利用ZK实现分布式锁

 

http://coolxing.iteye.com/blog/1871630

场景描述

在分布式应用, 往往存在多个进程提供同一服务. 这些进程有可能在相同的机器上, 也有可能分布在不同的机器上. 如果这些进程共享了一些资源, 可能就需要分布式锁来锁定对这些资源的访问.
本文将介绍如何利用zookeeper实现分布式锁.

思路

进程需要访问共享数据时, 就在"/locks"节点下创建一个sequence类型的子节点, 称为thisPath. 当thisPath在所有子节点中最小时, 说明该进程获得了锁. 进程获得锁之后, 就可以访问共享资源了. 访问完成后, 需要将thisPath删除. 锁由新的最小的子节点获得.
有了清晰的思路之后, 还需要补充一些细节. 进程如何知道thisPath是所有子节点中最小的呢? 可以在创建的时候, 通过getChildren方法获取子节点列表, 然后在列表中找到排名比thisPath前1位的节点, 称为waitPath, 然后在waitPath上注册监听, 当waitPath被删除后, 进程获得通知, 此时说明该进程获得了锁.

实现

以一个DistributedClient对象模拟一个进程的形式, 演示zookeeper分布式锁的实现.

public class DistributedClient {
    // 超时时间
    private static final int SESSION_TIMEOUT = 5000;
    // zookeeper server列表
    private String hosts = "localhost:4180,localhost:4181,localhost:4182";
    private String groupNode = "locks";
    private String subNode = "sub";

    private ZooKeeper zk;
    // 当前client创建的子节点
    private String thisPath;
    // 当前client等待的子节点
    private String waitPath;

    private CountDownLatch latch = new CountDownLatch(1);

    /**
     * 连接zookeeper
     */
    public void connectZookeeper() throws Exception {
        zk = new ZooKeeper(hosts, SESSION_TIMEOUT, new Watcher() {
            public void process(WatchedEvent event) {
                try {
                    // 连接建立时, 打开latch, 唤醒wait在该latch上的线程
                    if (event.getState() == KeeperState.SyncConnected) {
                        latch.countDown();
                    }

                    // 发生了waitPath的删除事件
                    if (event.getType() == EventType.NodeDeleted && event.getPath().equals(waitPath)) {
                        doSomething();
                    }
                } catch (Exception e) {
                    e.printStackTrace();
                }
            }
        });

        // 等待连接建立
        latch.await();

        // 创建子节点
        thisPath = zk.create("/" + groupNode + "/" + subNode, null, Ids.OPEN_ACL_UNSAFE,
                CreateMode.EPHEMERAL_SEQUENTIAL);

        // wait一小会, 让结果更清晰一些
        Thread.sleep(10);

        // 注意, 没有必要监听"/locks"的子节点的变化情况
        List<String> childrenNodes = zk.getChildren("/" + groupNode, false);

        // 列表中只有一个子节点, 那肯定就是thisPath, 说明client获得锁
        if (childrenNodes.size() == 1) {
            doSomething();
        } else {
            String thisNode = thisPath.substring(("/" + groupNode + "/").length());
            // 排序
            Collections.sort(childrenNodes);
            int index = childrenNodes.indexOf(thisNode);
            if (index == -1) {
                // never happened
            } else if (index == 0) {
                // inddx == 0, 说明thisNode在列表中最小, 当前client获得锁
                doSomething();
            } else {
                // 获得排名比thisPath前1位的节点
                this.waitPath = "/" + groupNode + "/" + childrenNodes.get(index - 1);
                // 在waitPath上注册监听器, 当waitPath被删除时, zookeeper会回调监听器的process方法
                zk.getData(waitPath, true, new Stat());
            }
        }
    }

    private void doSomething() throws Exception {
        try {
            System.out.println("gain lock: " + thisPath);
            Thread.sleep(2000);
            // do something
        } finally {
            System.out.println("finished: " + thisPath);
            // 将thisPath删除, 监听thisPath的client将获得通知
            // 相当于释放锁
            zk.delete(this.thisPath, -1);
        }
    }

    public static void main(String[] args) throws Exception {
        for (int i = 0; i < 10; i++) {
            new Thread() {
                public void run() {
                    try {
                        DistributedClient dl = new DistributedClient();
                        dl.connectZookeeper();
                    } catch (Exception e) {
                        e.printStackTrace();
                    }
                }
            }.start();
        }

        Thread.sleep(Long.MAX_VALUE);
    }
};