ReentrantLock(重入锁)以及公平性

简介

ReentrantLock的实现不仅可以替代隐式的synchronized关键字,而且能够提供超过关键字本身的多种功能。
这里提到一个锁获取的公平性问题,如果在绝对时间上,先对锁进行获取的请求一定被先满足,那么这个锁是公平的,反之,是不公平的,也就是说等待时间最长的线程最有机会获取锁,也可以说锁的获取是有序的。ReentrantLock这个锁提供了一个构造函数,能够控制这个锁是否是公平的。
而锁的名字也是说明了这个锁具备了重复进入的可能,也就是说能够让当前线程多次的进行对锁的获取操作,这样的最大次数限制是Integer.MAX_VALUE,约21亿次左右。
事实上公平的锁机制往往没有非公平的效率高,因为公平的获取锁没有考虑到操作系统对线程的调度因素,这样造成JVM对于等待中的线程调度次序和操作系统对线程的调度之间的不匹配。对于锁的快速且重复的获取过程中,连续获取的概率是非常高的,而公平锁会压制这种情况,虽然公平性得以保障,但是响应比却下降了,但是并不是任何场景都是以TPS作为唯一指标的,因为公平锁能够减少“饥饿”发生的概率,等待越久的请求越是能够得到优先满足。

实现分析

在ReentrantLock中,对于公平和非公平的定义是通过对同步器AbstractQueuedSynchronizer的扩展加以实现的,也就是在tryAcquire的实现上做了语义的控制。

非公平的获取语义:

01 final boolean nonfairTryAcquire(int acquires) {
02     final Thread current = Thread.currentThread();
03     int c = getState();
04     if (c == 0) {
05         if (compareAndSetState(0, acquires)) {
06             setExclusiveOwnerThread(current);
07             return true;
08         }
09     } else if (current == getExclusiveOwnerThread()) {
10         int nextc = c + acquires;
11                 if (nextc < 0) // overflow
12             throw new Error("Maximum lock count exceeded");
13         setState(nextc);
14         return true;
15     }
16     return false;
17 }

上述逻辑主要包括:

  • 如果当前状态为初始状态,那么尝试设置状态;
  • 如果状态设置成功后就返回;
  • 如果状态被设置,且获取锁的线程又是当前线程的时候,进行状态的自增;
  • 如果未设置成功状态且当前线程不是获取锁的线程,那么返回失败。

公平的获取语义:

01 protected final boolean tryAcquire(int acquires) {
02     final Thread current = Thread.currentThread();
03     int c = getState();
04     if (c == 0) {
05         if (!hasQueuedPredecessors() && compareAndSetState(0, acquires)) {
06             setExclusiveOwnerThread(current);
07             return true;
08         }
09     } else if (current == getExclusiveOwnerThread()) {
10         int nextc = c + acquires;
11         if (nextc < 0)
12             throw new Error("Maximum lock count exceeded");
13         setState(nextc);
14         return true;
15     }
16     return false;
17 }

上述逻辑相比较非公平的获取,仅加入了当前线程(Node)之前是否有前置节点在等待的判断。hasQueuedPredecessors()方法命名有些歧义,其实应该是currentThreadHasQueuedPredecessors()更为妥帖一些,也就是说当前面没有人排在该节点(Node)前面时候队且能够设置成功状态,才能够获取锁。

释放语义:

01 protected final boolean tryRelease(int releases) {
02     int c = getState() - releases;
03     if (Thread.currentThread() != getExclusiveOwnerThread())
04         throw new IllegalMonitorStateException();
05     boolean free = false;
06     if (c == 0) {
07         free = true;
08         setExclusiveOwnerThread(null);
09     }
10     setState(c);
11     return free;
12 }

上述逻辑主要主要计算了释放状态后的值,如果为0则完全释放,返回true,反之仅是设置状态,返回false。
下面将主要的笔墨放在公平性和非公平性上,首先看一下二者测试的对比:
测试用例如下:

01 public class ReentrantLockTest {
02     private static Lock fairLock = new ReentrantLock(true);
03     private static Lock unfairLock = new ReentrantLock();
04  
05     @Test
06     public void fair() {
07         System.out.println("fair version");
08         for (int i = 0; i < 5; i++) {
09             Thread thread = new Thread(new Job(fairLock));
10             thread.setName("" + i);
11             thread.start();
12         }
13  
14         try {
15             Thread.sleep(5000);
16         } catch (InterruptedException e) {
17             e.printStackTrace();
18         }
19     }
20  
21     @Test
22     public void unfair() {
23         System.out.println("unfair version");
24         for (int i = 0; i < 5; i++) {
25             Thread thread = new Thread(new Job(unfairLock));
26             thread.setName("" + i);
27             thread.start();
28         }
29  
30         try {
31             Thread.sleep(5000);
32         } catch (InterruptedException e) {
33             e.printStackTrace();
34         }
35     }
36  
37     private static class Job implements Runnable {
38         private Lock lock;
39         public Job(Lock lock) {
40             this.lock = lock;
41         }
42  
43         @Override
44         public void run() {
45             for (int i = 0; i < 5; i++) {
46                 lock.lock();
47                 try {
48                     System.out.println("Lock by:"
49                             + Thread.currentThread().getName());
50                 } finally {
51                     lock.unlock();
52                 }
53             }
54         }
55     }
56 }

调用非公平的测试方法,返回结果(部分):
unfair version
Lock by:0
Lock by:0
Lock by:2
Lock by:2
Lock by:2
Lock by:2
Lock by:2
Lock by:0
Lock by:0
Lock by:0
Lock by:1
Lock by:1
Lock by:1
调用公平的测试方法,返回结果:
fair version
Lock by:0
Lock by:1
Lock by:0
Lock by:2
Lock by:3
Lock by:4
Lock by:1
Lock by:0
Lock by:2
Lock by:3
Lock by:4
仔细观察返回的结果(其中每个数字代表一个线程),非公平的结果一个线程连续获取锁的情况非常多,而公平的结果连续获取的情况基本没有。那么在一个线程获取了锁的那一刻,究竟锁的公平性会导致锁有什么样的处理逻辑呢?
通过之前的同步器(AbstractQueuedSynchronizer)的介绍,在锁上是存在一个等待队列,sync队列,我们通过复写ReentrantLock的获取当前锁的sync队列,输出在ReentrantLock被获取时刻,当前的sync队列的状态。
修改测试如下:

01 public class ReentrantLockTest {
02     private static Lock fairLock = new ReentrantLock2(true);
03     private static Lock unfairLock = new ReentrantLock2();
04     @Test
05     public void fair() {
06         System.out.println("fair version");
07         for (int i = 0; i < 5; i++) {
08             Thread thread = new Thread(new Job(fairLock)) {
09                 public String toString() {
10                     return getName();
11                 }
12             };
13             thread.setName("" + i);
14             thread.start();
15         }
16         // sleep 5000ms
17     }
18  
19     @Test
20     public void unfair() {
21         System.out.println("unfair version");
22         for (int i = 0; i < 5; i++) {
23             Thread thread = new Thread(new Job(unfairLock)) {
24                 public String toString() {
25                     return getName();
26                 }
27             };
28             thread.setName("" + i);
29             thread.start();
30         }
31         // sleep 5000ms
32     }
33  
34     private static class Job implements Runnable {
35         private Lock lock;
36  
37         public Job(Lock lock) {
38             this.lock = lock;
39         }
40  
41         @Override
42         public void run() {
43             for (int i = 0; i < 5; i++) {
44                 lock.lock();
45                 try {
46                     System.out.println("Lock by:"
47                             + Thread.currentThread().getName() + " and "
48                             + ((ReentrantLock2) lock).getQueuedThreads()
49                             + " waits.");
50                 } finally {
51                     lock.unlock();
52                 }
53             }
54         }
55     }
56  
57     private static class ReentrantLock2 extends ReentrantLock {
58         // Constructor Override
59  
60         private static final long serialVersionUID = 1773716895097002072L;
61  
62         public Collection<Thread> getQueuedThreads() {
63             return super.getQueuedThreads();
64         }
65     }
66 }

上述逻辑主要是通过构造ReentrantLock2用来输出在sync队列中的线程内容,而且每个线程的toString方法被重写,这样当一个线程获取到锁时,sync队列里的内容也就可以得知了,运行结果如下:
调用非公平方法,返回结果:
unfair version
Lock by:0 and [] waits.
Lock by:0 and [] waits.
Lock by:3 and [2, 1] waits.
Lock by:3 and [4, 2, 1] waits.
Lock by:3 and [4, 2, 1] waits.
Lock by:3 and [0, 4, 2, 1] waits.
Lock by:3 and [0, 4, 2, 1] waits.
Lock by:1 and [0, 4, 2] waits.
Lock by:1 and [0, 4, 2] waits.
调用公平方法,返回结果:
fair version
Lock by:0 and [] waits.
Lock by:1 and [0, 4, 3, 2] waits.
Lock by:2 and [1, 0, 4, 3] waits.
Lock by:3 and [2, 1, 0, 4] waits.
Lock by:4 and [3, 2, 1, 0] waits.
Lock by:0 and [4, 3, 2, 1] waits.
Lock by:1 and [0, 4, 3, 2] waits.
Lock by:2 and [1, 0, 4, 3] waits.
可以明显看出,在非公平获取的过程中,“插队”现象非常严重,后续获取锁的线程根本不顾及sync队列中等待的线程,而是能获取就获取。反观公平获取的过程,锁的获取就类似线性化的,每次都由sync队列中等待最长的线程(链表的第一个,sync队列是由尾部结点添加,当前输出的sync队列是逆序输出)获取锁。一个 hasQueuedPredecessors方法能够获得公平性的特性,这点实际上是由AbstractQueuedSynchronizer来完成的,看一下acquire方法:

1 public final void acquire(int arg) {
2     if (!tryAcquire(arg) && acquireQueued(addWaiter(Node.EXCLUSIVE), arg))
3         selfInterrupt();
4 }

可以看到,如果获取状态和在sync队列中排队是短路的判断,也就是说如果tryAcquire成功,那么是不会进入sync队列的,可以通过下图来深刻的认识公平性和AbstractQueuedSynchronizer的获取过程。
非公平的,或者说默认的获取方式如下图所示:
ReentrantLock(重入锁)以及公平性_第1张图片
对于状态的获取,可以快速的通过tryAcquire的成功,也就是黄色的Fast路线,也可以由于tryAcquire的失败,构造节点,进入sync队列中排序后再次获取。因此可以理解为Fast就是一个快速通道,当例子中的线程释放锁之后,快速的通过Fast通道再次获取锁,就算当前sync队列中有排队等待的线程也会被忽略。这种模式,可以保证进入和退出锁的吞吐量,但是sync队列中过早排队的线程会一直处于阻塞状态,造成“饥饿”场景。
而公平性锁,就是在tryAcquire的调用中顾及当前sync队列中的等待节点(废弃了Fast通道),也就是任意请求都需要按照sync队列中既有的顺序进行,先到先得。这样很好的确保了公平性,但是可以从结果中看到,吞吐量就没有非公平的锁高了。

你可能感兴趣的:(java)