Zookeeper实现分布式锁（一）While版

前面文章讲解了用Redis实现分布式锁的方式：

分布式锁之Redis实现（acquire）
分布式锁之Redis实现（最终版）

这次我们来使用Zookeeper来实现分布式锁

核心逻辑

我们使用Zookeeper同名节点只能创建一次的特性，来实现独占锁，具体方法为：

try {
    zk.create(lockNameSpace, "value".getBytes(), ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
} catch (KeeperException e) {
    //如果节点已经被其他创建，捕获异常
    e.printStackTrace();
} catch (InterruptedException e) {
    e.printStackTrace();
}

所有客户端用create方法创建名称为lockNameSpace变量值的节点，如果创建成功则拿到锁，创建失败抛出NodeExists 异常：

org.apache.zookeeper.KeeperException$NodeExistsException: KeeperErrorCode = NodeExists for /mylock
    at org.apache.zookeeper.KeeperException.create(KeeperException.java:119)
    at org.apache.zookeeper.KeeperException.create(KeeperException.java:51)
    at org.apache.zookeeper.ZooKeeper.create(ZooKeeper.java:783)
    at zookeeper.zookeeper.tttt.main(tttt.java:49)

说明获取锁失败。

具体实现

根据create方法的特性，我们来实现一个最简版的分布式锁

1.获取锁

public boolean lock() throws InterruptedException {
        String path = null;
        watchNode(nodeString);
        while (true) {
            try {
                path = zk.create(nodeString, "".getBytes(), ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL);
            } catch (KeeperException e) {
                System.out.println(Thread.currentThread().getName() + "请求失败");
                try {
                    Thread.sleep(500);
                } catch (InterruptedException ex) {
                    System.out.println("thread is notify");
                }
            }
             if (path!=null&&!path.isEmpty()) {
                 System.out.println(Thread.currentThread().getName() + " getLock...");
                 return true;
             }
        }
    }

代码逻辑：

path是create方法的返回值，用来判断是否创建成功。
while循环用来获取锁失败后不断重试
watchNode（）方法用来监听nodeString的状态

watchNode：

private void watchNode(String nodeString) throws InterruptedException {
        try {
            zk.exists(nodeString, new Watcher() {
                @Override
                public void process(WatchedEvent watchedEvent) {
                    System.out.println("事件来了：" + watchedEvent.toString());
                    if (watchedEvent.getType() == Event.EventType.NodeDeleted) {
                        System.out.println("delete事件");
                    }
                }
            });
        } catch (KeeperException e) {
            e.printStackTrace();
        }
    }

2.释放锁

public void unlock() {
        try {
            zk.delete(nodeString, -1);
            System.out.println(Thread.currentThread().getName() +  "release Lock...");
        } catch (InterruptedException e) {
            e.printStackTrace();
        } catch (KeeperException e) {
            e.printStackTrace();
        }
    }

代码很简单，删除nodeString节点，以便其它客户端来竞争锁。

3.main方法

public static void main(String args[]) throws InterruptedException {
        Zookeeper_1_While test = new Zookeeper_1_While();
        try {
            Thread.sleep(100);
            zk.create(lockNameSpace, "".getBytes(), ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
        } catch (KeeperException e) {
        } catch (InterruptedException e) {
        }
        ExecutorService service = Executors.newFixedThreadPool(10);
        for (int i = 0; i < 4; i++) {
            service.execute(() -> {
                try {
                    test.lock();
                    Thread.sleep(1800);
                } catch (InterruptedException e) {
                    e.printStackTrace();
                }
                test.unlock();
            });
        }
        service.shutdown();
    }

创建四个客户端来竞争锁，竞争到以后执行1.8s的业务逻辑，然后释放锁。

4.输出结果

Receive event WatchedEvent state:SyncConnected type:None path:null
connection is ok
事件来了：WatchedEvent state:SyncConnected type:NodeCreated path:/mylock/lock1
事件来了：WatchedEvent state:SyncConnected type:NodeCreated path:/mylock/lock1
事件来了：WatchedEvent state:SyncConnected type:NodeCreated path:/mylock/lock1
事件来了：WatchedEvent state:SyncConnected type:NodeCreated path:/mylock/lock1
pool-1-thread-1 拿到 Lock...
pool-1-thread-2请求失败
pool-1-thread-3请求失败
pool-1-thread-4请求失败
pool-1-thread-2请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-2请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-2请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-1释放 Lock...
pool-1-thread-2 拿到 Lock...
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-4请求失败
pool-1-thread-3请求失败
pool-1-thread-2释放 Lock...
pool-1-thread-4 拿到 Lock...
pool-1-thread-3请求失败
pool-1-thread-3请求失败
pool-1-thread-3请求失败
pool-1-thread-3请求失败
pool-1-thread-4释放 Lock...
pool-1-thread-3 拿到 Lock...
pool-1-thread-3释放 Lock...

存在的问题

通过打印的输出结果我们可以看到，同时竞争锁的客户端越多，请求失败的输出越频繁，也就是while循环去请求create方法越频繁，这会导致两个问题

1.while循环浪费cpu
2.频繁进行create重试会对Zookeeper服务器造成压力

githup代码：https://github.com/granett/zookeeper/blob/master/src/main/java/zookeeper/zookeeper/Zookeeper_1_While.java
针对以上两个问题，我们后续在下一篇Zookeeper实现分布式锁（二）Watcher版进行解决和优化。