Skip to content

读写锁&&邮戳锁

1.ReentrantReadWriteLock

1.1.读写锁ReentrantReadWriteLock

读写锁:一个资源能够被多个读线程访问,或者被一个写线程访问但是不能同时存在读写线程。

它只允许读读共存,而读写和写写依然是互斥的,大多实际场景是“读/读”线程间并不存在互斥关系,只有"读/写"线程或"写/写"线程间的操作需要互斥的。因此引入ReentrantReadWriteLock 一个ReentrantReadWriteLock同时只能存在一个写锁但是可以存在多个读锁,但不能同时存在写锁和读锁(切菜还是拍蒜选一个)。也即一个资源可以被多个读操作访问―或一个写操作访问,但两者不能同时进行。只有在读多写少情景下,读写锁才具有较高的性能体现。

image-20221225145606143

1.2.锁降级

image-20221225163602324

写锁的降级,降级成为了读锁 1)如果同一个线程持有了写锁,在没有释放写锁的情况下,它还可以继续获得读锁。这就是写锁的降级,降级成为了读锁。 2)规则惯例,先获取写锁,然后获取读锁,再释放写锁的次序。 3)如果释放了写锁,那么就完全转换为读锁。

image-20221225163648565

锁降级是为了让当前线程感知到数据的变化,目的是保证数据可见性

如果有线程在读,那么写线程是无法获取写锁的,是悲观锁的策略

在ReentrantReadWriteLock中,当读锁被使用时,如果有线程尝试获取写锁,该写线程会被阻塞。所以,需要释放所有读锁,才可获取写锁。

写锁和读锁是互斥的(这里的互斥是指线程间的互斥,当前线程可以获取到写锁又获取到读锁,但是获取到了读锁不能继续获取写锁),这是因为读写锁要保持写操作的可见性。因为,如果允许读锁在被获取的情况下对写锁的获取,那么正在运行的其他读线程无法感知到当前写线程的操作。

ReentrantReadWriteLock读的过程中不允许写,只有等待线程都释放了读锁,当前线程才能获取写锁,也就是写入必须等待,这是一种悲观的读锁,人家还在读着那,你先别去写,省的数据乱。

分析StampedLock,会发现它改进之处在于: 读的过程中也允许获取写锁介入(相当牛B,读和写两个操作也让你“共享”(注意引号)),这样会导致我们读的数据就可能不一致所以,需要额外的方法来判断读的过程中是否有写入,这是一种乐观的读锁。 显然乐观锁的并发效率更高,但一旦有小概率的写入导致读取的数据不一致,需要能检测出来,再读一遍就行。

1.3.为什么要锁降级?

image-20221225164151822

锁降级确实不太贴切,明明是锁切换,在写锁释放前由写锁切换成了读锁。问题的关键其实是为什么要在锁切换前就加上读锁呢?防止释放写锁的瞬间被其他线程拿到写锁然后修改了数据,然后本线程在拿到读锁后读取数据就发生了错乱。但是,我把锁的范围加大一点不就行了吗?在写锁的范围里面完成读锁里面要干的事。缺点呢就是延长了写锁的占用时长,导致性能下降。

image-20221225164258718

1.4.锁饥饿问题

ReentrantReadWriteLock实现了读写分离,但是一旦读操作比较多的时候,想要获取写锁就变得比较困难了,假如当前1000个线程,999个读,1个写,有可能999个读取线程长时间抢到了锁,那1个写线程就悲剧了因为当前有可能会一直存在读锁,而无法获得写锁,根本没机会写。

如何缓解锁饥饿问题? 使用"公平"策略可以一定程度上缓解这个问题,但是"公平"策略是以牺牲系统吞吐量为代价的

StampedLock类的乐观读锁闪亮登场

2.邮戳锁StampedLock

2.1 StampedLock横空出世

StampedLock(也叫票据锁)是JDK1.8中新增的一个读写锁,也是对JDK1.5中的读写锁ReentrantReadWriteLock的优化。

stamp(戳记,long类型) 代表了锁的状态。当stamp返回零时,表示线程获取锁失败。并且,当释放锁或者转换锁的时候,需要传入最初获取的stamp值。

ReentrantReadWriteLock的读锁被占用的时候,其他线程尝试获取写锁的时候会被阻塞。但是,StampedLock采取乐观获取锁后,其他线程尝试获取写锁时不会被阻塞,这其实是对读锁的优化,所以,在获取乐观读锁后,还需要对结果进行校验。

ReentrantReadWriteLock 允许多个线程向时读,但是只允许一个线程写,在线程获取到写锁的时候,其他写操作和读操作都会处于阻塞状态, 读锁和写锁也是互斥的,所以在读的时候是不允许写的,读写锁比传统的synchronized速度要快很多,原因就是在于ReentrantReadWriteLock支持读并发,读读可以共享。

对短的只读代码段,使用乐观模式通常可以减少争用并提高吞吐量

2.2 ReentrantLock、ReentrantReadWriteLock、StampedLock性能比较

java
package com.bilibili.juc.rwlock;

import java.util.concurrent.CountDownLatch;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import java.util.concurrent.locks.StampedLock;

public class ReentrantReadWriteLockTest {
    static Lock lock = new ReentrantLock();
    static ReentrantReadWriteLock readWriteLock = new ReentrantReadWriteLock();
    static StampedLock stampedLock = new StampedLock();
    static int read = 1000;
    static int write = 3;
    static long mills = 10;

    public static void main(String[] args) {
        testReentrantLock();
        testReentrantReadWriteLock();
//        System.out.println("=========================");
        testStampedLock();
    }

    public static void testStampedLock() {
        ExecutorService executorService = Executors.newFixedThreadPool(100);
        ExecutorService executorServiceWrite = Executors.newFixedThreadPool(3);
        CountDownLatch latch = new CountDownLatch(read + write);
        long l = System.currentTimeMillis();
        for (int i = 0; i < read; i++) {
            executorService.execute(() -> {
//                tryOptimisticRead();
                readStampedLock();
                latch.countDown();
            });
        }
        for (int i = 0; i < write; i++) {
            executorServiceWrite.execute(() -> {
                writeStampedLock();
                latch.countDown();
//                System.out.println("时间间隔:"+(System.currentTimeMillis()-l));
            });
        }
        try {
            latch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        executorService.shutdown();
        executorServiceWrite.shutdown();
        System.out.println("testStampedLock执行耗时:" + (System.currentTimeMillis() - l));
    }

    public static void testReentrantLock() {
        ExecutorService executorService = Executors.newFixedThreadPool(100);
        ExecutorService executorServiceWrite = Executors.newFixedThreadPool(3);
        CountDownLatch latch = new CountDownLatch(read + write);
        long l = System.currentTimeMillis();
        for (int i = 0; i < read; i++) {
            executorService.execute(() -> {
                read();
                latch.countDown();
            });
        }
        for (int i = 0; i < write; i++) {
            executorServiceWrite.execute(() -> {
                write();
                latch.countDown();
            });
        }
        try {
            latch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        executorService.shutdown();
        executorServiceWrite.shutdown();
        System.out.println("testReentrantLock执行耗时:" + (System.currentTimeMillis() - l));
    }

    public static void testReentrantReadWriteLock() {
        ExecutorService executorService = Executors.newFixedThreadPool(100);
        ExecutorService executorServiceWrite = Executors.newFixedThreadPool(3);
        CountDownLatch latch = new CountDownLatch(read + write);
        long l = System.currentTimeMillis();
        for (int i = 0; i < read; i++) {
            executorService.execute(() -> {
                readLock();
                latch.countDown();
            });
        }
        for (int i = 0; i < write; i++) {
            executorServiceWrite.execute(() -> {
                writeLock();
                latch.countDown();
//                System.out.println("时间间隔:"+(System.currentTimeMillis()-l));
            });
        }
        try {
            latch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        executorService.shutdown();
        executorServiceWrite.shutdown();
        System.out.println("testReentrantReadWriteLock执行耗时:" + (System.currentTimeMillis() - l));
    }

    public static void tryOptimisticRead() {
        long stamp = stampedLock.tryOptimisticRead();
        try {
            Thread.sleep(mills);
            if (!stampedLock.validate(stamp)) {
                long readLock = stampedLock.readLock();
                try {
                } finally {
                    stampedLock.unlock(readLock);
                }
            }
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }

    public static void readStampedLock() {
        long stamp = stampedLock.readLock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            stampedLock.unlock(stamp);
        }
    }

    public static void writeStampedLock() {
        long stamp = stampedLock.writeLock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            stampedLock.unlock(stamp);
        }
    }

    public static void readLock() {
        readWriteLock.readLock().lock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            readWriteLock.readLock().unlock();
        }
    }

    public static void writeLock() {
        readWriteLock.writeLock().lock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            readWriteLock.writeLock().unlock();
        }
    }

    public static void read() {
        lock.lock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            lock.unlock();
        }
    }

    public static void write() {
        lock.lock();
        try {
            Thread.sleep(mills);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            lock.unlock();
        }
    }
}

//输出结果
testReentrantLock执行耗时:15658
testReentrantReadWriteLock执行耗时:265
testStampedLock执行耗时:237

根据执行结果可以明显看出在读多写少的情况下,ReentrantLock的性能是比较差的,而ReentrantReadWriteLock和StampedLock性能差不多相同,而StampedLock主要是为了解决ReentrantReadWriteLock可能出现的锁饥饿问题。

java
/**
 * <p>
 * StampedLock = ReentrantReadWriteLock + 读的过程中也允许获取写锁介入
 */
public class StampedLockDemo {
    static int number = 37;
    static StampedLock stampedLock = new StampedLock();

    public void write() {
        long stamp = stampedLock.writeLock();
        System.out.println(Thread.currentThread().getName() + "\t" + "写线程准备修改");
        try {
            number = number + 13;
        } finally {
            stampedLock.unlockWrite(stamp);
        }
        System.out.println(Thread.currentThread().getName() + "\t" + "写线程结束修改");
    }

    //悲观读,读没有完成时候写锁无法获得锁
    public void read() {
        long stamp = stampedLock.readLock();
        System.out.println(Thread.currentThread().getName() + "\t" + " come in readlock code block,4 seconds continue...");
        for (int i = 0; i < 4; i++) {
            //暂停几秒钟线程
            try {
                TimeUnit.SECONDS.sleep(1);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + "\t" + " 正在读取中......");
        }

        try {
            int result = number;
            System.out.println(Thread.currentThread().getName() + "\t" + " 获得成员变量值result:" + result);
            System.out.println("写线程没有修改成功,读锁时候写锁无法介入,传统的读写互斥");
        } finally {
            stampedLock.unlockRead(stamp);
        }
    }

    //乐观读,读的过程中也允许获取写锁介入
    public void tryOptimisticRead() {
        long stamp = stampedLock.tryOptimisticRead();
        int result = number;
        //故意间隔4秒钟,很乐观认为读取中没有其它线程修改过number值,具体靠判断
        System.out.println("4秒前stampedLock.validate方法值(true无修改,false有修改)" + "\t" + stampedLock.validate(stamp));
        for (int i = 0; i < 4; i++) {
            try {
                TimeUnit.SECONDS.sleep(1);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
            System.out.println(Thread.currentThread().getName() + "\t" + "正在读取... " + i + " 秒" +
                    "后stampedLock.validate方法值(true无修改,false有修改)" + "\t" + stampedLock.validate(stamp));
        }
        if (!stampedLock.validate(stamp)) {
            System.out.println("有人修改过------有写操作");
            stamp = stampedLock.readLock();
            try {
                System.out.println("从乐观读 升级为 悲观读");
                result = number;
                System.out.println("重新悲观读后result:" + result);
            } finally {
                stampedLock.unlockRead(stamp);
            }
        }
        System.out.println(Thread.currentThread().getName() + "\t" + " finally value: " + result);
    }


    public static void main(String[] args) {
        StampedLockDemo resource = new StampedLockDemo();

        /*传统版
        new Thread(() -> {
            resource.read();
        },"readThread").start();

        //暂停几秒钟线程
        try { TimeUnit.SECONDS.sleep(1); } catch (InterruptedException e) { e.printStackTrace(); }

        new Thread(() -> {
            System.out.println(Thread.currentThread().getName()+"\t"+"----come in");
            resource.write();
        },"writeThread").start();

        //暂停几秒钟线程
        try { TimeUnit.SECONDS.sleep(4); } catch (InterruptedException e) { e.printStackTrace(); }

        System.out.println(Thread.currentThread().getName()+"\t"+"number:" +number);*/

        new Thread(() -> {
            resource.tryOptimisticRead();
        }, "readThread").start();

        //暂停2秒钟线程,读过程可以写介入,演示
        //try { TimeUnit.SECONDS.sleep(2); } catch (InterruptedException e) { e.printStackTrace(); }

        //暂停6秒钟线程
        try {
            TimeUnit.SECONDS.sleep(6);
        } catch (InterruptedException e) {
            e.printStackTrace();
        }

        new Thread(() -> {
            System.out.println(Thread.currentThread().getName() + "\t" + "----come in");
            resource.write();
        }, "writeThread").start();


    }
}

//输出结果
4秒前stampedLock.validate方法值(true无修改,false有修改)	true
readThread	正在读取... 0 秒后stampedLock.validate方法值(true无修改,false有修改)	true
readThread	正在读取... 1 秒后stampedLock.validate方法值(true无修改,false有修改)	true
readThread	正在读取... 2 秒后stampedLock.validate方法值(true无修改,false有修改)	true
readThread	正在读取... 3 秒后stampedLock.validate方法值(true无修改,false有修改)	true
readThread	 finally value: 37
writeThread	----come in
writeThread	写线程准备修改
writeThread	写线程结束修改

2.3 StampedLock总结

StampedLock的特点 所有获取锁的方法,都返回一个邮戳( Stamp) , Stamp为零表示获取失败,其余都表示成功; 所有释放锁的方法,都需要一个邮戳(Stamp),这个Stamp必须是和成功获取锁时得到的Stamp一致; StampedLock是不可重入的,危险(如果一个线程已经持有了写锁,再去获取写锁的话就会造成死锁)

StampedLock有三种访问模式 Reading (读模式悲观):功能和ReentrantReadWriteLock的读锁类似 Writing(写模式):功能和ReentrantRedWriteLock的写锁类似 Optimistic reading(乐观读模式):无锁机制,类似于数据库中的乐观锁,支持读写并发,很乐观认对为读取时没人修改,假如被修改再实现升级为悲观读模式

主要API tryOptimisticRead():加乐观读锁 validate(long stamp):校验乐观读锁执行过程中有无写锁搅局

StampedLock的缺点 StampedLock 不支持重入,没有Re开头 StampedLock的悲观读锁和写锁都不支持条件变量(Condition),这个也需要注意。 使用StampedLock一定不要调用中断操作,即不要调用interrupt()方法。