【分布式锁】07-Zookeeper实现分布式锁:Semaphore、读写锁实现原理

前言

前面已经讲解了Zookeeper可重入锁的实现原理,自己对分布式锁也有了更深的认知。

我在公众号中发了一个疑问,相比于Redis来说,Zookeeper的实现方式要更好一些,即便Redis作者实现了RedLock算法来解决Redis集群模式下分布式锁的弊端,但Redis实现的分布式锁仍然不是那么完美。

比如有5台Redis集群,按照n/2 + 1代表获取锁成功,如果客户端A此时获取锁,Redis集群(1,2,3)返回成功,客户端A获取锁成功。

此时Redis 1 master宕机,切换到slave,而slave并未来得及同步客户端A加锁成功的信息到slave。

客户端B获取锁,Redis集群(1,4,5)返回成功,客户端B仍然可以成功获取锁。

即使如此,为何在实际生产项目中分布式锁大多还是由Redis来完成?

这一点我仍然有些疑惑,我接触过的公司和项目都普遍用Redis来实现分布式锁。

这里就不再纠结了,接着继续学习Zookeeper剩下几个实现分布式锁的组件吧。

Semaphore实现原理

前面已经讲过Redisson中Semaphore的实现原理( 【分布式锁】05-使用Redisson中Semaphore和CountDownLatch原理 ),现在学习下ZK中Semaphore是如何实现的

Semaphore 使用案例

使用示例很简单,Curator官网上有对应代码,使用 InterProcessSemaphoreV2 类即可,代码如下:

/**
*  Zookeeper分布式锁测试代码
*
 * @author wangmeng
 * @date 2020/03/30 18:59
 */
public class Application {

    /** Zookeeper info */
    private static final String ZK_ADDRESS = "YourZkIP:2181";
    private static final String ZK_LOCK_PATH = "/locks/lock_01";
    private static final String ZK_SEMAPHORE_LOCK_PATH = "/semaphore/semaphore_01";

    public static void main(String[] args) throws InterruptedException {
        // 1.Connect to zk
        CuratorFramework client = CuratorFrameworkFactory.newClient(
                ZK_ADDRESS,
                new RetryNTimes(10, 5000)
        );
        client.start();
        System.out.println("zk client start successfully!");

        Thread t1 = new Thread(() -> {
            testSemaphore(client);
        }, "t1");
        Thread t2 = new Thread(() -> {
            testSemaphore(client);
        }, "t2");
        Thread t3 = new Thread(() -> {
            testSemaphore(client);
        }, "t3");

        t1.start();
        t2.start();
        t3.start();
    }


    /**
     * 测试Semaphore
     */
    private static void testSemaphore(CuratorFramework client) {
        InterProcessSemaphoreV2 semaphore = new InterProcessSemaphoreV2(client, ZK_SEMAPHORE_LOCK_PATH, 2);
        try {
            Lease lease = semaphore.acquire();
            System.out.println(Thread.currentThread().getName() + " hold lock");
            Thread.sleep(5000L);
            semaphore.returnLease(lease);
            System.out.println(Thread.currentThread().getName() + " release lock");
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

打印结果为:

image.png

因为设置的只允许最多2个客户端同时获取锁。

从效果上看t3和t2同时获取到了锁,接着t3释放了锁后t1才获取锁。

Semaphore加锁源码解析

源码面前出真知,我们直接看下源码:

public class InterProcessSemaphoreV2 {

    private static final String LOCK_PARENT = "locks";
    private static final String LEASE_PARENT = "leases";
    private static final String LEASE_BASE_NAME = "lease-";

    public Collection acquire(int qty, long time, TimeUnit unit) throws Exception
    {
        long startMs = System.currentTimeMillis();
        boolean hasWait = (unit != null);
        long waitMs = hasWait ? TimeUnit.MILLISECONDS.convert(time, unit) : 0;

        Preconditions.checkArgument(qty > 0, "qty cannot be 0");

        ImmutableList.Builder builder = ImmutableList.builder();
        boolean success = false;
        try
        {
            while ( qty-- > 0 )
            {
                int retryCount = 0;
                long startMillis = System.currentTimeMillis();
                boolean isDone = false;
                while ( !isDone )
                {
                    switch ( internalAcquire1Lease(builder, startMs, hasWait, waitMs) )
                    {
                        case CONTINUE:
                        {
                            isDone = true;
                            break;
                        }

                        // 省略其他分支逻辑
                    }
                }
            }
            success = true;
        }
        finally
        {
            if ( !success )
            {
                returnAll(builder.build());
            }
        }

        return builder.build();
    }

    private InternalAcquireResult internalAcquire1Lease(ImmutableList.Builder builder, long startMs, boolean hasWait, long waitMs) throws Exception
    {
        if ( client.getState() != CuratorFrameworkState.STARTED )
        {
            return InternalAcquireResult.RETURN_NULL;
        }

        if ( hasWait )
        {
            long thisWaitMs = getThisWaitMs(startMs, waitMs);
            if ( !lock.acquire(thisWaitMs, TimeUnit.MILLISECONDS) )
            {
                return InternalAcquireResult.RETURN_NULL;
            }
        }
        else
        {
            lock.acquire();
        }
        try
        {
            PathAndBytesable createBuilder = client.create().creatingParentsIfNeeded().withProtection().withMode(CreateMode.EPHEMERAL_SEQUENTIAL);
            String path = (nodeData != null) ? createBuilder.forPath(ZKPaths.makePath(leasesPath, LEASE_BASE_NAME), nodeData) : createBuilder.forPath(ZKPaths.makePath(leasesPath, LEASE_BASE_NAME));
            String nodeName = ZKPaths.getNodeFromPath(path);
            builder.add(makeLease(path));

            synchronized(this)
            {
                for(;;)
                {
                    List children = client.getChildren().usingWatcher(watcher).forPath(leasesPath);
                    if ( !children.contains(nodeName) )
                    {
                        log.error("Sequential path not found: " + path);
                        return InternalAcquireResult.RETRY_DUE_TO_MISSING_NODE;
                    }

                    if ( children.size() <= maxLeases )
                    {
                        break;
                    }
                    if ( hasWait )
                    {
                        long thisWaitMs = getThisWaitMs(startMs, waitMs);
                        if ( thisWaitMs <= 0 )
                        {
                            return InternalAcquireResult.RETURN_NULL;
                        }
                        wait(thisWaitMs);
                    }
                    else
                    {
                        wait();
                    }
                }
            }
        }
        finally
        {
            lock.release();
        }
        return InternalAcquireResult.CONTINUE;
    }
}

代码有点长,我们一点点分析,我们以客户端A、B、C同时进入获取锁逻辑来举例,这里Semaphore最大可允许2个客户端同时获取锁。

  1. 三个客户端同时进入switch逻辑,执行 internalAcquire1Lease() 方法
  2. internalAcquire1Lease() 方法中,先使用 lock.acquire() 执行加锁逻辑,这个lock是我们上一章讲的可重入锁逻辑,不再赘述
  3. 这个 lock 是哪里初始化的呢?在 InterProcessSemaphoreV2 构造函数中:
    lock = new InterProcessMutex(client, ZKPaths.makePath(path, LOCK_PARENT));
    this.maxLeases = (count != null) ? count.getCount() : maxLeases;
  4. 注意lock的path为: /semaphore/semaphore_01/locks , maxLeases为传入的3
  5. 此时客户端A、B、C执行 lock.acquire() 只会有一个可以成功获取锁,其他两个客户端会 wait()

到了这里,Zookeeper中就会有三条类似于:

/semaphores/semaphore_01/locks/_c_a9302e20-de9c-4356-923a-274664d7676c-lock-0000000001 的数据

接着客户端A继续往下执行,具体逻辑如图:

  1. 首先是客户端A创建一个/locks/lock-xxxx01节点,获取锁成功过
  2. 接着创建临时顺序节点/leases/lease-xxxx01
  3. 判断/leases目录下节点数量(数量为1)是否小于等于maxLeases(maxLeases=2)
  4. 如果成功则退出循环,释放/locks加的锁,返回 InternalAcquireResult.CONTINUE ,状态,执行 lock.release() 通知客户端B、C争抢 /locks 节点下的锁
  5. 此时如果客户端B抢到锁,然后同样创建/leases/lease-xxxx02,
  6. 判断/leases目录下节点数量(数量为2)是否小于等于maxLeases(maxLeases=2)
  7. 客户端B也退出循环,返回 InternalAcquireResult.CONTINUE ,接着客户端C来获取锁
  8. 客户端C执行时,判断/leases目录下节点数量(数量为3)是否小于等于maxLeases(maxLeases=2)

此时客户端C会进入到 wait() 方法,直到客户端A或者客户端B释放 leases 节点下锁时才会重试获取锁。

返回 InternalAcquireResult.CONTINUE 后,就标志获取锁成功。

Semaphore释放锁源码分析

我们直接看代码,释放锁代码很简单:

/**
 * Convenience method. Closes the lease
 *
 * @param lease lease to close
 */
public void returnLease(Lease lease)
{
    Closeables.closeQuietly(lease);
}

一路跟下去,可以看到 closeQuietly 实现方法:

image.png

最后用到 Lease 中的 close() 方法,删除创建的 /leases/lease-xxxx 节点数据,然后通知其他节点客户端,使用 notifyAll()

ZK-Semaphore总结

一张图总结下:

05_Zookeeper中Semaphore实现原理 _1_.jpg

Zookeeper 非重入锁实现原理

之前听小伙伴说过一个面试题,请说出你所知道的非重入锁?

在脑子中搜索JDK中非重入锁?好像没有?

Zookeeper中提供了一个非重入锁的实现方式,实现原理使用Semaphore,最大允许1个客户端获取锁

按理说JDK中的Semaphore也可以实现此功能,哈哈,感觉自己被忽悠了,接着还是勉为其难的看下ZK中”非重入锁”的实现方式吧:

使用示例

/**
 * 测试非重入锁
 */
private static void testSemaphoreMutex(CuratorFramework client) {
    InterProcessSemaphoreMutex semaphoreMutex = new InterProcessSemaphoreMutex(client, ZK_SEMAPHORE_LOCK_PATH);
    try {
        semaphoreMutex.acquire();
        Thread.sleep(5000L);
        semaphoreMutex.release();
    } catch (Exception e) {
        e.printStackTrace();
    }
}

源码分析

image.png

实际上就是设置 maxLeases 为1,原理同上面的 Semaphore 源码分析

Zookeeper读写锁原理

之前在Redisson中已经见过它对读写锁的实现,分别举例了读读、写写、读写、写读这几种场景锁的互斥性以及可重入性,这里也采用类似的场景分析。

读写锁使用案例

直接看案例,可以针对案例修改几种场景进行测试:

/**
 * @author wangmeng
 * @date 2020/03/30 18:59
 */
public class Application {

    /** Zookeeper info */
    private static final String ZK_ADDRESS = "yourZkIP:2181";
    private static final String ZK_LOCK_PATH = "/locks/lock_01";
    private static final String ZK_SEMAPHORE_LOCK_PATH = "/semaphore/semaphore_01";
    private static final String ZK_READ_WRITE_LOCK_PATH = "/readwrite/readwrite_01";

    public static void main(String[] args) throws InterruptedException {
        // 1.Connect to zk
        CuratorFramework client = CuratorFrameworkFactory.newClient(
                ZK_ADDRESS,
                new RetryNTimes(10, 5000)
        );
        client.start();
        System.out.println("zk client start successfully!");

        Thread t1 = new Thread(() -> {
            testReadWriteLock(client);
        }, "t1");
        Thread t2 = new Thread(() -> {
            testReadWriteLock(client);
        }, "t2");

        t1.start();
        t2.start();
    }


    /**
     * 测试读写锁
     */
    private static void testReadWriteLock(CuratorFramework client) {
        InterProcessReadWriteLock readWriteLock = new InterProcessReadWriteLock(client, ZK_READ_WRITE_LOCK_PATH);
        try {
            // 获取读锁
            InterProcessMutex readLock = readWriteLock.readLock();
            readLock.acquire();
            System.out.println(Thread.currentThread().getName() + " hold read lock");
            Thread.sleep(5000);
            readLock.release();
            System.out.println(Thread.currentThread().getName() + " release read lock");

            // 获取写锁
            InterProcessMutex writeLock = readWriteLock.writeLock();
            writeLock.acquire();
            System.out.println(Thread.currentThread().getName() + " hold write lock");
            Thread.sleep(5000);
            writeLock.release();
            System.out.println(Thread.currentThread().getName() + " release write lock");
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

运行后结果:

image.png

从结果可以看出来: 读读不互斥、 写写互斥

读写锁源码解析

首先看下 InterProcessReadWrite 的构造函数:

public class InterProcessReadWriteLock {    
    public InterProcessReadWriteLock(CuratorFramework client, String basePath)
    {
        writeMutex = new InternalInterProcessMutex
        (
            client,
            basePath,
            WRITE_LOCK_NAME,
            1,
            new SortingLockInternalsDriver()
            {
                @Override
                public PredicateResults getsTheLock(CuratorFramework client, List children, String sequenceNodeName, int maxLeases) throws Exception
                {
                    return super.getsTheLock(client, children, sequenceNodeName, maxLeases);
                }
            }
        );

        readMutex = new InternalInterProcessMutex
        (
            client,
            basePath,
            READ_LOCK_NAME,
            Integer.MAX_VALUE,
            new SortingLockInternalsDriver()
            {
                @Override
                public PredicateResults getsTheLock(CuratorFramework client, List children, String sequenceNodeName, int maxLeases) throws Exception
                {
                    return readLockPredicate(children, sequenceNodeName);
                }
            }
        );
    }
}

因为ZK中的读写锁底层也是基于第一讲中 InterProcessMutex.internalLock() 去实现的,所以 InterProcessReadWriteLock 读锁和写锁分别初始化了 maxLeases 及重写了 getsTheLock() 方法,这个方法是判断是否可以获取锁的核心代码,类似于:

int ourIndex = children.indexOf(sequenceNodeName);
boolean getsTheLock = ourIndex < maxLeases;

不清楚的可以回头看看: 【分布式锁】06-Zookeeper实现分布式锁:可重入锁源码分析

另外写锁和读锁的path会有区别:

private static final String READ_LOCK_NAME  = "__READ__";
private static final String WRITE_LOCK_NAME = "__WRIT__";

写锁的maxLeases是1,加了写锁就不允许再加其他读锁(但可重入加写锁和读锁)

读锁的maxLeases是Integer.MAX_VALUE,读读锁不互斥

读读互斥及重入

查看读锁中判断获取锁成功的核心逻辑:

private PredicateResults readLockPredicate(List children, String sequenceNodeName) throws Exception
{
    // 如果当前线程获取写锁,那么直接返回true,获取读锁成功
    if ( writeMutex.isOwnedByCurrentThread() )
    {
        return new PredicateResults(null, true);
    }

    int         index = 0;
    int         firstWriteIndex = Integer.MAX_VALUE;
    int         ourIndex = Integer.MAX_VALUE;
    for ( String node : children )
    {
        if ( node.contains(WRITE_LOCK_NAME) )
        {
            firstWriteIndex = Math.min(index, firstWriteIndex);
        }
        else if ( node.startsWith(sequenceNodeName) )
        {
            ourIndex = index;
            break;
        }

        ++index;
    }
    StandardLockInternalsDriver.validateOurIndex(sequenceNodeName, ourIndex);

    boolean     getsTheLock = (ourIndex < firstWriteIndex);
    String      pathToWatch = getsTheLock ? null : children.get(firstWriteIndex);
    return new PredicateResults(pathToWatch, getsTheLock);
}

如果客户端A已经获取了读锁

此时客户端B再来获取读锁

  1. children:[xxx_READ_0001, xxxx_READ_0002],此时都是读锁,不包含WRITE锁标识
  2. sequenceNodeName就是node创建的节点名称,这里ourIndex=index0
  3. ourIndex<firstWriteIndex = Integer.MAX_VALUE 获取锁成功

执行debug流程如下图:

image.png

因为读读不互斥,所以这里读锁也是可重入的

写读互斥及重入

上面已经分析过读读的逻辑了,这里接着按照上面的代码分析下读写的逻辑:

客户端A加写锁成功

客户端B加读锁

  1. node.contains(WRITE_LOCK_NAME) ,此时客户端B中含有WRITE标识
  2. firstWriteIndex = Math.min(index, firstWriteIndex)=0
  3. boolean getsTheLock = (ourIndex < firstWriteIndex);
    此时ourIndex = Integer.MAX_VALUE,判断条件不成立,所以加写锁失败

不同客户端写读锁互斥

接着看看同一个客户端逻辑:

if ( writeMutex.isOwnedByCurrentThread() )
{
    return new PredicateResults(null, true);
}

如果当前线程获取了写锁,那么再加读写直接返回成功。

所以同一个客户端同一线程:先加写锁、再加读锁可重入,这一点和Redisson中是一致的,具体可以看: 【分布式锁】04-使用Redisson实现ReadWriteLock原理

写写互斥及重入

写锁完全可以看做成 InterProcessMutex ,这里 maxLeases 为1,所以同一个线程写是可重入的,不同客户端获取锁时互斥的

读写互斥及重入

客户端A加读锁

客户端B加写锁

同样道理,此时children数据结构如:

[_c_13bf63d6-43f3-4c2f-ba98-07a641d351f2-__READ__0000000004,

_c_73b60882-9361-4fb7-8420-a8d4911d2c99-__WRIT__0000000005]

判断写锁在”/readwrite/readwrite_01″目录下的位置,不是在首位,加锁失败

可重入锁也是同样原理,不可重入

Zookeeper中MultiLock实现原理

我们在Redisson中已经见过MultiLock原理,其中Redissoon为了实现 RedLock 算法,也有MultiLock的实现(可以参考 【分布式锁】03-使用Redisson实现RedLock原理 )当多个资源需要统一加锁的时候,我们就可以使用MultiLock

Zookeeper中的MultiLock实现非常简单,就是依次加锁,实现如下图:

image.png

总结

Zookeeper实现分布式锁的相关原理全都讲完了,仔细阅读Curator源码觉得还挺有意思,再来会先Curator官网那句话:

Guava is to Java what Curator is to Zookeeper

Curator真的很强,分布式锁实现的很棒!

申明

本文章首发自本人公众号: 壹枝花算不算浪漫 ,如若转载请标明来源!

感兴趣的小伙伴可关注个人公众号:壹枝花算不算浪漫