自旋分布式锁实现

最近都在说AQS，因为手头有地方要实现一个自旋分布式锁，就不得不说一下ReentrantLock的AQS了，有关锁的通常特性能够参考线程，JVM锁整理

AQS的全称为AbstractQueuedSynchronizer，抽象队列同步器

在ReentrantLock类中，咱们来看一下加锁是怎么来实现的。

private final Sync sync;

public void lock() {
    sync.lock();
}

这个sync就是一个AQS的子类，而且是一个抽象类

abstract static class Sync extends AbstractQueuedSynchronizer

它的lock()方法是一个抽象方法

abstract void lock();

具体实现sync的是两个子类，公平锁类

static final class FairSync extends Sync

和非公平锁类

static final class NonfairSync extends Sync

这里咱们主要以非公平锁来讲明，由于咱们日常用的大部分都是非公平锁，在非公平锁中，lock()方法的实现以下

final void lock() {
    //AQS的内部方法，无锁竞争AQS中state的状态，state的初始值为0，得到锁的将0变为1
    if (compareAndSetState(0, 1))
        //竞争到state为1的将当前线程设为AQS的独家主线程
        setExclusiveOwnerThread(Thread.currentThread());
    else
        acquire(1);
}

在AbstractQueuedSynchronizer类中

private static final long stateOffset;

在静态代码块中，咱们能够看到这个stateOffset取的就是state,而且这个state是多线程可见的volatile。

stateOffset = unsafe.objectFieldOffset
    (AbstractQueuedSynchronizer.class.getDeclaredField("state"));

private volatile int state;

protected final boolean compareAndSetState(int expect, int update) {
    // See below for intrinsics setup to support this
    return unsafe.compareAndSwapInt(this, stateOffset, expect, update);
}

private transient Thread exclusiveOwnerThread;

protected final void setExclusiveOwnerThread(Thread thread) {
    exclusiveOwnerThread = thread;
}

这里unsafe.compareAndSwapInt()是用C来实现的，咱们能够用java来模拟该方法

@Slf4j
@Getter
public class GetState {
    private AtomicReference<Integer> state = new AtomicReference<>(0);
    private boolean lockState() {
        while (true) {
            if (state.compareAndSet(0,1)) {
                return true;
            }
        }
    }

    private void unlockState() {
        state.set(0);
    }

    @AllArgsConstructor
    private static class Task implements Runnable {
        private GetState getState;

        @Override
        public void run() {
            if (getState.lockState()) {
                log.info(Thread.currentThread().getName() + "获取锁");
            }
        }
    }
    public static void main(String[] args) throws InterruptedException {
        ExecutorService service = Executors.newFixedThreadPool(16);
        GetState state = new GetState();
        for (int i = 0;i < 10;i++) {
            service.execute(new Task(state));
        }
        while (state.getState().get() == 1) {
            Thread.sleep(1000);
            state.unlockState();
        }
        service.shutdown();
    }
}

打印日志(每秒打印一条)

15:35:42.953 [pool-1-thread-1] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-1获取锁
15:35:43.953 [pool-1-thread-9] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-9获取锁
15:35:44.957 [pool-1-thread-5] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-5获取锁
15:35:45.962 [pool-1-thread-2] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-2获取锁
15:35:46.962 [pool-1-thread-7] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-7获取锁
15:35:47.962 [pool-1-thread-3] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-3获取锁
15:35:48.967 [pool-1-thread-8] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-8获取锁
15:35:49.969 [pool-1-thread-6] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-6获取锁
15:35:50.970 [pool-1-thread-4] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-4获取锁
15:35:51.971 [pool-1-thread-10] INFO com.guanjian.websocket.tomic.GetState - pool-1-thread-10获取锁

Process finished with exit code 0

如今咱们能够来写一个支持自旋的分布式锁了。

public class SpinDistributedLock {
    private volatile AtomicReference<Boolean> state = new AtomicReference<>(false);

    public  boolean lock(RedisService redisService,String key,String value,int expire) {
        while (true) {
            if (state.compareAndSet(false,
                    RedisTool.tryGetDistributedLock(redisService,key,value,expire))) {
                if (state.get()) {
                    return true;
                }
            }
        }
    }

    public void unlock(RedisService redisService,String key,String value) {
        state.set(!RedisTool.releaseDistributedLock(redisService,key,value));
    }
}

常规分布式锁能够参考采用redis token,分布式锁的接口幂等性实现

如今咱们来进行一个简单的测试，先不使用分布式锁

咱们在redis中手动设置一个键count,0

127.0.0.1:6379> set count 0
OK

咱们的目的是累加这个count，但不能让其超过10

@Service
public class NoDistributedTest {
    @Autowired
    private RedisService redisService;

    private class Task implements Runnable {
        @Override
        public void run() {
            if (Integer.valueOf(redisService.get("count")) < 10) {
                redisService.incr("count");
            }
        }
    }

    @PostConstruct
    public void test() {
        ExecutorService service = Executors.newFixedThreadPool(16);
        for (int i = 0;i < 100000;i++) {
            service.execute(new Task());
        }
        service.shutdown();
    }
}

咱们启动两个进程，两个进程启动完成后，咱们再来看一下该键的值。

127.0.0.1:6379> get count
"15"

这个时候咱们能够看到，在没有锁的状况下，数量超过了10.

如今用分布式锁来进行测试。

将count键从新设为0

127.0.0.1:6379> set count 0
OK

@Slf4j
@Service
public class DistributedTest {
    private SpinDistributedLock lock = new SpinDistributedLock();
    @Autowired
    private RedisService redisService;

    private class Task implements Runnable {
        @Override
        public void run() {
            try {
                lock.lock(redisService,"countlock","countlock",3);
                log.info(Thread.currentThread().getName() + "进入锁");
                if (Integer.valueOf(redisService.get("count")) < 10) {
                    redisService.incr("count");
                }
            } finally {
                lock.unlock(redisService,"countlock","countlock");
                log.info(Thread.currentThread().getName() + "释放锁");
            }
        }
    }

    @PostConstruct
    public void test() {
        ExecutorService service = Executors.newFixedThreadPool(16);
        for (int i = 0;i < 100000;i++) {
            service.execute(new Task());
        }
        service.shutdown();
    }
}

一样启动两个进程或者更多进程，启动完成后，咱们来看一下count键的值

127.0.0.1:6379> get count
"10"

根据两个进程的日志也能够看到，两个进程会分别获取锁以及释放锁，但只有一个进程能在一个时间点内拿到分布式锁。

可是如今已经符合了咱们的需求，不让其累加超过10.