概述
为什么要使用线程池?
合理利用线程池能够带来三个好处。第一:降低资源消耗。通过重复利用已创建的线程降低线程创建和销毁造成的消耗。第二:提高响应速度。当任务到达时,任务可以不需要等到线程创建就能立即执行。第三:提高线程的可管理性。线程是稀缺资源,如果无限制的创建,不仅会消耗系统资源,还会降低系统的稳定性,使用线程池可以进行统一的分配,调优和监控。但是要做到合理的利用线程池,必须对其原理了如指掌。——摘自http://www.infoq.com/cn/articles/java-threadPool。
类图
使用
线程池的监控
可以通过线程池的以下属性监控线程池的当前状态:
getTaskCount():线程池已经执行的和未执行的任务总数,因为统计的过程中可能会发生变化,该值是个近似值;
getCompletedTaskCount():已完成的任务数量,是个近似值,该值小于等于TaskCount;
getLargestPoolSize():线程池曾经的最大线程数量,可以通过该值判断线程池是否满过。如该数值等于线程池的最大大小,则表示线程池曾经满过;
getPoolSize():线程池当前的线程数量;
getActiveCount():线程池中活动的线程数(正在执行任务),是个近似值。
还可以通过重写线程池提供的hook方法(beforeExecute、afterExecute和terminated)进行监控,例如监控任务的平均执行时间、最大执行时间和最小执行时间等。
程序员可以通过重写钩子 hook 方法(如beforeExecute)实现ThreadPoolExecutor的扩展。
扩展示例:添加了简单的暂停/恢复功能的子类
1 class PausableThreadPoolExecutor extends ThreadPoolExecutor { 2 private boolean isPaused; //标志是否被暂停 3 private ReentrantLock pauseLock = new ReentrantLock(); //访问isPaused时需要加锁,保证线程安全 4 private Condition unpaused = pauseLock.newCondition(); 5 6 public PausableThreadPoolExecutor(...) { super(...); } 7 8 //beforeExecute为ThreadPoolExecutor提供的hood方法 9 protected void beforeExecute(Thread t, Runnable r) { 10 super.beforeExecute(t, r); 11 pauseLock.lock(); 12 try { 13 while (isPaused) 14 unpaused.await(); 15 } catch(InterruptedException ie) { 16 t.interrupt(); 17 } finally { 18 pauseLock.unlock(); 19 } 20 } 21 //暂停 22 public void pause() { 23 pauseLock.lock(); 24 try { 25 isPaused = true; 26 } finally { 27 pauseLock.unlock(); 28 } 29 } 30 //取消暂停 31 public void resume() { 32 pauseLock.lock(); 33 try { 34 isPaused = false; 35 unpaused.signalAll(); 36 } finally { 37 pauseLock.unlock(); 38 } 39 } 40 }
实现原理
ThreadPoolExecutor源码分析
域
1 //ctl是控制线程池状态的一个变量,包含有效的线程数(workerCount)和线程池的运行状态(runState)两部分信息。高3位表示runState,低29位表示workerCount。 2 private final AtomicInteger ctl = new AtomicInteger(ctlOf(RUNNING, 0)); 3 private static final int COUNT_BITS = Integer.SIZE - 3; //表示workerCount的位数,29位。 4 private static final int CAPACITY = (1 << COUNT_BITS) - 1; //线程数的上限,(2^29)-1,大约5亿 5 6 // runState is stored in the high-order bits 7 private static final int RUNNING = -1 << COUNT_BITS; //能接收新任务和处理队列中的任务 8 private static final int SHUTDOWN = 0 << COUNT_BITS; //不能接收新任务,但可以处理队列中的任务 9 private static final int STOP = 1 << COUNT_BITS; //不能接收新任务,不能处理队列中的任务,中断正在执行的任务 10 private static final int TIDYING = 2 << COUNT_BITS; //所有的线程都被终止,workerCount为0时会进入该状态. 11 private static final int TERMINATED = 3 << COUNT_BITS; //terminated()方法完成后将进入该状态。
以上ThreadPoolExecutor的成员变量表示线程池的状态,状态信息存储在ctl变量中,ctl包含有效线程数(workerCount)和线程池运行状态(runState)两部分信息,ctl的高3位表示runState,低29位表示workerCount。ctl初始值为RUNNING状态且线程数为0。
线程池运行状态的转换如下:
1)线程池在RUNNING状态下调用shutdown()方法会进入到SHUTDOWN状态,(finalize()方法也会调用shutdownNow())。
2)在RUNNING和SHUTDOWN状态下调用 shutdownNow() 方法会进入到STOP状态。
3)在SHUTDOWN状态下,当阻塞队列为空且线程数为0时进入TIDYING状态;在STOP状态下,当线程数为0时进入TIDYING状态。
4)在TIDYING状态,调用terminated()方法完成后进入TERMINATED状态。
1 //阻塞队列 2 private final BlockingQueue<Runnable> workQueue; 3 //可重入锁。访问woker线程和相关记录信息时需要获取该锁 4 private final ReentrantLock mainLock = new ReentrantLock(); 5 //包含全部worker线程集合,Accessed only under mainLock,HashSet是非线程安全的. 6 private final HashSet<Worker> workers = new HashSet<Worker>(); 7 private final Condition termination = mainLock.newCondition(); 8 //记录最大的线程数量,Accessed only under mainLock. 9 private int largestPoolSize; 10 //完成任务的数量,Accessed only under mainLock. 11 private long completedTaskCount; 12 13 14 //以下所有程序员可以控制的参数都被声明为volatile变量,保证可见性。 15 16 //创建线程的工厂 17 private volatile ThreadFactory threadFactory; 18 //线程池饱和或关闭时的处理策略(提供了四种饱和策略) 19 private volatile RejectedExecutionHandler handler; 20 //超出corePoolSize数量的空闲线程存活时间(allowCoreThreadTimeOut=true时有效) 21 private volatile long keepAliveTime; 22 //allowCoreThreadTimeOut=false,线程不会因为空闲时间超过keepAliveTime而被停止 23 private volatile boolean allowCoreThreadTimeOut; 24 //核心线程数 25 private volatile int corePoolSize; 26 //最大线程数,此变量的最大上限为CAPACITY 27 private volatile int maximumPoolSize;
一、线程池核心线程数和最大线程数
ThreadPoolExecutor 将根据 corePoolSize (核心线程数)和 maximumPoolSize(最大线程数)设置的边界自动调整线程池大小。当新任务在方法 execute(java.lang.Runnable) 中提交时,如果运行的线程少于 corePoolSize,则创建新线程来处理请求,即使其他辅助线程是空闲的。如果运行的线程多于 corePoolSize 而少于 maximumPoolSize,则仅当队列满时才创建新线程。如果设置的 corePoolSize 和 maximumPoolSize 相同,则创建了固定大小的线程池。如果将 maximumPoolSize 设置为基本的无界值(如 Integer.MAX_VALUE),则允许池适应任意数量的并发任务。在大多数情况下,核心和最大池大小仅基于构造函数来设置,不过也可以使用 setCorePoolSize(int) 和 setMaximumPoolSize(int) 进行动态更改。
二、任务队列
workQueue是一个阻塞队列,用来存储执行的任务。所有的BlockingQueue都可用于workQueue。
如果有效的线程数小于 corePoolSize,则线程池首选添加新线程,而不进行排队。
如果有效的线程数大于等于 corePoolSize,则线程池首选将任务加入队列,而不添加新的线程。
如果队列已满,则创建新的线程,当线程数超出 maximumPoolSize 时,任务将被拒绝。
常用的三种阻塞队列的实现:
1)直接提交。SynchronousQueue是一个不存储元素的阻塞队列。每个插入操作必须等到另一个线程调用移除操作,否则插入操作一直处于阻塞状态。它将任务直接提交给线程而不存储任务。直接提交通常要求不限制 maximumPoolSizes 以避免拒绝新提交的任务。Executors.newCachedThreadPool使用了这个队列。
2)无界队列。LinkedBlockingQueue是一个基于链表结构的阻塞队列,默认的大小是Integer.MAX_VALUE。创建的线程就不会超过 corePoolSize,会使maximumPoolSize 的值无效。
3)有界队列。ArrayBlockingQueue是一个基于数组结构的有界阻塞队列。有助于防止资源耗尽,但是可能较难调整和控制。
三、饱和策略
当 Executor 已经关闭,或者 Executor 将有限边界用于最大线程和工作队列容量且已经饱和时,在方法 execute(Runnable) 中提交的新任务将被拒绝。线程池提供了4种饱和策略:
1)AbortPolicy。默认的饱和策略,直接抛出RejectedExecutionException异常。
2)CallerRunsPolicy。用调用者所在的线程来执行任务,此策略提供简单的反馈控制机制,能够减缓新任务的提交速度。
3)DiscardPolicy。直接丢弃任务。
4)DiscardOldestPolicy。如果执行程序尚未关闭,则丢弃阻塞队列中最靠前的任务,然后重试执行新任务(如果再次失败,则重复此过程)。
也可以使用自定义的 RejectedExecutionHandler 类,但需要非常小心,尤其是当策略仅用于特定容量或排队策略时。
四、threadFactory
使用 ThreadFactory 创建新线程,默认情况下在同一个 ThreadGroup 中一律使用 Executors.defaultThreadFactory() 创建线程,这些线程具有相同的 NORM_PRIORITY 优先级和非守护进程状态。通过自定义的 ThreadFactory创建新线程,可以改变线程的名称、线程组、优先级、守护进程状态等。
五、workers用来存储工作线程,注意HashSet<Worker>是非线程安全的,访问时需要获取mainLock;
六、mainLock是一个独占式可重入锁,用来保证访问workers和其他监控变量(如largestPoolSize、completedTaskCount等)的线程安全。
七、keepAliveTime为线程池的工作线程空闲后,保持存活的时间。所以如果任务很多,并且每个任务执行的时间比较短,可以调大这个时间,提高线程的利用率。allowCoreThreadTimeout变量表示是否允许核心线程超时,如果allowCoreThreadTimeOut=false,那么当线程空闲时间达到keepAliveTime时,线程会退出,直到线程数量=corePoolSize;如果allowCoreThreadTimeOut=true,那么当线程空闲时间达到keepAliveTime时,线程会退出,直到线程数量=0。
执行任务(execute)
1 public void execute(Runnable command) { 2 if (command == null) 3 throw new NullPointerException(); 4 /* 5 * Proceed in 3 steps: 6 * 7 * 1. If fewer than corePoolSize threads are running, try to 8 * start a new thread with the given command as its first 9 * task. The call to addWorker atomically checks runState and 10 * workerCount, and so prevents false alarms that would add 11 * threads when it shouldn't, by returning false. 12 * 13 * 2. If a task can be successfully queued, then we still need 14 * to double-check whether we should have added a thread 15 * (because existing ones died since last checking) or that 16 * the pool shut down since entry into this method. So we 17 * recheck state and if necessary roll back the enqueuing if 18 * stopped, or start a new thread if there are none. 19 * 20 * 3. If we cannot queue task, then we try to add a new 21 * thread. If it fails, we know we are shut down or saturated 22 * and so reject the task. 23 */ 24 int c = ctl.get(); //获取线程池的状态(runState和workerCount) 25 //如果线程数小于corePoolSize,新建一个线程执行该任务。 26 if (workerCountOf(c) < corePoolSize) { 27 if (addWorker(command, true)) 28 return; 29 c = ctl.get(); 30 } 31 //如果线程池是运行状态,并且添加任务到队列成功(队列未满) 32 if (isRunning(c) && workQueue.offer(command)) { 33 int recheck = ctl.get(); 34 //再次判断线程池的运行状态,如果不是运行状态,需要从队列删除该任务。使用拒绝策略处理该任务。 35 if (! isRunning(recheck) && remove(command)) 36 reject(command); 37 //如果线程数为0,执行addWorker方法。参数为null的原因是任务已经加入到队列,新建的线程从队列取任务执行即可。 38 else if (workerCountOf(recheck) == 0) 39 addWorker(null, false); 40 } 41 //线程池不是RUNNING状态或队列已满,尝试新建一个线程执行该任务。如果失败则拒绝该任务。 42 else if (!addWorker(command, false)) 43 reject(command); 44 }
新增线程(addWorker)
线程被封装在Worker类中。
1 //参数firstTask表示新建线程执行的第一个任务。如果firstTask为null,表示 2 //如果参数core=true,把corePoolSize作为线程数上限的判断条件;如果为false,把maximumPoolSize作为线程数上限的判断条件 3 private boolean addWorker(Runnable firstTask, boolean core) { 4 retry: 5 for (;;) { 6 int c = ctl.get(); 7 int rs = runStateOf(c); 8 /* 9 * rs >= SHUTDOWN表示不再接受新任务。 10 * 1)线程池的运行状态为SHUTDOWN;2)firstTask == null;3)阻塞队列不为空,只有这三个条件同时满足才不返回false 11 */ 12 // Check if queue empty only if necessary. 13 if (rs >= SHUTDOWN && 14 ! (rs == SHUTDOWN && 15 firstTask == null && 16 ! workQueue.isEmpty())) 17 return false; 18 19 //自旋CAS递增workerCount 20 for (;;) { 21 int wc = workerCountOf(c); 22 //如果线程数超过上限,返回false。如果参数core=true,把corePoolSize作为线程数上限的判断条件;如果为false,把maximumPoolSize作为线程数上限的判断条件 23 if (wc >= CAPACITY || 24 wc >= (core ? corePoolSize : maximumPoolSize)) 25 return false; 26 //CAS递增线程数。如果成功,跳出最外层循环;如果失败,且运行状态没有改变,继续内层循环直到成功。 27 if (compareAndIncrementWorkerCount(c)) 28 break retry; 29 //判断runState是否改变,如果改变则继续外层循环 30 c = ctl.get(); // Re-read ctl 31 if (runStateOf(c) != rs) 32 continue retry; 33 // else CAS failed due to workerCount change; retry inner loop 34 } 35 } 36 37 //走到这说明需要新建线程,且workerCount更新成功 38 //下面是新建Worker的过程。 39 boolean workerStarted = false; //新建的Worker是否启动标识 40 boolean workerAdded = false; //新建的Worker是否被添加到workers标识 41 Worker w = null; 42 try { 43 final ReentrantLock mainLock = this.mainLock; 44 w = new Worker(firstTask); //新建Worker 45 final Thread t = w.thread; 46 //什么情况下线程会为null呢?在ThreadFactory创建线程失败时可能会出现。 47 if (t != null) { 48 mainLock.lock(); //获取mainLock锁。对workers(HashSet非线程安全)和largestPoolSize更新必须加锁 49 try { 50 // Recheck while holding lock. 51 // Back out on ThreadFactory failure or if 52 // shut down before lock acquired. 53 int c = ctl.get(); 54 int rs = runStateOf(c); 55 /* 56 * 如果运行状态是RUNNING,或者运行状态是SHUTDOWN且firstTask为null,才将新建的Worker添加到workers 57 */ 58 if (rs < SHUTDOWN || 59 (rs == SHUTDOWN && firstTask == null)) { 60 if (t.isAlive()) // precheck that t is startable 61 throw new IllegalThreadStateException(); 62 workers.add(w); 63 //更新largestPoolSize,标识线程池曾经出现过的最大线程数 64 int s = workers.size(); 65 if (s > largestPoolSize) 66 largestPoolSize = s; 67 workerAdded = true; 68 } 69 } finally { 70 mainLock.unlock(); //释放mainLock锁 71 } 72 if (workerAdded) { 73 //启动线程 74 t.start(); 75 workerStarted = true; 76 } 77 } 78 } finally { 79 //新建的Worker未启动,进行失败处理 80 if (! workerStarted) 81 addWorkerFailed(w); 82 } 83 return workerStarted; 84 }
Worker类
每个线程被封装为一个Worker类实例。Worker类继承了AbstractQueuedSynchronizer,并实现了一个互斥非重入锁。Worker类同时继承了Runnable,Worker类的实例也是一个线程。
1 private final class Worker 2 extends AbstractQueuedSynchronizer 3 implements Runnable 4 { 5 /** 6 * This class will never be serialized, but we provide a 7 * serialVersionUID to suppress a javac warning. 8 */ 9 private static final long serialVersionUID = 6138294804551838833L; 10 11 /** Thread this worker is running in. Null if factory fails. */ 12 final Thread thread; //处理任务的线程 13 /** Initial task to run. Possibly null. */ 14 Runnable firstTask; //传入的任务 15 /** Per-thread task counter */ 16 volatile long completedTasks; //完成的任务数 17 18 /** 19 * Creates with given first task and thread from ThreadFactory. 20 * @param firstTask the first task (null if none) 21 */ 22 Worker(Runnable firstTask) { 23 //同步状态初始化为-1,在执行runWorker方法前禁止中断当前线程 24 setState(-1); // inhibit interrupts until runWorker 25 this.firstTask = firstTask; 26 this.thread = getThreadFactory().newThread(this); //通过ThreadFactory创建线程 27 } 28 29 /** Delegates main run loop to outer runWorker */ 30 public void run() { 31 runWorker(this); 32 } 33 34 // Lock methods 35 // 36 // The value 0 represents the unlocked state. 37 // The value 1 represents the locked state. 38 //实现了一个非重入互斥锁,state=0表示解锁状态,state=1表示加锁状态 39 protected boolean isHeldExclusively() { 40 return getState() != 0; 41 } 42 43 protected boolean tryAcquire(int unused) { 44 if (compareAndSetState(0, 1)) { 45 setExclusiveOwnerThread(Thread.currentThread()); 46 return true; 47 } 48 return false; 49 } 50 51 protected boolean tryRelease(int unused) { 52 setExclusiveOwnerThread(null); 53 setState(0); 54 return true; 55 } 56 57 public void lock() { acquire(1); } 58 public boolean tryLock() { return tryAcquire(1); } 59 public void unlock() { release(1); } 60 public boolean isLocked() { return isHeldExclusively(); } 61 62 void interruptIfStarted() { 63 Thread t; 64 if (getState() >= 0 && (t = thread) != null && !t.isInterrupted()) { 65 try { 66 t.interrupt(); 67 } catch (SecurityException ignore) { 68 } 69 } 70 } 71 }
runWorker方法
1 final void runWorker(Worker w) { 2 Thread wt = Thread.currentThread(); 3 Runnable task = w.firstTask; 4 w.firstTask = null; 5 //Worker初始化时同步状态置为-1,此处进行解锁操作目的是将同步状态置为0,允许中断。 6 w.unlock(); // allow interrupts 7 boolean completedAbruptly = true; //是否因为异常跳出循环 8 try { 9 //如果firstTask为null则通过getTask()方法从队列中获取。 10 //正常情况下,会一直执行While循环,如果队列为空,getTask()方法中会阻塞当前线程,getTask()返回null时会跳出循环 11 while (task != null || (task = getTask()) != null) { 12 w.lock(); //加Worker锁 13 // If pool is stopping, ensure thread is interrupted; 14 // if not, ensure thread is not interrupted. This 15 // requires a recheck in second case to deal with 16 // shutdownNow race while clearing interrupt 17 /* 18 * 如果线程池正在停止,要保证当前线程是中断状态 19 * 如果不是,则要保证当前线程不是中断状态 20 * STOP状态要中断线程池中的所有线程,而这里使用Thread.interrupted()来判断是否中断是为了确保在RUNNING或者SHUTDOWN状态时线程是非中断状态的,因为Thread.interrupted()方法会复位中断的状态。 21 */ 22 if ((runStateAtLeast(ctl.get(), STOP) || 23 (Thread.interrupted() && 24 runStateAtLeast(ctl.get(), STOP))) && 25 !wt.isInterrupted()) 26 wt.interrupt(); 27 try { 28 beforeExecute(wt, task); //钩子方法 29 Throwable thrown = null; 30 try { 31 task.run(); //调用任务的run方法,而不是start()方法,因为Worker本身就是一个线程类 32 } catch (RuntimeException x) { 33 thrown = x; throw x; 34 } catch (Error x) { 35 thrown = x; throw x; 36 } catch (Throwable x) { 37 thrown = x; throw new Error(x); 38 } finally { 39 afterExecute(task, thrown); //钩子方法 40 } 41 } finally { 42 task = null; 43 w.completedTasks++; 44 w.unlock(); //释放Worker锁 45 } 46 } 47 completedAbruptly = false; 48 } finally { 49 //跳出循环,执行processWorkerExit()方法 50 processWorkerExit(w, completedAbruptly); 51 } 52 }
getTask()方法
1 //如果返回null,在runWorker方法中会执行processWorkerExit,即关闭该线程。 2 private Runnable getTask() { 3 //表示上次从队列获取任务是否超时 4 boolean timedOut = false; // Did the last poll() time out? 5 6 retry: 7 for (;;) { 8 int c = ctl.get(); 9 int rs = runStateOf(c); 10 11 // Check if queue empty only if necessary. 12 // 如果rs >= STOP,或者 rs=SHUTDOWN且队列为空,此时不再接收新任务,将WorkerCount递减并返回null。 13 if (rs >= SHUTDOWN && (rs >= STOP || workQueue.isEmpty())) { 14 decrementWorkerCount(); //自旋CAS递减workerCount直到成功 15 return null; 16 } 17 18 //timed用于判断是否需要重试控制 19 boolean timed; // Are workers subject to culling? 20 21 for (;;) { 22 //allowCoreThreadTimeOut默认是false,核心线程不进行超时控制,当线程数量大于corePoolSize时需要进行超时控制 23 int wc = workerCountOf(c); 24 timed = allowCoreThreadTimeOut || wc > corePoolSize; 25 26 //如果wc <= maximumPoolSize ,且上次从队列获取任务超时或本次需要进行超时控制,则跳出内层循环。 27 //timedOut=true表示上次从队列获取元素超时,说明队列在上次获取的keepAliveTime时间内是空的。 28 //timed=true说明线程数量大于corePoolSize。 29 //所以timedOut=true和timed=true同时满足则说明当前线程已经空闲了keepAliveTime时间,并且线程池的数量大于corePoolSize。这时就需要关闭多余的空闲线程(即compareAndDecrementWorkerCount并返回null)。 30 if (wc <= maximumPoolSize && ! (timedOut && timed)) 31 break; 32 //如果线程数量大于maximumPoolSize,或者上次从队列获取任务超时且本次需要进行超时控制。需要递减WorkerCount,如果递减成功则返回null 33 if (compareAndDecrementWorkerCount(c)) 34 return null; 35 //检查线程池运行状态是否改变。如果改变,那么继续外层循环,如果未改变,那么继续内层循环。 36 c = ctl.get(); // Re-read ctl 37 if (runStateOf(c) != rs) 38 continue retry; 39 // else CAS failed due to workerCount change; retry inner loop 40 } 41 42 try { 43 Runnable r = timed ? 44 workQueue.poll(keepAliveTime, TimeUnit.NANOSECONDS) : 45 //超时方式获取,注意keepAliveTime为超出corePoolSize大小的线程的空闲存活时间 46 workQueue.take(); //阻塞方式获取,如果队列为空阻塞当前线程 47 if (r != null) 48 return r; 49 timedOut = true; //如果超时,继续循环。 50 } catch (InterruptedException retry) { 51 //如果发生中断,则将timedOut置为false,继续循环 52 timedOut = false; 53 } 54 } 55 }
processWorkerExit方法
1 private void processWorkerExit(Worker w, boolean completedAbruptly) { 2 //如果completedAbruptly=false,说明是由getTask返回null导致的,WorkerCount递减的操作已经执行。 3 //如果completedAbruptly=true,说明是由执行任务的过程中发生异常导致,需要进行WorkerCount递减的操作。 4 if (completedAbruptly) // If abrupt, then workerCount wasn't adjusted 5 decrementWorkerCount(); 6 7 final ReentrantLock mainLock = this.mainLock; 8 mainLock.lock(); 9 try { 10 completedTaskCount += w.completedTasks; 11 workers.remove(w); //从workers中删除当前worker,对workers更新需要加mainLock锁。 12 } finally { 13 mainLock.unlock(); 14 } 15 16 tryTerminate(); 17 18 //如果是异常结束(completedAbruptly=true),需要重新调用addWorker()增加一个线程,保持线程数量。 19 //如果是由getTask()返回null导致的线程结束,需要进行以下判断: 20 // 1)如果allowCoreThreadTimeOut=true且队列不为空,那么需要至少保证有一个线程。 21 // 2)如果allowCoreThreadTimeOut=false,那么需要保证线程数大于等于corePoolSize。 22 // 23 int c = ctl.get(); 24 if (runStateLessThan(c, STOP)) { 25 if (!completedAbruptly) { 26 int min = allowCoreThreadTimeOut ? 0 : corePoolSize; 27 if (min == 0 && ! workQueue.isEmpty()) 28 min = 1; 29 if (workerCountOf(c) >= min) 30 return; // replacement not needed 31 } 32 addWorker(null, false); 33 } 34 }
tryTerminate()方法
1 //根据线程池状态判断是否结束线程池 2 final void tryTerminate() { 3 for (;;) { 4 int c = ctl.get(); 5 //如果线程池运行状态是RUNNING,或者大于等于TIDYING,或者运行状态为SHUTDOWN且队列为空,则直接return。 6 if (isRunning(c) || 7 runStateAtLeast(c, TIDYING) || 8 (runStateOf(c) == SHUTDOWN && ! workQueue.isEmpty())) 9 return; 10 //如果线程数不为0,则中断一个空闲线程并return。为什么有这一步操作。 11 if (workerCountOf(c) != 0) { // Eligible to terminate 12 interruptIdleWorkers(ONLY_ONE); 13 return; 14 } 15 16 final ReentrantLock mainLock = this.mainLock; 17 mainLock.lock(); 18 try { 19 //尝试将状态设置为TIDYING状态, 20 if (ctl.compareAndSet(c, ctlOf(TIDYING, 0))) { 21 try { 22 //如果CAS成功,执行terminated()方法 23 terminated(); 24 } finally { 25 ctl.set(ctlOf(TERMINATED, 0)); 26 termination.signalAll(); 27 } 28 return; 29 } 30 } finally { 31 mainLock.unlock(); 32 } 33 // else retry on failed CAS 34 } 35 }
shutdown()方法
线程池运行状态由RUNNING到SHUTDOWN的转换。
1 public void shutdown() { 2 final ReentrantLock mainLock = this.mainLock; 3 mainLock.lock(); 4 try { 5 //安全管理,检查方法调用者是否有权限中断Worker线程 6 checkShutdownAccess(); 7 //运行状态改为SHUTDOWN 8 advanceRunState(SHUTDOWN); //自旋CAS 9 //中断空闲线程 10 interruptIdleWorkers(); 11 onShutdown(); // hook for ScheduledThreadPoolExecutor 12 } finally { 13 mainLock.unlock(); 14 } 15 //尝试结束线程池 16 tryTerminate(); 17 } 18 19 private void interruptIdleWorkers() { 20 interruptIdleWorkers(false); 21 } 22 23 private void interruptIdleWorkers(boolean onlyOne) { 24 final ReentrantLock mainLock = this.mainLock; 25 mainLock.lock(); //对workers的操作需要获取mainLock 26 try { 27 //遍历所有的线程,如果没有被中断且获取锁成功则中断线程。获取锁失败时很可能该线程正在执行任务(woker执行任务时需要对woker加锁)。 28 for (Worker w : workers) { 29 Thread t = w.thread; 30 if (!t.isInterrupted() && w.tryLock()) { 31 try { 32 t.interrupt(); 33 } catch (SecurityException ignore) { 34 } finally { 35 w.unlock(); 36 } 37 } 38 if (onlyOne) 39 break; 40 } 41 } finally { 42 mainLock.unlock(); 43 } 44 }
shutdownNow()方法
1 public List<Runnable> shutdownNow() { 2 List<Runnable> tasks; 3 final ReentrantLock mainLock = this.mainLock; 4 mainLock.lock(); 5 try { 6 checkShutdownAccess(); 7 advanceRunState(STOP); 8 //中断所有线程,即使线程正在执行任务 9 interruptWorkers(); 10 //取出队列中的任务 11 tasks = drainQueue(); 12 } finally { 13 mainLock.unlock(); 14 } 15 //尝试结束线程池 16 tryTerminate(); 17 return tasks; 18 } 19 20 private void interruptWorkers() { 21 final ReentrantLock mainLock = this.mainLock; 22 mainLock.lock(); 23 try { 24 for (Worker w : workers) 25 w.interruptIfStarted(); 26 } finally { 27 mainLock.unlock(); 28 } 29 } 30 31 private List<Runnable> drainQueue() { 32 BlockingQueue<Runnable> q = workQueue; 33 List<Runnable> taskList = new ArrayList<Runnable>(); 34 q.drainTo(taskList); 35 if (!q.isEmpty()) { 36 for (Runnable r : q.toArray(new Runnable[0])) { 37 if (q.remove(r)) 38 taskList.add(r); 39 } 40 } 41 return taskList; 42 }
FutureTask源码分析
利用FutureTask可以实现获取异步任务的返回值、取消异步任务等功能。看一下ThreadPoolExecutor的submit方法。submit方法根据任务构造一个FutureTask对象并返回,在主线程中可以根据FutureTask提供的方法进行任务取消和获取异步任务的返回值。
1 public <T> Future<T> submit(Callable<T> task) { 2 if (task == null) throw new NullPointerException(); 3 RunnableFuture<T> ftask = newTaskFor(task); 4 execute(ftask); //实际执行的任务是ftask 5 return ftask; 6 }
域
private volatile int state; //状态,新创建时状态为NEW private static final int NEW = 0; //新创建 private static final int COMPLETING = 1; //正在执行 private static final int NORMAL = 2; //正常完成 private static final int EXCEPTIONAL = 3; //执行过程中出现异常 private static final int CANCELLED = 4; //被取消 private static final int INTERRUPTING = 5; // private static final int INTERRUPTED = 6; /** The underlying callable; nulled out after running */ private Callable<V> callable; //要执行的任务 /** The result to return or exception to throw from get() */ private Object outcome; // non-volatile, protected by state reads/writes /** The thread running the callable; CASed during run() */ private volatile Thread runner; //执行callable的线程 /** Treiber stack of waiting threads */ private volatile WaitNode waiters; //Treiber算法实现的栈,用于存储等待的线程 static final class WaitNode { volatile Thread thread; volatile WaitNode next; WaitNode() { thread = Thread.currentThread(); } }
状态的转换有以下几种情况:
1)NEW -> COMPLETING -> NORMAL 正常执行并返回;
2)NEW -> COMPLETING -> EXCEPTIONAL 执行过程中出现异常;
3)NEW -> CANCELLED 执行前被取消
4)NEW -> INTERRUPTING -> INTERRUPTED 取消时被中断。
初始化
1 public FutureTask(Callable<V> callable) { 2 if (callable == null) 3 throw new NullPointerException(); 4 this.callable = callable; 5 this.state = NEW; // ensure visibility of callable 6 } 7 8 public FutureTask(Runnable runnable, V result) { 9 //由于Runnable没有返回值,通过Executors将Runnable转换为Callable。 10 this.callable = Executors.callable(runnable, result); 11 this.state = NEW; // ensure visibility of callable 12 }
执行任务-run()方法
1 public void run() { 2 //只执行state=NEW的任务。如果state!=NEW说明任务已经执行。 3 //如果state=NEW,则通过CAS将runner置为当前线程。如果失败说明其他线程已经执行。 4 if (state != NEW || 5 !UNSAFE.compareAndSwapObject(this, runnerOffset, 6 null, Thread.currentThread())) 7 return; 8 try { 9 Callable<V> c = callable; 10 if (c != null && state == NEW) { 11 V result; //任务执行结果 12 boolean ran; //任务执行期间是否发生异常 13 try { 14 result = c.call(); //执行任务 15 ran = true; 16 } catch (Throwable ex) { 17 result = null; 18 ran = false; 19 //如果发生异常,执行setException(ex) 20 setException(ex); 21 } 22 //如果正常结束,执行set(result). 23 if (ran) 24 set(result); 25 } 26 } finally { 27 // runner must be non-null until state is settled to 28 // prevent concurrent calls to run() 29 //不管任务执行是否正常,都需要将runner置为null 30 runner = null; 31 // state must be re-read after nulling runner to prevent 32 // leaked interrupts 33 //防止中断泄露,需要结合cancel方法研究 34 //如果s>=INTERRUPTING,说明状态变换为NEW -> INTERRUPTING -> INTERRUPTED,即在取消时被中断。 35 int s = state; 36 if (s >= INTERRUPTING) 37 handlePossibleCancellationInterrupt(s); 38 } 39 }
任务执行正常结束:
1 //任务正常结束,通过CAS更新state为COMPLETING,如果成功,将state更新为NORMAL,唤醒等待线程。 2 protected void set(V v) { 3 if (UNSAFE.compareAndSwapInt(this, stateOffset, NEW, COMPLETING)) { 4 outcome = v; //将运行结果result赋给outcome 5 UNSAFE.putOrderedInt(this, stateOffset, NORMAL); // final state 6 //删除和唤醒所有的等待线程 7 finishCompletion(); 8 } 9 }
任务执行时发生异常:
1 //任务执行时发生异常,通过CAS更新state为COMPLETING,如果成功,将state更新为EXCEPTIONAL,唤醒等待线程 2 protected void setException(Throwable t) { 3 if (UNSAFE.compareAndSwapInt(this, stateOffset, NEW, COMPLETING)) { 4 outcome = t; //将异常信息赋给outcome 5 UNSAFE.putOrderedInt(this, stateOffset, EXCEPTIONAL); // final state 6 finishCompletion(); 7 } 8 }
唤醒等待获取任务运行结果的线程:
1 private void finishCompletion() { 2 // assert state > COMPLETING; 3 //自旋CAS更新waiters为null直到成功 4 for (WaitNode q; (q = waiters) != null;) { 5 if (UNSAFE.compareAndSwapObject(this, waitersOffset, q, null)) { 6 for (;;) { 7 Thread t = q.thread; 8 if (t != null) { 9 q.thread = null; 10 LockSupport.unpark(t); //唤醒等待线程,WaitNode是在get方法中添加的 11 } 12 WaitNode next = q.next; 13 if (next == null) 14 break; 15 q.next = null; // unlink to help gc 16 q = next; 17 } 18 break; 19 } 20 } 21 22 done(); //hook方法,默认不执行任何操作,子类可以重写该方法完成指定的功能(例如:回调) 23 24 callable = null; // to reduce footprint 25 }
handlePossibleCancellationInterrupt方法要确保cancel(true)产生的中断发生在run或runAndReset方法执行的过程中。这里会循环的调用Thread.yield()来确保状态在cancel方法中被设置为INTERRUPTED。
1 private void handlePossibleCancellationInterrupt(int s) { 2 // It is possible for our interrupter to stall before getting a 3 // chance to interrupt us. Let's spin-wait patiently. 4 if (s == INTERRUPTING) 5 while (state == INTERRUPTING) 6 Thread.yield(); // wait out pending interrupt 7 8 // assert state == INTERRUPTED; 9 10 // We want to clear any interrupt we may have received from 11 // cancel(true). However, it is permissible to use interrupts 12 // as an independent mechanism for a task to communicate with 13 // its caller, and there is no way to clear only the 14 // cancellation interrupt. 15 // 16 // Thread.interrupted(); 17 }
获取运行结果-get()方法
1 public V get() throws InterruptedException, ExecutionException { 2 int s = state; 3 //如果state为NEW或COMPLETING,调用awaitDone方法将当前线程添加到waiters中并阻塞 4 if (s <= COMPLETING) 5 s = awaitDone(false, 0L); 6 //如果已经完成(包括正常结束或异常结束),返回 7 return report(s); 8 } 9 10 //如果超时则抛出TimeoutException异常 11 public V get(long timeout, TimeUnit unit) 12 throws InterruptedException, ExecutionException, TimeoutException { 13 if (unit == null) 14 throw new NullPointerException(); 15 int s = state; 16 if (s <= COMPLETING && 17 (s = awaitDone(true, unit.toNanos(timeout))) <= COMPLETING) 18 throw new TimeoutException(); 19 return report(s); 20 }
awaitDone方法,阻塞线程。
1 //timed参数表示是否使用超时机制 2 private int awaitDone(boolean timed, long nanos) 3 throws InterruptedException { 4 final long deadline = timed ? System.nanoTime() + nanos : 0L; 5 WaitNode q = null; 6 boolean queued = false; //是否已经入栈 7 for (;;) { 8 //若当前线程被中断,则删除q并抛出InterruptedException() 9 if (Thread.interrupted()) { 10 removeWaiter(q); 11 throw new InterruptedException(); 12 } 13 14 int s = state; 15 //如果state大于COMPLETING,表明任务已经完成,则将节点q的线程置为null并返回状态值。 16 if (s > COMPLETING) { 17 if (q != null) 18 q.thread = null; 19 return s; 20 } 21 //s==COMPLETING,说明任务已经执行完成但还没有设置最终状态。 22 //Thread.yield();让当前正在运行的线程回到可运行状态,以允许其他线程(包括当前线程)获得运行的机会。注意目的是尝试让状态改变,继续下个循环。 23 else if (s == COMPLETING) // cannot time out yet 24 Thread.yield(); 25 else if (q == null) 26 q = new WaitNode(); //新建WaitNode节点 27 //CAS添加到waiters栈,在阻塞之前先将节点q添加栈,入栈成功后queued更新为true。 28 else if (!queued) 29 queued = UNSAFE.compareAndSwapObject(this, waitersOffset, 30 q.next = waiters, q); 31 else if (timed) { 32 nanos = deadline - System.nanoTime(); 33 //如果已经过期,则删除节点q并返回 34 if (nanos <= 0L) { 35 removeWaiter(q); 36 return state; 37 } 38 LockSupport.parkNanos(this, nanos); //超时机制阻塞当前线程 39 } 40 else 41 LockSupport.park(this); //阻塞当前线程 42 } 43 } 44 45 //删除指定节点(Treiber算法实现的栈) 46 private void removeWaiter(WaitNode node) { 47 if (node != null) { 48 node.thread = null; //将线程置为null,因为下面要根据thread是否为null判断是否要把node移出 49 retry: 50 for (;;) { // restart on removeWaiter race 51 for (WaitNode pred = null, q = waiters, s; q != null; q = s) { 52 s = q.next; 53 if (q.thread != null) 54 pred = q; 55 else if (pred != null) { 56 pred.next = s; 57 if (pred.thread == null) // check for race 58 continue retry; 59 } 60 else if (!UNSAFE.compareAndSwapObject(this, waitersOffset, q, s)) 61 continue retry; 62 } 63 break; 64 } 65 } 66 }
report方法,返回运行结果或抛出异常。
1 //任务完成返回执行结果或抛出异常 2 private V report(int s) throws ExecutionException { 3 Object x = outcome; 4 //如果任务正常完成,返回执行结果 5 if (s == NORMAL) 6 return (V)x; 7 //如果s >= CANCELLED,说明任务被取消,那么就抛出CancellationException 8 if (s >= CANCELLED) 9 throw new CancellationException(); 10 //最后s==EXCEPTIONAL,任务执行时发生异常,抛出该异常 11 throw new ExecutionException((Throwable)x); 12 }
取消任务-cancel方法
试图取消对此任务的执行。如果任务已完成、或已取消,或者由于某些其他原因而无法取消,则此尝试将失败。当调用 cancel 时,如果调用成功,而此任务尚未启动,则此任务将永不运行。如果任务已经启动,则 mayInterruptIfRunning 参数确定是否应该以试图停止任务的方式来中断执行此任务的线程。
1 public boolean cancel(boolean mayInterruptIfRunning) { 2 //若state != NEW,说明任务已经启动,则直接返回失败。 3 if (state != NEW) 4 return false; 5 //如果mayInterruptIfRunning为true,要中断当前执行任务的线程。 6 if (mayInterruptIfRunning) { 7 //CAS更新state为INTERRUPTING不成功,说明state已被改变(即state != NEW),则直接返回失败。如果成功则中断正在执行任务的线程,并唤醒等待获取结果的线程。 8 if (!UNSAFE.compareAndSwapInt(this, stateOffset, NEW, INTERRUPTING)) 9 return false; 10 Thread t = runner; 11 if (t != null) 12 t.interrupt(); //中断当前线程 13 //更新state为INTERRUPTED 14 UNSAFE.putOrderedInt(this, stateOffset, INTERRUPTED); // final state 15 } 16 //mayInterruptIfRunning=flase,CAS更新state为CANCELLED,若成功则唤醒等待的线程(不中断正在执行任务的线程),若失败返回false。 17 else if (!UNSAFE.compareAndSwapInt(this, stateOffset, NEW, CANCELLED)) 18 return false; 19 finishCompletion(); 20 return true; 21 }
Executors源码解析
Executors是一个工具类,提供了公共的静态方法,例如创建默认线程工厂、创建线程池、把Runnable包装成Callable的方法等。
创建默认线程工厂
DefaultThreadFactory类
1 static class DefaultThreadFactory implements ThreadFactory { 2 private static final AtomicInteger poolNumber = new AtomicInteger(1); //线程池序号 3 private final ThreadGroup group; //线程组 4 private final AtomicInteger threadNumber = new AtomicInteger(1); //线程号 5 private final String namePrefix; 6 7 DefaultThreadFactory() { 8 SecurityManager s = System.getSecurityManager(); 9 group = (s != null) ? s.getThreadGroup() : 10 Thread.currentThread().getThreadGroup(); 11 namePrefix = "pool-" + poolNumber.getAndIncrement() + "-thread-"; 12 } 13 14 public Thread newThread(Runnable r) { 15 Thread t = new Thread(group, r, 16 namePrefix + threadNumber.getAndIncrement(), //线程名 17 0); 18 //非守护线程 19 if (t.isDaemon()) 20 t.setDaemon(false); 21 //相同的优先级 22 if (t.getPriority() != Thread.NORM_PRIORITY) 23 t.setPriority(Thread.NORM_PRIORITY); 24 return t; 25 } 26 }
创建默认工厂方法:
1 public static ThreadFactory defaultThreadFactory() { 2 return new DefaultThreadFactory(); 3 }
创建线程池
1) newFixedThreadPool方法
public static ExecutorService newFixedThreadPool(int nThreads) { return new ThreadPoolExecutor(nThreads, nThreads, 0L, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<Runnable>()); }
固定线程数的线程池,corePoolSize和 maximumPoolSize 都被设置为nThreads,keepAliveTime=0,由于corePoolSize等于maximumPoolSize,所以keepAliveTime和maximumPoolSize参数是无效的。阻塞队列是LinkedBlockingQueue,是一个无界队列。正常情况下(未执行方法shutdown()或shutdownNow()),不会调用饱和策略。
2)newSingleThreadExecutor方法
1 public static ExecutorService newSingleThreadExecutor() { 2 return new FinalizableDelegatedExecutorService 3 (new ThreadPoolExecutor(1, 1, 4 0L, TimeUnit.MILLISECONDS, 5 new LinkedBlockingQueue<Runnable>())); 6 }
单个线程的线程池,corePoolSize和maximumPoolSize都为1,其他同FixedThreadPool。能保证任务按顺序执行。
3)newCachedThreadPool方法
1 public static ExecutorService newCachedThreadPool() { 2 return new ThreadPoolExecutor(0, Integer.MAX_VALUE, 3 60L, TimeUnit.SECONDS, 4 new SynchronousQueue<Runnable>()); 5 }
线程数可改变的线程池,corePoolSize=0,maximumPoolSize=Integer.MAX_VALUE,核心线程数为0,最大线程数为CAPACITY(因为CAPACITY<Integer.MAX_VALUE).keepAliveTime=60L,意味着CachedThreadPool中的空闲线程等待新任务的最长时间为60秒,空闲线程超过60秒后将会被终止。CachedThreadPool使用没有容量的SynchronousQueue作为线程池的工作队列.这意味着,如果主线程提交任务的速度高于maximumPool中线程处理任务的速度时,CachedThreadPool会不断创建新线程。极端情况下,CachedThreadPool会因为创建过多线程而耗尽CPU和内存资源。
把Runnable包装成Callable的方法
1 public static <T> Callable<T> callable(Runnable task, T result) { 2 if (task == null) 3 throw new NullPointerException(); 4 return new RunnableAdapter<T>(task, result); 5 } 6 7 public static Callable<Object> callable(Runnable task) { 8 if (task == null) 9 throw new NullPointerException(); 10 return new RunnableAdapter<Object>(task, null); 11 }