wc > corePoolSize;

                if (wc <= maximumPoolSize && ! (timedOut && timed))

                    break;

                if (compareAndDecrementWorkerCount(c))

                    return null;

                c = ctl.get();  // Re-read ctl

                if (runStateOf(c) != rs)

                    continue retry;

                // else CAS failed due to workerCount change; retry inner loop

            }

            try {

                Runnable r = timed ?

                    workQueue.poll(keepAliveTime, TimeUnit.NANOSECONDS) :

                    workQueue.take();

                if (r != null)

                    return r;

                timedOut = true;

            } catch (InterruptedException retry) {

                timedOut = false;

            }

        }

    }

会有两阶段的Worker:

刚进入getTask(),还没进行状态判断。

block在poll或者take上的Worker。

当调用ShutDown方法时,首先设置了线程池的状态为ShutDown,此时1阶段的worker进入到状态判断时会返回null,此时Worker退出。

因为getTask的时候是不加锁的,所以在shutdown时可以调用worker.Interrupt.此时会中断退出,Loop到状态判断时,同时workQueue为empty。那么抛出中断异常,导致重新Loop,在检测线程池状态时,Worker退出。如果workQueue不为null就不会退出,此处有些疑问,因为没有看见中断标志位清除的逻辑,那么这里就会不停的循环直到workQueue为Empty退出。

这里也能看出来SHUTDOWN只是清除一些空闲Worker,并且拒绝新Task加入,对于workQueue中的线程还是继续处理的。

对于shutdown中获取mainLock而addWorker中也做了mainLock的获取,这么做主要是因为Works是HashSet类型的,是线程不安全的,我们也看到在addWorker后面也是对线程池状态做了判断,将Worker添加和中断逻辑分离开。

接下来做了tryTerminate()操作,这操作是进行了后面状态的转换,在shutdownNow后面说。

接下来看看shutdownNow:

    /**

     * Attempts to stop all actively executing tasks, halts the

     * processing of waiting tasks, and returns a list of the tasks

     * that were awaiting execution. These tasks are drained (removed)

     * from the task queue upon return from this method.

     *

     * <p>This method does not wait for actively executing tasks to

     * terminate.  Use {@link #awaitTermination awaitTermination} to

     * do that.

     *

     * <p>There are no guarantees beyond best-effort attempts to stop

     * processing actively executing tasks.  This implementation

     * cancels tasks via {@link Thread#interrupt}, so any task that

     * fails to respond to interrupts may never terminate.

     *

     * @throws SecurityException {@inheritDoc}

     */

    public List<Runnable> shutdownNow() {

        List<Runnable> tasks;

        final ReentrantLock mainLock = this.mainLock;

        mainLock.lock();

        try {

            checkShutdownAccess();

            advanceRunState(STOP);

            interruptWorkers();

            tasks = drainQueue();

        } finally {

            mainLock.unlock();

        }

        tryTerminate();

        return tasks;

    }

shutdownNow和shutdown代码类似,但是实现却很不相同。首先是设置线程池状态为STOP,前面的代码我们可以看到,是对SHUTDOWN有一些额外的判断逻辑,但是对于>=STOP,基本都是reject,STOP也是比SHUTDOWN更加严格的一种状态。此时不会有新Worker加入,所有刚执行完一个线程后去GetTask的Worker都会退出。

之后调用interruptWorkers:

    /**

     * Interrupts all threads, even if active. Ignores SecurityExceptions

     * (in which case some threads may remain uninterrupted).

     */

    private void interruptWorkers() {

        final ReentrantLock mainLock = this.mainLock;

        mainLock.lock();

        try {

            for (Worker w : workers) {

                try {

                    w.thread.interrupt();

                } catch (SecurityException ignore) {

                }

            }

        } finally {

            mainLock.unlock();

        }

    }

©著作权归作者所有,转载或内容合作请联系作者
平台声明:文章内容(如有图片或视频亦包括在内)由作者上传并发布,文章内容仅代表作者本人观点,简书系信息发布平台,仅提供信息存储服务。

推荐阅读更多精彩内容