Zookeeper开发总结 授权与验证 自动重连 递归删除 事务

                Zookeeper开发小结

一.实例化zookeeper与自动重连代码样例

<span style="font-size:18px;">public class ZkClient {    private ZooKeeper zooKeeper;    private String connectString;    private Integer sessionTimeout;    private Object <span style="color:#ff0000;"><strong>waiter = new Object();//simple object-lock</strong></span>    //"zk-client.properties";//classpath下    public ZkClient(String configLocation) throws  Exception{        Properties config = new Properties();        config.load(ClassLoader.getSystemResourceAsStream(configLocation));        connectString = config.getProperty("zk.connectString");        if(connectString == null){            throw new NullPointerException("'zk.connectString' cant be empty.. ");        }        sessionTimeout = Integer.parseInt(config.getProperty("zk.sessionTimeout","-1"));        connectZK();    }    /**     * core method,启动zk服务 本实例基于自动重连策略<span style="color:#ff0000;"><strong>,如果zk连接没有建立成功或者在运行时断开,将会自动重连.     */</strong></span>    private void connectZK() {</span>

<span style="font-size:18px;">//<span style="font-size:18px;">private Object <span style="color:#ff0000;background-color: rgb(240, 240, 240);"><strong>waiter = new Object();//simple object-lock</strong></span></span></span>

<span style="font-size:18px;">        synchronized (waiter) {            try {                SessionWatcher watcher =<span style="color:#ff0000;"><strong> new SessionWatcher();</strong></span>                // session的构建是异步的                this.zooKeeper = new ZooKeeper(connectString, sessionTimeout, watcher, false);            } catch (Exception e) {                e.printStackTrace();            }            <span style="color:#ff0000;"><strong>waiter.notifyAll();</strong></span>        }    }    class SessionWatcher implements Watcher {        public void process(WatchedEvent<strong> event)</strong> {            // 如果是“数据变更”事件            if (event.getType() != Event.EventType.None) {                return;            }            <span style="color:#ff0000;"><strong>// 如果是链接状态迁移            // 参见keeperState            synchronized (waiter) {</strong></span>                switch (event.getState()) {                    // zk连接建立成功,或者重连成功                    case SyncConnected:                        System.out.println("Connected...");                        waiter.notifyAll();                        break;                    // session过期,这是个非常严重的问题,有可能client端出现了问题,也有可能zk环境故障                    // 此处仅仅是重新实例化zk client                    case Expired:                        System.out.println("Expired...");                        // 重连                        connectZK();                        break;                    // session过期                    case Disconnected:                        // 链接断开，或session迁移                        System.out.println("Connecting....");                        break;                    case AuthFailed:                        close();                        throw new RuntimeException("ZK Connection auth failed...");                    default:                        break;                }            }        }    }    private void close(){        try {            synchronized (waiter) {                if (this.zooKeeper != null) {                    zooKeeper.close();                }                waiter.notifyAll();            }        } catch (Exception e) {            e.printStackTrace();        }    }}</span>

二.授权与验证

<span style="font-size:18px;">String auth = "admin:admin";//anywhere,but before znode operation//can addauth more than oncezooKeeper.addAuthInfo("digest", auth.getBytes("UTF-8"));</span>

 

<span style="font-size:18px;">Id id = new Id("digest", DigestAuthenticationProvider.generateDigest(auth));ACL acl = new ACL(ZooDefs.Perms.ALL, id);List<ACL> acls = Collections.singletonList(acl);//如果不需要访问控制,可以使用acls = ZooDefs.Ids.OPEN_ACL_UNSAFEzooKeeper.create("/singleWorker", null, acls, CreateMode.PERSISTENT);</span>

三.创建znode节点

<span style="font-size:18px;">try{ Id id = new Id("digest", DigestAuthenticationProvider.generateDigest(auth)); ACL acl = new ACL(ZooDefs.Perms.ALL, id); List<ACL> acls = Collections.singletonList(acl); zooKeeper.create("/workers", null, acls, CreateMode.PERSISTENT);}catch(KeeperException.NodeExistsException e){ //在并发环境中,节点创建有可能已经被创建,即使使用exist方法检测也不能确保.} catch (Exception e){ e.printStackTrace();}</span>

四.删除节点

<span style="font-size:18px;">//zookeeper不允许直接删除含有子节点的节点;//如果你需要删除当前节点以及其所有子<span style="color:#ff0000;"><strong>节点,需要递归来做</strong></span>private void deletePath(String path,ZooKeeper zooKeeper) throws Exception{ List<String> children = zooKeeper.getChildren(path,false); for(String child : children){  <strong><span style="color:#ff0000;">String childPath = path + "/" + child;  deletePath(childPath,zooKeeper);</span></strong> } try{  zooKeeper.delete(path,-1); }catch(KeeperException.NoNodeException e){  //ignore }}</span>

<span style="font-size:18px;">//删除节点,删除时比较version,避免删除时被其他client修改public boolean delete(String path){ try{  Stat stat = zooKeeper.exists(path,false);  //如果节点已经存在  if(stat != null){   zooKeeper.delete(path,stat.getVersion());  } }catch(KeeperException.NoNodeException e){  //igore }catch (Exception e){  e.printStackTrace();                return false; } return true;}</span>

五.修改数据

<span style="font-size:18px;">public boolean update(String path,byte[] data){   try{    Stat stat = zooKeeper.exists(path,false);    //如果节点已经存在    if(stat != null){     zooKeeper.setData(path,data,stat.getVersion());     return true;    }   }catch (KeeperException.NoNodeException e){     //ignore   }catch (Exception e){    e.printStackTrace();   } return false;}</span>

五.事务

<span style="font-size:18px;">public boolean create(String name){ try{  Transaction tx = zooKeeper.transaction();  tx.create("/workers/servers/" + name,null, ZooDefs.Ids.OPEN_ACL_UNSAFE,null);  tx.create("/workers/schedule/" + name,null, ZooDefs.Ids.OPEN_ACL_UNSAFE,null);  tx.commit();  return true; } catch (Exception e){  e.printStackTrace(); } return false;}</span>

    备注:zookeeper中的watcher机制非常好,但是重要的数据变更,不能完全依赖watcher通知,因为对于zkClient而言,网络异常都将会导致watcher有丢失的潜在风险,而且watcher是"即发即失",当你接收到watcher通知之后,在处理过程中,数据仍然有变更的可能,因此在时间线上,不可能做到完全准确.