[TOC]

前言

紧跟上文的：分布式锁实现（一）：Redis ，这篇我们用Zookeeper来设计和实现分布式锁，并且研究下开源客户端工具Curator的分布式锁源码

设计实现

一、基本算法

1.在某父节点下创建临时有序节点
2.判断创建的节点是否是当前父节点下所有子节点中序号最小的
3.是序号最小的成功获取锁，否则监听比自己小的那个节点，进行watch，当该节点被删除的时候通知当前节点，重新获取锁
4.解锁的时候删除当前节点

二、关键点

临时有序节点

实现Zookeeper分布式锁关键就在于其[临时有序节点]的特性，在Zookeeper中有四种节点
1.PERSISTENT 持久，若不手动删除就永久存在
2.PERSISTENT_SEQUENTIAL 持久有序节点，zookeeper会为节点编号（保证有序）
3.EPHEMERAL 临时，一个客户端会话断开后会自动删除
4.EPHEMERAL_SEQUENTIAL 临时有序节点，zookeeper会为节点编号（保证有序）

监听

Zookeeper提供事件监听机制，通过对节点、节点数据、子节点都提供了监听，我们通过这种监听watcher机制实现锁的等待

三、代码实现

我们基于ZkClient这个客户端来实现，当然也可以用原生Zookeeper API,大致是一样的
坐标如下：
  <dependency>
        <groupId>com.101tec</groupId>
        <artifactId>zkclient</artifactId>
        <version>0.2</version>
    </dependency>

代码如下：

public class MyDistributedLock {


    private ZkClient zkClient;
    private String name;
    private String currentLockPath;
    private CountDownLatch countDownLatch;

    private static final String PARENT_LOCK_PATH = "/distribute_lock";

    public MyDistributedLock(ZkClient zkClient, String name) {
        this.zkClient = zkClient;
        this.name = name;
    }

	//加锁
    public void lock() {
    	//判断父节点是否存在，不存在就创建
        if (!zkClient.exists(PARENT_LOCK_PATH)) {
            try {
            	//多个线程只会成功建立一次
                zkClient.createPersistent(PARENT_LOCK_PATH);
            } catch (Exception ignored) {
            }
        }
        //创建当前目录下的临时有序节点
        currentLockPath = zkClient.createEphemeralSequential(PARENT_LOCK_PATH + "/", System.currentTimeMillis());
        //校验是否最小节点
        checkMinNode(currentLockPath);
    }

	//解锁
    public void unlock() {
        System.out.println("delete : " + currentLockPath);
        zkClient.delete(currentLockPath);
    }


    private boolean checkMinNode(String lockPath) {
		//获取当前目录下所有子节点
        List<String> children = zkClient.getChildren(PARENT_LOCK_PATH);
        Collections.sort(children);
        int index = children.indexOf(lockPath.substring(PARENT_LOCK_PATH.length() + 1));
        if (index == 0) {
            System.out.println(name + "：success");
            if (countDownLatch != null) {
                countDownLatch.countDown();
            }
            return true;
        } else {
            String waitPath = PARENT_LOCK_PATH + "/" + children.get(index - 1);
            //等待前一个节点释放的监听
            waitForLock(waitPath);
            return false;
        }
    }


    private void waitForLock(String prev) {
        System.out.println(name + " current path :" + currentLockPath + "：fail add listener" + " wait path :" + prev);
        countDownLatch = new CountDownLatch(1);
        zkClient.subscribeDataChanges(prev, new IZkDataListener() {
            @Override
            public void handleDataChange(String s, Object o) throws Exception {

            }

            @Override
            public void handleDataDeleted(String s) throws Exception {
                System.out.println("prev node is done");
                checkMinNode(currentLockPath);
            }
        });
        if (!zkClient.exists(prev)) {
            return;
        }
        try {
            countDownLatch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        countDownLatch = null;
    }
}

加锁

zkClient.exists先判断父节点是否存在，不存在就创建，zookeeper可以保证只会创建成功一次
在当前目录下zkClient.createEphemeralSequential创建临时有序节点，再判断当前目录下此节点是否为序号最小的，如果是，成功获取锁，否则的话拿比自己小的节点，并做监听
waitForLock等待比自己小的节点，subscribeDataChanges监听一个节点的变化，handleDataDeleted里面再次做checkMinNode的判断
监听完毕后，再判断一次此节点是否存在，因为在监听的过程中有可能之前小的那个节点重新释放了锁，如果之前节点不存在的话，无需在这里等待，这里的等待是通过countDownLatch实现的

解锁

解锁就是通过zkClient的delete删除当前节点

测试用例

通过启动多个线程来测试lock、unlock的过程，查看是否有序

public class MyDistributedLockTest {


    public static void main(String[] args) {

        ZkClient zk = new ZkClient("127.0.0.1:2181", 5 * 10000);

        for (int i = 0; i < 20; i++) {

            String name = "thread" + i;
            Thread thread = new Thread(() -> {
                MyDistributedLock myDistributedLock = new MyDistributedLock(zk, name);
                myDistributedLock.lock();
//                try {
//                    Thread.sleep(1 * 1000);
//                } catch (InterruptedException e) {
//                    e.printStackTrace();
//                }
                myDistributedLock.unlock();
            });
            thread.start();
        }

    }
}

执行结果如下，多线程情况下lock/unlock和监听一切正常：

thread1 current path :/distribute_lock2/0000000007：fail add listener wait path :/distribute_lock2/0000000006
thread6 current path :/distribute_lock2/0000000006：fail add listener wait path :/distribute_lock2/0000000005
thread3：success
delete : /distribute_lock2/0000000000
thread2 current path :/distribute_lock2/0000000005：fail add listener wait path :/distribute_lock2/0000000004
thread7 current path :/distribute_lock2/0000000004：fail add listener wait path :/distribute_lock2/0000000003
thread9 current path :/distribute_lock2/0000000009：fail add listener wait path :/distribute_lock2/0000000008
thread5 current path :/distribute_lock2/0000000008：fail add listener wait path :/distribute_lock2/0000000007
thread0 current path :/distribute_lock2/0000000001：fail add listener wait path :/distribute_lock2/0000000000
thread8 current path :/distribute_lock2/0000000002：fail add listener wait path :/distribute_lock2/0000000001
thread4 current path :/distribute_lock2/0000000003：fail add listener wait path :/distribute_lock2/0000000002
delete : /distribute_lock2/0000000001
prev node is done
thread8：success
delete : /distribute_lock2/0000000002
prev node is done
thread4：success
delete : /distribute_lock2/0000000003
prev node is done
thread7：success
delete : /distribute_lock2/0000000004
prev node is done
thread2：success
delete : /distribute_lock2/0000000005
prev node is done
thread6：success
delete : /distribute_lock2/0000000006
prev node is done
thread1：success
delete : /distribute_lock2/0000000007
prev node is done
thread5：success
delete : /distribute_lock2/0000000008
prev node is done
thread9：success
delete : /distribute_lock2/0000000009

Curator源码分析

一、基本使用

 		RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
        CuratorFramework client = CuratorFrameworkFactory.newClient("127.0.0.1:2181", retryPolicy);
        client.start();
        InterProcessMutex lock2 = new InterProcessMutex(client, "/test");

        try {
            lock.acquire();
            //业务
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            lock.release();
        }

CuratorFrameworkFactory.newClient获取zookeeper的客户端，retryPolicy指定重试策略，开启客户端
Curator本身提供了多种锁的实现，这里我们以InterProcessMutex可重入锁为例， lock.acquire()方法获取锁，lock.release()来释放锁，acquire方法也提供了重载的等待时间参数

二、源码分析

加锁

acquire内部就直接internalLock方法，传了-1的等待时间

 public void acquire() throws Exception {
        if(!this.internalLock(-1L, (TimeUnit)null)) {
            throw new IOException("Lost connection while trying to acquire lock: " + this.basePath);
        }
    }

internalLock方法首先判断是否是重入锁，通过ConcurrentMap维护线程和一个原子计数器，非重入锁的话，再通过attemptLock去获取锁

 private boolean internalLock(long time, TimeUnit unit) throws Exception
    {
        /*
           Note on concurrency: a given lockData instance
           can be only acted on by a single thread so locking isn't necessary
        */

        Thread currentThread = Thread.currentThread();

        LockData lockData = threadData.get(currentThread);
        if ( lockData != null )
        {
            // re-entering
            lockData.lockCount.incrementAndGet();
            return true;
        }

        String lockPath = internals.attemptLock(time, unit, getLockNodeBytes());
        if ( lockPath != null )
        {
            LockData newLockData = new LockData(currentThread, lockPath);
            threadData.put(currentThread, newLockData);
            return true;
        }

        return false;
    }

attemptLock在这里进行循环等待，createsTheLock方法去创建节点，internalLockLoop去判断当前节点是否是最小节点

String attemptLock(long time, TimeUnit unit, byte[] lockNodeBytes) throws Exception
    {
        final long      startMillis = System.currentTimeMillis();
        final Long      millisToWait = (unit != null) ? unit.toMillis(time) : null;
        final byte[]    localLockNodeBytes = (revocable.get() != null) ? new byte[0] : lockNodeBytes;
        int             retryCount = 0;

        String          ourPath = null;
        boolean         hasTheLock = false;
        boolean         isDone = false;
        while ( !isDone )
        {
            isDone = true;

            try
            {
                ourPath = driver.createsTheLock(client, path, localLockNodeBytes);
                hasTheLock = internalLockLoop(startMillis, millisToWait, ourPath);
            }
            catch ( KeeperException.NoNodeException e )
            {
                // gets thrown by StandardLockInternalsDriver when it can't find the lock node
                // this can happen when the session expires, etc. So, if the retry allows, just try it all again
                if ( client.getZookeeperClient().getRetryPolicy().allowRetry(retryCount++, System.currentTimeMillis() - startMillis, RetryLoop.getDefaultRetrySleeper()) )
                {
                    isDone = false;
                }
                else
                {
                    throw e;
                }
            }
        }

        if ( hasTheLock )
        {
            return ourPath;
        }

        return null;
    }

createsTheLock就是调用curator封装的api去创建临时有序节点

   public String createsTheLock(CuratorFramework client, String path, byte[] lockNodeBytes) throws Exception
    {
        String ourPath;
        if ( lockNodeBytes != null )
        {
            ourPath = client.create().creatingParentContainersIfNeeded().withProtection().withMode(CreateMode.EPHEMERAL_SEQUENTIAL).forPath(path, lockNodeBytes);
        }
        else
        {
            ourPath = client.create().creatingParentContainersIfNeeded().withProtection().withMode(CreateMode.EPHEMERAL_SEQUENTIAL).forPath(path);
        }
        return ourPath;
    }

internalLockLoop锁判断，内部就是driver.getsTheLock去判断是否是当前目录下最小节点，如果是的话，返回获取锁成功，否则的话对previousSequencePath进行监听，监听动作完成后再对等待时间进行重新判断

private boolean internalLockLoop(long startMillis, Long millisToWait, String ourPath) throws Exception
    {
        boolean     haveTheLock = false;
        boolean     doDelete = false;
        try
        {
            if ( revocable.get() != null )
            {
                client.getData().usingWatcher(revocableWatcher).forPath(ourPath);
            }

            while ( (client.getState() == CuratorFrameworkState.STARTED) && !haveTheLock )
            {
                List<String>        children = getSortedChildren();
                String              sequenceNodeName = ourPath.substring(basePath.length() + 1); // +1 to include the slash

                PredicateResults    predicateResults = driver.getsTheLock(client, children, sequenceNodeName, maxLeases);
                if ( predicateResults.getsTheLock() )
                {
                    haveTheLock = true;
                }
                else
                {
                    String  previousSequencePath = basePath + "/" + predicateResults.getPathToWatch();

                    synchronized(this)
                    {
                        try 
                        {
                            // use getData() instead of exists() to avoid leaving unneeded watchers which is a type of resource leak
                            client.getData().usingWatcher(watcher).forPath(previousSequencePath);
                            if ( millisToWait != null )
                            {
                                millisToWait -= (System.currentTimeMillis() - startMillis);
                                startMillis = System.currentTimeMillis();
                                if ( millisToWait <= 0 )
                                {
                                    doDelete = true;    // timed out - delete our node
                                    break;
                                }

                                wait(millisToWait);
                            }
                            else
                            {
                                wait();
                            }
                        }
                        catch ( KeeperException.NoNodeException e ) 
                        {
                            // it has been deleted (i.e. lock released). Try to acquire again
                        }
                    }
                }
            }
        }
        catch ( Exception e )
        {
            ThreadUtils.checkInterrupted(e);
            doDelete = true;
            throw e;
        }
        finally
        {
            if ( doDelete )
            {
                deleteOurPath(ourPath);
            }
        }
        return haveTheLock;
    }

解锁

release代码相对来说比较简单，就是先判断map里面是否存在当前线程的锁计数，不存在抛出异常，存在的话，进行原子减一操作，releaseLock内部就是删除节点操作，小于0的时候，从map里面移除

  public void release() throws Exception
    {
        /*
            Note on concurrency: a given lockData instance
            can be only acted on by a single thread so locking isn't necessary
         */

        Thread currentThread = Thread.currentThread();
        LockData lockData = threadData.get(currentThread);
        if ( lockData == null )
        {
            throw new IllegalMonitorStateException("You do not own the lock: " + basePath);
        }

        int newLockCount = lockData.lockCount.decrementAndGet();
        if ( newLockCount > 0 )
        {
            return;
        }
        if ( newLockCount < 0 )
        {
            throw new IllegalMonitorStateException("Lock count has gone negative for lock: " + basePath);
        }
        try
        {
            internals.releaseLock(lockData.lockPath);
        }
        finally
        {
            threadData.remove(currentThread);
        }
    }

后记

分布式锁的实现目前主流比较常用的实现就是Redis和Zookeeper了，相比较自己的实现，Redission和Curator的设计实现更为优秀，也更值得我们借鉴和学习

千里之行，积于跬步;万里之船，成于罗盘，共勉。