问题

(1)ReentrantLock有哪些优点?

(2)ReentrantLock有哪些缺点?

(3)ReentrantLock是否可以完全替代synchronized?

简介

synchronized是Java原生提供的用于在多线程环境中保证同步的关键字,底层是通过修改对象头中的MarkWord来实现的。

ReentrantLock是Java语言层面提供的用于在多线程环境中保证同步的类,底层是通过原子更新状态变量state来实现的。

既然有了synchronized的关键字来保证同步了,为什么还要实现一个ReentrantLock类呢?它们之间有什么异同呢?

ReentrantLock VS synchronized

直接上表格:(手机横屏查看更方便)

功能ReentrantLocksynchronized可重入支持支持非公平支持(默认)支持加锁/解锁方式需要手动加锁、解锁,一般使用try..finally..保证锁能够释放手动加锁,无需刻意解锁按key锁不支持,比如按用户id加锁支持,synchronized加锁时需要传入一个对象公平锁支持,new ReentrantLock(true)不支持中断支持,lockInterruptibly()不支持尝试加锁支持,tryLock()不支持超时锁支持,tryLock(timeout, unit)不支持获取当前线程获取锁的次数支持,getHoldCount()不支持获取等待的线程支持,getWaitingThreads()不支持检测是否被当前线程占有支持,isHeldByCurrentThread()不支持检测是否被任意线程占有支持,isLocked()不支持条件锁可支持多个条件,condition.await(),condition.signal(),condition.signalAll()只支持一个,obj.wait(),obj.notify(),obj.notifyAll()对比测试

在测试之前,我们先预想一下结果,随着线程数的不断增加,ReentrantLock(fair)、ReentrantLock(unfair)、synchronized三者的效率怎样呢?

我猜测应该是ReentrantLock(unfair)> synchronized > ReentrantLock(fair)。

到底是不是这样呢?

直接上测试代码:(为了全面对比,彤哥这里把AtomicInteger和LongAdder也拿来一起对比了)

public class ReentrantLockVsSynchronizedTest { public static AtomicInteger a = new AtomicInteger(0); public static LongAdder b = new LongAdder(); public static int c = 0; public static int d = 0; public static int e = 0; public static final ReentrantLock fairLock = new ReentrantLock(true); public static final ReentrantLock unfairLock = new ReentrantLock(); public static void main(String[] args) throws InterruptedException { System.out.println("-------------------------------------"); testAll(1, 100000); System.out.println("-------------------------------------"); testAll(2, 100000); System.out.println("-------------------------------------"); testAll(4, 100000); System.out.println("-------------------------------------"); testAll(6, 100000); System.out.println("-------------------------------------"); testAll(8, 100000); System.out.println("-------------------------------------"); testAll(10, 100000); System.out.println("-------------------------------------"); testAll(50, 100000); System.out.println("-------------------------------------"); testAll(100, 100000); System.out.println("-------------------------------------"); testAll(200, 100000); System.out.println("-------------------------------------"); testAll(500, 100000); System.out.println("-------------------------------------");// testAll(1000, 1000000); System.out.println("-------------------------------------"); testAll(500, 10000); System.out.println("-------------------------------------"); testAll(500, 1000); System.out.println("-------------------------------------"); testAll(500, 100); System.out.println("-------------------------------------"); testAll(500, 10); System.out.println("-------------------------------------"); testAll(500, 1); System.out.println("-------------------------------------"); } public static void testAll(int threadCount, int loopCount) throws InterruptedException { testAtomicInteger(threadCount, loopCount); testLongAdder(threadCount, loopCount); testSynchronized(threadCount, loopCount); testReentrantLockUnfair(threadCount, loopCount);// testReentrantLockFair(threadCount, loopCount); } public static void testAtomicInteger(int threadCount, int loopCount) throws InterruptedException { long start = System.currentTimeMillis(); CountDownLatch countDownLatch = new CountDownLatch(threadCount); for (int i = 0; i < threadCount; i++) { new Thread(() -> { for (int j = 0; j < loopCount; j++) { a.incrementAndGet(); } countDownLatch.countDown(); }).start(); } countDownLatch.await(); System.out.println("testAtomicInteger: result=" + a.get() + ", threadCount=" + threadCount + ", loopCount=" + loopCount + ", elapse=" + (System.currentTimeMillis() - start)); } public static void testLongAdder(int threadCount, int loopCount) throws InterruptedException { long start = System.currentTimeMillis(); CountDownLatch countDownLatch = new CountDownLatch(threadCount); for (int i = 0; i < threadCount; i++) { new Thread(() -> { for (int j = 0; j < loopCount; j++) { b.increment(); } countDownLatch.countDown(); }).start(); } countDownLatch.await(); System.out.println("testLongAdder: result=" + b.sum() + ", threadCount=" + threadCount + ", loopCount=" + loopCount + ", elapse=" + (System.currentTimeMillis() - start)); } public static void testReentrantLockFair(int threadCount, int loopCount) throws InterruptedException { long start = System.currentTimeMillis(); CountDownLatch countDownLatch = new CountDownLatch(threadCount); for (int i = 0; i < threadCount; i++) { new Thread(() -> { for (int j = 0; j < loopCount; j++) { fairLock.lock(); // 消除try的性能影响// try { c++;// } finally { fairLock.unlock();// } } countDownLatch.countDown(); }).start(); } countDownLatch.await(); System.out.println("testReentrantLockFair: result=" + c + ", threadCount=" + threadCount + ", loopCount=" + loopCount + ", elapse=" + (System.currentTimeMillis() - start)); } public static void testReentrantLockUnfair(int threadCount, int loopCount) throws InterruptedException { long start = System.currentTimeMillis(); CountDownLatch countDownLatch = new CountDownLatch(threadCount); for (int i = 0; i < threadCount; i++) { new Thread(() -> { for (int j = 0; j < loopCount; j++) { unfairLock.lock(); // 消除try的性能影响// try { d++;// } finally { unfairLock.unlock();// } } countDownLatch.countDown(); }).start(); } countDownLatch.await(); System.out.println("testReentrantLockUnfair: result=" + d + ", threadCount=" + threadCount + ", loopCount=" + loopCount + ", elapse=" + (System.currentTimeMillis() - start)); } public static void testSynchronized(int threadCount, int loopCount) throws InterruptedException { long start = System.currentTimeMillis(); CountDownLatch countDownLatch = new CountDownLatch(threadCount); for (int i = 0; i < threadCount; i++) { new Thread(() -> { for (int j = 0; j < loopCount; j++) { synchronized (ReentrantLockVsSynchronizedTest.class) { e++; } } countDownLatch.countDown(); }).start(); } countDownLatch.await(); System.out.println("testSynchronized: result=" + e + ", threadCount=" + threadCount + ", loopCount=" + loopCount + ", elapse=" + (System.currentTimeMillis() - start)); }}

运行这段代码,你会发现结果大大出乎意料,真的是不测不知道,一测吓一跳,运行后发现以下规律:

随着线程数的不断增加,synchronized的效率竟然比ReentrantLock非公平模式要高!

彤哥的电脑上大概是高3倍左右,我的运行环境是4核8G,java版本是8,请大家一定要在自己电脑上运行一下,并且最好能给我反馈一下。

彤哥又使用Java7及以下的版本运行了,发现在Java7及以下版本中synchronized的效率确实比ReentrantLock的效率低一些。

总结

(1)synchronized是Java原生关键字锁;

(2)ReentrantLock是Java语言层面提供的锁;

(3)ReentrantLock的功能非常丰富,解决了很多synchronized的局限性;

(4)至于在非公平模式下,ReentrantLock与synchronized的效率孰高孰低,彤哥给出的结论是随着Java版本的不断升级,synchronized的效率只会越来越高;

彩蛋

既然ReentrantLock的功能更丰富,而且效率也不低,我们是不是可以放弃使用synchronized了呢?

答:我认为不是。因为synchronized是Java原生支持的,随着Java版本的不断升级,Java团队也是在不断优化synchronized,所以我认为在功能相同的前提下,最好还是使用原生的synchronized关键字来加锁,这样我们就能获得Java版本升级带来的免费的性能提升的空间。

另外,在Java8的ConcurrentHashMap中已经把ReentrantLock换成了synchronized来分段加锁了,这也是Java版本不断升级带来的免费的synchronized的性能提升。

推荐阅读

死磕 java同步系列之ReentrantLock源码解析(二)——条件锁

死磕 java同步系列之ReentrantLock源码解析(一)——公平锁、非公平锁

死磕 java同步系列之AQS起篇

死磕 java同步系列之自己动手写一个锁Lock

死磕 java魔法类之Unsafe解析

死磕 java同步系列之JMM(Java Memory Model)

死磕 java同步系列之volatile解析

死磕 java同步系列之synchronized解析

欢迎关注我的公众号“彤哥读源码”,查看更多源码系列文章, 与彤哥一起畅游源码的海洋。