Curator是Netflix公司开源的一套zookeeper客户端框架.解决了很多Zookeeper客户端非常底层的细节开发工作,包括连接重连、反复注册Watcher和NodeExistsException异常等等。Curator被看做是zookeeper客户端框里面的瑞士军刀(牛逼了)。Curator使得我们开发zookeeper客户端程序变的很容易。
Curator框架包含三个主要的包:
- curator-framework:对zookeeper的底层api的一些封装。
- curator-client:提供一些客户端的操作,例如重试策略等。
- curator-recipes:封装了一些高级特性,例如Cache事件监听、选举、分布式锁、分布式计数器、分布式Barrier等。
Curator的引入(pom方式,版本可能有变化)。
org.apache.curator
curator-framework
4.0.1
org.apache.curator
curator-client
4.0.1
com.google.guava
guava
org.apache.curator
curator-recipes
4.0.1
org.apache.zookeeper
zookeeper
3.4.10
org.slf4j
slf4j-log4j12
log4j
log4j
一 Curator的基本用法
1.1 创建zookeeper客户端
在Curator中CuratorFramework对象就代表一个zookeeper客户端。所以创建创建zookeeper客户端就是创建CuratorFramework对象。CuratorFramework对象又可以通过CuratorFrameworkFactory来创建。
CuratorFramework api介绍如下
public interface CuratorFramework {
/**
* 启动zookeeper客户端
*/
public void start();
/**
* 关闭zookeeper客户端
*/
public void close();
/**
* 返回客户端状态:LATENT、STARTED、STOPPED
*/
public CuratorFrameworkState getState();
/**
* 客户端是否已经启动
*/
@Deprecated
public boolean isStarted();
/**
* 创建节点的建造器
*/
public CreateBuilder create();
/**
* 删除节点的建造器
*/
public DeleteBuilder delete();
/**
* 检查节点是否存在的建造器
*/
public ExistsBuilder checkExists();
/**
* 获取接连数据的建造器
*/
public GetDataBuilder getData();
/**
* 设置节点数据的建造器
*/
public SetDataBuilder setData();
/**
* 获取子节点的建造器
*/
public GetChildrenBuilder getChildren();
/**
* 获取权限的构造器
*/
public GetACLBuilder getACL();
/**
* 设置权限的构造器
*/
public SetACLBuilder setACL();
/**
* 重新配置的建造器
*/
public ReconfigBuilder reconfig();
/**
* 获取配置的建造器
*/
public GetConfigBuilder getConfig();
/**
* 事务构造器
* @deprecated use {@link #transaction()} instead
*/
public CuratorTransaction inTransaction();
/**
* 事务构造器
*/
public CuratorMultiTransaction transaction();
/**
* 分配可与{transaction()}一起使用的操作
*/
public TransactionOp transactionOp();
/**
* 如果路径不存在,则创建路径对应的节点
*/
public void createContainers(String path) throws Exception;
/**
* 启动同步构建器。注意:即使您不使用其中一种background()方法,同步也始终在后台
*/
public SyncBuilder sync();
/**
* 启动remove watch builder,有节点删除的时候会调用
*/
public RemoveWatchesBuilder watches();
/**
*
* 返回Connect State的可侦听接口
*/
public Listenable getConnectionStateListenable();
/**
* 返回事件的可侦听接口
*/
public Listenable getCuratorListenable();
/**
* 返回未处理错误的可侦听接口
*/
public Listenable getUnhandledErrorListenable();
/**
* 返回一个新的CuratorFramework,该CuratorFramework指定了一个新的命名空间
*/
public CuratorFramework usingNamespace(String newNamespace);
/**
* 获取命名空间
*/
public String getNamespace();
/**
* 返回托管的zookeeper客户端
*/
public CuratorZookeeperClient getZookeeperClient();
/**
* 阻塞,直到与ZooKeeper的连接可用或已超过maxWaitTime
*/
public boolean blockUntilConnected(int maxWaitTime, TimeUnit units) throws InterruptedException;
/**
* 阻塞,直到与ZooKeeper的连接可用。在连接可用或中断之前,此方法不会返回,在这种情况下,将抛出InterruptedException
*/
public void blockUntilConnected() throws InterruptedException;
/**
* 返回跟踪观察者创建的当前实例的外观,并允许一次性删除所有观察者
*/
public WatcherRemoveCuratorFramework newWatcherRemoveCuratorFramework();
/**
* 返回配置的错误策略
*/
public ConnectionStateErrorPolicy getConnectionStateErrorPolicy();
/**
*
* Current维护Zookeeper仲裁配置的缓存视图。
*/
public QuorumVerifier getCurrentConfig();
/**
* 获取SchemaSet
*/
SchemaSet getSchemaSet();
/**
* 如果此实例在ZK 3.4.x兼容模式下运行,则返回true
*/
boolean isZk34CompatibilityMode();
}
CuratorFrameworkFactory api介绍如下
public class CuratorFrameworkFactory {
/**
* 用于通过建造者模式创建zookeeper客户端
*/
public static Builder builder();
/**
* 创建zookeeper客户端
*/
public static CuratorFramework newClient(String connectString, RetryPolicy retryPolicy);
/**
* 创建zookeeper客户端
*/
public static CuratorFramework newClient(String connectString, int sessionTimeoutMs, int connectionTimeoutMs, RetryPolicy retryPolicy);
/**
* 将本地地址作为可用作节点有效负载的字节返回
*/
public static byte[] getLocalAddress();
public static class Builder {
/**
* build CuratorFramework对象 -- zookeeper客户端
*/
public CuratorFramework build();
/**
* 创建一个临时的CuratorFramework客户端,CuratorFramework,默认3分钟不活动客户端连接就被关闭
*/
public CuratorTempFramework buildTemp();
/**
* 创建一个临时的CuratorFramework客户端,CuratorFramework,可以自己指定多长时间不活动客户端连接就被关闭
*/
public CuratorTempFramework buildTemp(long inactiveThreshold, TimeUnit unit);
/**
* 添加zookeeper 访问权限
*/
public Builder authorization(String scheme, byte[] auth);
public Builder authorization(List authInfos);
/**
* 设置zookeeper服务器列表
*/
public Builder connectString(String connectString);
/**
* zookeeper服务器地址通过EnsembleProvider(配置提供者)来提供,不能和connectString共同使用
*/
public Builder ensembleProvider(EnsembleProvider ensembleProvider);
/**
* 为每次新建的节点设置一个默认值
*/
public Builder defaultData(byte[] defaultData);
/**
* 设置命名空间,为了实现不同的Zookeeper业务之间的隔离,有的时候需要为每个业务分配一个独立的命名空间
*/
public Builder namespace(String namespace)
/**
* 会话超时时间,单位毫秒,默认60000ms
*/
public Builder sessionTimeoutMs(int sessionTimeoutMs);
/**
* 连接创建超时时间,单位毫秒,默认60000ms
*/
public Builder connectionTimeoutMs(int connectionTimeoutMs);
/**
* @param maxCloseWaitMs time to wait during close to join background threads
* @return this
*/
public Builder maxCloseWaitMs(int maxCloseWaitMs);
/**
* 设置客户端重连策略
*/
public Builder retryPolicy(RetryPolicy retryPolicy);
/**
* Executor Services的线程工厂
*/
public Builder threadFactory(ThreadFactory threadFactory);
/**
* 压缩器,用于压缩和解压数据
*/
public Builder compressionProvider(CompressionProvider compressionProvider);
/**
* ZookeeperFactory 用于创建ZooKeeper
*/
public Builder zookeeperFactory(ZookeeperFactory zookeeperFactory);
/**
* 权限控制器
*/
public Builder aclProvider(ACLProvider aclProvider);
/**
* 设置只读模式
*/
public Builder canBeReadOnly(boolean canBeReadOnly);
/**
* 不让客户端,创建节点的时候顺带创建父节点
*/
public Builder dontUseContainerParents();
/**
* 默认是StandardConnectionStateErrorPolicy,设置要使用的错误策略
*/
public Builder connectionStateErrorPolicy(ConnectionStateErrorPolicy connectionStateErrorPolicy);
/**
* 如果mode为true,则创建ZooKeeper 3.4.x兼容客户端。如果使用的客户端库是ZooKeeper 3.4.x 默认情况下已启用
*/
public Builder zk34CompatibilityMode(boolean mode);
/**
* 更改连接处理策略,默认StandardConnectionHandlingPolicy
*/
public Builder connectionHandlingPolicy(ConnectionHandlingPolicy connectionHandlingPolicy);
/**
* 添加强制架构集
*/
public Builder schemaSet(SchemaSet schemaSet);
}
}
从上面的CuratorFrameworkFactory api的介绍可以看出CuratorFrameworkFactory对象的创建有两种方式:
- 通过过构造函数创建
参数 | 类型 | 含义 |
---|---|---|
connectString | String | 服务器列表,格式host1:port1,host2:port2,… |
sessionTimeoutMs | int | 会话超时时间,单位毫秒,默认60000ms |
connectionTimeoutMs | int | 连接创建超时时间,单位毫秒,默认60000ms |
retryPolicy | RetryPolicy | 重试策略,curator已经提供了多种重试策略,也可以自行实现RetryPolicy接口 |
curator提供的重试策略有:ExponentialBackoffRetry、BoundedExponentialBackoffRetry、RetryForever、RetryNTimes、RetryOneTime、RetryUntilElapsed
- 通过build创建,关于build里面的各个参数在CuratorFrameworkFactory api里面都顺带介绍了哦。
比如如下的实例代码,连接到127.0.0.1:2181服务端。
关于zookeeper的安装大家可以自己去网上搜下。
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
1.2 启动客户端
调用start()方法启动客户端。这个时候zookeeper客户端才会去连接zookeeper服务端。在zookeeper客户端上做的所有动作都需要在start()之后执行。如果你不想连接服务端的时候可以调用close()方法断开连接.
1.3 节点操作
首先我们要明确zookeeper里面的节点结构类似于我们文件系统的结构(就像一棵树样的)。除此之外zookeeper的每个节点上还可以保存数据。zookeeper里面的节点有四种,不同的节点类型都有自己的特点:
- CreateMode.PERSISTENT:持久化节点。
- CreateMode.PERSISTENT_SEQUENTIAL:持久化并且带序列号节点。
- CreateMode.EPHEMERAL:临时节点(客户端断开了节点也就删除了)
- CreateMode.EPHEMERAL_SEQUENTIAL:临时并且带序列号(客户端断开了节点也就删除了)
1.3.1 创建节点
创建节点很简单,我们前面已经创建了zookeeper客户端,并且调了start()方法把客户端启动起来了。
比如我们可以用如下的代码创建一个持久化的节点。通过withMode(CreateMode.PERSISTENT)来指定节点的类型。
/**
* 同步 创建持久化节点
*
* @param path 节点路径
* @throws Exception errors
*/
public void createPersistentNodeSync(String path) throws Exception {
client.create()
.creatingParentContainersIfNeeded() // 自动递归创建父节点
.withMode(CreateMode.PERSISTENT)
.forPath(path);
}
我们也可以在创建节点的同时,给节点设置数据。
/**
* 同步-创建持久化节点
*
* @param path 节点路径
* @param data 节点对应的值
* @throws Exception errors
*/
public void createPersistentNodeSync(String path, byte[] data) throws Exception {
client.create()
.creatingParentContainersIfNeeded() // 自动递归创建父节点
.withMode(CreateMode.PERSISTENT)
.forPath(path, data);
}
1.3.2 删除节点
删除叶子节点。(如果不是叶子节点是会报异常的)
/**
* 同步-删除一个叶子节点(注意哦,只能删除叶子节点否则报错的)
*
* @param path 需要删除的节点对应的路径
* @throws Exception errors
*/
public void deleteNodeSync(String path) throws Exception {
client.delete()
.forPath(path);
}
我们也可以删除整个节点(包括节点下的子节点)。
/**
* 同步-删除一个节点,并且递归删除其所有的子节点
*
* @param path 需要删除的节点对应的路基
* @throws Exception errors
*/
public void deleteNodeRecursivelySync(String path) throws Exception {
client.delete()
.deletingChildrenIfNeeded()
.forPath(path);
}
1.3.3 判断节点是否存在
通过节点的Stat来判断节点是否存在。
/**
* 同步-检查节点是否存在
*
* @param path 节点路径
* @return 节点是否存在
* @throws Exception errors
*/
public boolean isNodeExistSync(String path) throws Exception {
Stat state = client.checkExists()
.forPath(path);
return state != null;
}
1.3.4 节点数据操作
读取节点数据。
/**
* 同步-读取一个节点的数据内容
*
* @param path 节点路基
* @return 节点内容
* @throws Exception errors
*/
public byte[] getNodeDataSync(String path) throws Exception {
return client.getData()
.forPath(path);
}
更新节点数据,或者设置null删除节点数据
/**
* 同步-更新一个节点的数据内容
*
* @param path 节点路径
* @param data 节点对应数据
* @throws Exception errors
*/
public void updateNodeDataSync(String path, byte[] data) throws Exception {
client.setData()
.forPath(path, data);
}
1.3.5 获取节点的所有子节点
/**
* 同步-获取某个节点的所有子节点路径
*
* @param path 目录
* @return children
* @throws Exception errors
*/
public List getChildrenSync(String path) throws Exception {
return client.getChildren()
.forPath(path);
}
1.4 事务
事务相信大家都非常的熟悉。Curator也提供了事务的支持,一组crud操作要么都成功,要么都失败。使用起来也非常的简单。
一个事务里面肯定是有多个操作的,我们首先要把每个操作都封装成CuratorOp。比如如下的实例,我们把多个操作放到一个事务里面去执行.
@Test
public void transaction() throws Exception {
CuratorOp createOp = client.transactionOp().create().forPath("/a/path", "some data".getBytes());
CuratorOp setDataOp = client.transactionOp().setData().forPath("/another/path", "other data".getBytes());
CuratorOp deleteOp = client.transactionOp().delete().forPath("/yet/another/path");
Collection results = client.transaction().forOperations(createOp, setDataOp, deleteOp);
for (CuratorTransactionResult result : results) {
System.out.println(result.getForPath() + " - " + result.getType());
}
}
1.5 异步操作
因为zookeeper客户端的操作都是在和zookeeper服务端打交道的。涉及到网络的调用。所以有些操作的响应就不会那么及时了。Curator就给提供了异步操作。异步响应操作结果。
既然是异步操作,那么肯定需要BackgroundCallback来异步接收操作结果了。关于异步操作,我们也举一个简单的例子,我们以创建节点来举例(删除节点,修改节点数据,事务等等其他操作都是一样的使用)。
/**
* 异步-获取某个节点的所有子节点路径
*
* @param path 目录
* @param callback 回调
* @throws Exception errors
*/
public void getChildrenAsync(String path, BackgroundCallback callback) throws Exception {
client.getChildren()
.inBackground(callback)
.forPath(path);
}
/**
* 异步-获取某个节点的所有子节点路径
*
* @param path 目录
* @param callback 回调
* @param executor 回调在哪里执行
* @throws Exception errors
*/
public void getChildrenAsync(String path, BackgroundCallback callback, Executor executor) throws Exception {
client.getChildren()
.inBackground(callback, executor)
.forPath(path);
}
二 Curator高级特性
Curator里面的curator-recipes ja包封装了一些高级特性,如:Cache事件监听、选举、分布式锁、分布式计数器、分布式Barrier等等。而且这些特性都是在分布式系统里面常用的功能了。
2.1 Cache事件监听
Zookeeper原生支持通过注册Watcher来进行事件监听,但是开发者需要反复注册(Watcher只能单次注册单次使用)。Cache是Curator中对事件监听的包装,可以看作是对事件监听的本地缓存视图,能够自动为开发者处理反复注册监听。Curator提供了三种Watcher(Cache)来监听结点的变化。
2.1.1 Path Cache
Path Cache用来监控子节点.当一个子节点增加, 更新,删除时, Path Cache会改变它的状态,会包含最新子节点的数据和状态,而状态的更变将通过PathChildrenCacheListener通知。
Path Cache的使用非常的简单,主要涉及到四个类:
- PathChildrenCache:Path Cache听实现类
- PathChildrenCacheEvent:子节点事件
- PathChildrenCacheListener: 子节点监听
- ChildData:子节点信息
关于Path Cache的使用,我们用一个实例来简单的说明下,实例里面也只是简单的创建了一个节点。最终监听到节点的创建.
@Test
public void pathChildrenCache() throws Exception {
// 创建zookeeper客户端
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
PathChildrenCache cache = new PathChildrenCache(client, "/tuacy/pathCache", true);
// 添加监听
cache.getListenable().addListener(new PathChildrenCacheListener() {
@Override
public void childEvent(CuratorFramework client, PathChildrenCacheEvent event) throws Exception {
System.out.println("事件类型:" + event.getType());
if (null != event.getData()) {
System.out.println("节点数据:" + event.getData().getPath() + " = " + new String(event.getData().getData()));
}
}
});
cache.start();
// 添加节点
client.create().creatingParentContainersIfNeeded().forPath("/tuacy/pathCache/001");
Uninterruptibles.sleepUninterruptibly(30, TimeUnit.SECONDS);
cache.close();
}
2.1.2 Node Cache
Node Cache与Path Cache类似,Node Cache只是监听某一指定的节点。子节点的变化它是不会管的。
Node Cache的使用涉及到下面的三个类:
- NodeCache - Node Cache实现类
- NodeCacheListener - 节点监听器
- ChildData - 节点数据
我们还是用一个简单的实例来说明。
@Test
public void nodeCache() throws Exception {
// 创建zookeeper客户端
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
final NodeCache cache = new NodeCache(client, "/tuacy/nodeCache");
cache.start();
// 添加监听
cache.getListenable().addListener(new NodeCacheListener() {
@Override
public void nodeChanged() throws Exception {
ChildData data = cache.getCurrentData();
if (null != data) {
System.out.println("节点数据:" + new String(cache.getCurrentData().getData()));
} else {
System.out.println("节点被删除!");
}
}
});
// 添加节点
client.create().creatingParentsIfNeeded().forPath("/tuacy/nodeCache");
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
client.setData().forPath("/tuacy/nodeCache", "abc".getBytes());
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
client.delete().forPath("/tuacy/nodeCache");
Uninterruptibles.sleepUninterruptibly(30, TimeUnit.SECONDS);
cache.close();
}
2.1.3 Tree Cache
Tree Cache可以监控整个树上的所有节点,就是PathCache和NodeCache的组合功能。
Tree Cache的使用涉及到下面四个类。
- TreeCache - Tree Cache实现类
- TreeCacheListener - 监听器类
- TreeCacheEvent - 触发的事件类
- ChildData - 节点数据
我们还是以具体的实例来说明Tree Cache的使用。
@Test
public void nodeCache() throws Exception {
// 创建zookeeper客户端
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
final TreeCache cache = TreeCache.newBuilder(client, "/tuacy/treeCache")
.setCacheData(true)
.build();
// 添加监听
cache.getListenable().addListener(new TreeCacheListener() {
@Override
public void childEvent(CuratorFramework client, TreeCacheEvent event) throws Exception {
System.out.println("事件类型:" + event.getType() + " | 路径:" + (null != event.getData() ? event.getData().getPath() : null));
}
});
cache.start();
// 添加节点
client.create().creatingParentsIfNeeded().forPath("/tuacy/treeCache");
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
// 给节点设置数据
client.setData().forPath("/tuacy/treeCache", "abc".getBytes());
// 创建子节点
client.create().creatingParentsIfNeeded().forPath("/tuacy/treeCache/001");
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
// 修改子节点的数据
client.setData().forPath("/tuacy/treeCache/001", "abc".getBytes());
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
// 删除子节点
client.delete().forPath("/tuacy/treeCache/001");
Uninterruptibles.sleepUninterruptibly(10, TimeUnit.SECONDS);
// 删除节点
client.delete().forPath("/tuacy/treeCache");
Uninterruptibles.sleepUninterruptibly(30, TimeUnit.SECONDS);
cache.close();
client.close();
}
2.2 Leader选举
在分布式系统中,选主是一个很常见的场景(Leader,Slaver真的真的是非常的常见)。
- 主节点是唯一的。
- 各个节点获取主节点的概率是一样的,一旦某个节点被选为了主节点(Leader),其他的从节点(Slaver)也要能感知到。
- 一旦主节点断开,其他的从节点重新选出一个主节点。
2.2.1 LeaderLatch
在不同的zookeeper客户端,使用了相同latch path的LeaderLatch,当中的一个最终会被选举为leader,可以通过hasLeadership方法查看LeaderLatch实例是否leade。也可以在LeaderLatchListener里面监听当前节点是否是leader。使用LeaderLatch的时候如果不想参与选举了要调用close()方法退出选举。
LeaderLatch api介绍
public class LeaderLatch {
/**
* 构造函数
*
* @param client CuratorFramework
* @param latchPath 路径,所有参与者同一个路径
*/
public LeaderLatch(CuratorFramework client, String latchPath);
public LeaderLatch(CuratorFramework client, String latchPath, String id);
public LeaderLatch(CuratorFramework client, String latchPath, String id, CloseMode closeMode);
/**
* 参与选举
*/
public void start() throws Exception;
/**
* 退出选举
*/
@Override
public void close() throws IOException;
/**
* 退出选举
* 关闭方式:SILENT : 静默关闭,不触发相关监听器、NOTIFY_LEADER :关闭时触发监听器
*/
public synchronized void close(CloseMode closeMode) throws IOException;
/**
* 添加监听器,监听是否当选为leader
*/
public void addListener(LeaderLatchListener listener);
public void addListener(LeaderLatchListener listener, Executor executor);
/**
* 移除监听器
*/
public void removeListener(LeaderLatchListener listener);
/**
* 尝试让当前LeaderLatch实例为leader
*/
public void await() throws InterruptedException, EOFException
public boolean await(long timeout, TimeUnit unit) throws InterruptedException;
/**
* 获取构造函数里面这是的id
*/
public String getId();
/**
* 获取当前LeaderLatch实例的状态
*/
public State getState();
/**
* 返回所有的参与者
*/
public Collection getParticipants() throws Exception;
/**
* 返回当前leader节点信息
*/
public Participant getLeader() throws Exception;
/**
* 判断实例是否是leader
*/
public boolean hasLeadership();
}
我们用一个简单的实例来说明LeaderLatch用法,比如我们创建10个zookeeper客户端来进行选举。
@Test
public void leaderLatch() throws Exception {
List zookeeperClientList = Lists.newArrayList();
List leaderLatchList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 这里我们所有的客户端都参与leader选举
for (int index = 0; index < zookeeperClientList.size(); index++) {
// 所有的客户端都参与leader选举
final LeaderLatch latch = new LeaderLatch(zookeeperClientList.get(index), LEADER_PATH, index + "");
latch.addListener(new LeaderLatchListener() {
@Override
public void isLeader() {
System.out.println("我是leader: " + latch.getId());
}
@Override
public void notLeader() {
System.out.println("我不是leader: " + latch.getId());
}
});
latch.start();
leaderLatchList.add(latch);
}
// 30S之后
Uninterruptibles.sleepUninterruptibly(30, TimeUnit.SECONDS);
// 我们找到谁是leader
String leaderId = leaderLatchList.get(0).getLeader().getId();
System.out.println("当前leader id : " + leaderId);
leaderLatchList.forEach(item -> {
// 这里我们吧leader退出选举,让剩下的重新选举
if (item.getId().equals(leaderId)) {
try {
item.close();
} catch (IOException e) {
e.printStackTrace();
}
}
});
Uninterruptibles.sleepUninterruptibly(1, TimeUnit.MINUTES);
leaderLatchList.forEach(curatorFramework -> {
// 退出选举
try {
curatorFramework.close();
} catch (IOException e) {
e.printStackTrace();
}
});
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
2.2.2 LeaderSelector
LeaderSelector也是一个用于分布式选举的类,相对于LeaderLatch来说,LeaderSelector更加的灵活点。LeaderSelector使用的时候主要涉及下面几个类:
- LeaderSelector:LeaderSelector选举实例类。
- LeaderSelectorListener:监听选举状态和连接状态
- LeaderSelectorListenerAdapter:实现了LeaderSelectorListener接口的一个抽象类,封装了客户端与zk服务器连接挂起或者断开时的处理逻辑(抛出抢主失败CancelLeadershipException),一般监听器推荐实现该类。
- CancelLeadershipException: 抢主失败异常。
LeaderSelector api 介绍
public class LeaderSelector {
/**
* 构造函数
* @param client CuratorFramework
* @param leaderPath 路径
* @param listener 监听器
*/
public LeaderSelector(CuratorFramework client, String leaderPath, LeaderSelectorListener listener);
public LeaderSelector(CuratorFramework client, String leaderPath, ExecutorService executorService, LeaderSelectorListener listener);
public LeaderSelector(CuratorFramework client, String leaderPath, CloseableExecutorService executorService, LeaderSelectorListener listener);
/**
* 保证在此实例释放领导权之后还可能获得领导权
*/
public void autoRequeue();
/**
* 设置获取当前实例对应的id
*/
public void setId(String id);
public String getId();
/**
* 当前实例参与选举
*/
public void start();
/**
* 重新键入到参与者队列里面去选举,如果此实例已在参与者排队里面,则不会发生任何操作并返回false。如果实例未排队,则重新执行该操作并返回true
*/
public boolean requeue();
/**
* 退出选举
*/
public synchronized void close();
/**
* 获取所有的参与者
*/
public Collection getParticipants() throws Exception;
/**
* 获取leader
*/
public Participant getLeader() throws Exception;
/**
* 当前节点是否是leader
*/
public boolean hasLeadership();
/**
* 如果当前实例是leader的话,尝试终断领导权
*/
public synchronized void interruptLeadership();
}
ConnectionStateListener、LeaderSelectorListener
public interface ConnectionStateListener {
/**
* 监听网络连接问题
*/
public void stateChanged(CuratorFramework client, ConnectionState newState);
}
/**
* Notification for leadership
*
* @see LeaderSelector
*/
public interface LeaderSelectorListener extends ConnectionStateListener {
/**
* 当前节点获取到leader权之后调用,注意:在您希望释放领导力之前,此方法不应返回
* 所以说如果你想一直占有leader权利,就在里面写个无限循环吧
*/
public void takeLeadership(CuratorFramework client) throws Exception;
}
LeaderSelectorListenerAdapter
/**
* 实现了LeaderSelectorListener接口的一个抽象类,封装了客户端与zk服务器连接挂起或者断开时的处理逻辑(抛出抢主失败CancelLeadershipException),一般监听器推荐实现该类
*/
public abstract class LeaderSelectorListenerAdapter implements LeaderSelectorListener {
/**
* 当遇到SUSPENDED以及LOST时直接抛出CancelLeadershipException从而去引发LeaderSelector.interruptLeadership()调用
*/
@Override
public void stateChanged(CuratorFramework client, ConnectionState newState) {
if ( client.getConnectionStateErrorPolicy().isErrorState(newState) ) {
throw new CancelLeadershipException();
}
}
}
我们还是用一个简单的实例来说明LeaderSelector的用法,我们还是创建10个zookeeper客户端。并且我们创建一个LeaderSelectorAdapter类,在里面当是leader之后的一些处理,如果是leader 10s之后,释放leader权力重新选举。
public class LeaderSelectorAdapter extends LeaderSelectorListenerAdapter {
private final LeaderSelector leaderSelector;
public LeaderSelectorAdapter(CuratorFramework client, String path, String id) {
// 创建一个LeaderSelector对象
leaderSelector = new LeaderSelector(client, path, this);
// 设置id
leaderSelector.setId(id);
// 保证在此实例释放领导权之后还可能获得领导权
leaderSelector.autoRequeue();
}
/**
* 参与选举
*/
public void start() {
// 参与选举
leaderSelector.start();
}
/**
* 退出选举
*/
public void close() {
// 退出选举
leaderSelector.close();
}
/**
* 当获得leader的时候,这个方法会被调用。如果还想继续当leader,这个方法不能返回。如果你想要要此实例一直是leader的话可以加一个死循环
*/
@Override
public void takeLeadership(CuratorFramework client) throws Exception {
System.out.println(leaderSelector.getId() + " 是leader");
try {
// 当上leader 5s之后,释放leader权利
Thread.sleep(TimeUnit.SECONDS.toMillis(10));
} catch (InterruptedException e) {
System.err.println(leaderSelector.getId() + " 被中断.");
Thread.currentThread().interrupt();
} finally {
System.out.println(leaderSelector.getId() + " 释放leader的权力。");
}
}
}
private static final String LEADER_PATH = "/tuacy/leaderSelector";
@Test
public void leaderSelector() throws Exception {
List zookeeperClientList = Lists.newArrayList();
List leaderLatchList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 这里我们所有的客户端都参与leader选举
for (int index = 0; index < zookeeperClientList.size(); index++) {
// 所有的客户端都参与leader选举
final LeaderSelectorAdapter latch = new LeaderSelectorAdapter(zookeeperClientList.get(index), LEADER_PATH, index + "");
latch.start();
leaderLatchList.add(latch);
}
// 1分钟之后关掉程序
Uninterruptibles.sleepUninterruptibly(1, TimeUnit.MINUTES);
leaderLatchList.forEach(curatorFramework -> {
// 退出选举
curatorFramework.close();
});
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
2.3 分布式锁
分布式锁也是咱们分布式系统里面非常常见的功能了。Curator直接就帮我们做到了,省的我们自己去实现分布式锁。
2.3.1 InterProcessMutex
InterProcessMutex公平锁、可重入锁。和ReentrantLock类似。
InterProcessMutex api 介绍
public class InterProcessMutex implements InterProcessLock, Revocable {
/**
* InterProcessMutex的构造函数,
*/
public InterProcessMutex(CuratorFramework client, String path);
public InterProcessMutex(CuratorFramework client, String path, LockInternalsDriver driver);
/**
* 申请获取锁
*/
@Override
public void acquire() throws Exception;
@Override
public boolean acquire(long time, TimeUnit unit) throws Exception;
/**
*
* 如果此JVM中的线程获取了互斥锁,则返回true
*/
@Override
public boolean isAcquiredInThisProcess();
/**
* 释放锁
*/
@Override
public void release() throws Exception;
/**
* 返回所有参与获取锁的所有当前节点的排序列表
*/
public Collection getParticipantNodes() throws Exception;
/**
* 将锁设为可撤销的. 当别的进程或线程想让你释放锁是Listener会被调用
*/
@Override
public void makeRevocable(RevocationListener listener);
@Override
public void makeRevocable(final RevocationListener listener, Executor executor);
/**
* 如果调用线程获取互斥锁,则返回true
*/
public boolean isOwnedByCurrentThread();
}
2.3.2 InterProcessSemaphoreMutex
InterProcessSemaphoreMutex不可重入锁。
InterProcessSemaphoreMutex api介绍
public class InterProcessSemaphoreMutex implements InterProcessLock {
/**
* 构造函数
*/
public InterProcessSemaphoreMutex(CuratorFramework client, String path);
/**
* 申请获取锁
*/
@Override
public void acquire() throws Exception;
@Override
public boolean acquire(long time, TimeUnit unit) throws Exception;
/**
* 释放锁
*/
@Override
public void release() throws Exception;
/**
* 如果此JVM中的线程获取了互斥锁,则返回true
*/
@Override
public boolean isAcquiredInThisProcess();
}
2.3.3 InterProcessReadWriteLock
InterProcessReadWriteLock 读写锁。和ReadWriteLock类似。
InterProcessReadWriteLock api 介绍
public class InterProcessReadWriteLock {
/**
* 读锁
*/
private final InterProcessMutex readMutex;
/**
* 写锁
*/
private final InterProcessMutex writeMutex;
/**
* 构造函数
*/
public InterProcessReadWriteLock(CuratorFramework client, String basePath)
/**
* 构造函数
* lockData是存储在节点上的数据
*/
public InterProcessReadWriteLock(CuratorFramework client, String basePath, byte[] lockData);
/**
* 获取读锁
*/
public InterProcessMutex readLock();
/**
* 获取写锁
*/
public InterProcessMutex writeLock();
}
2.3.4 信号量(InterProcessSemaphoreV2)
InterProcessSemaphoreV2实现了一个跨jvm的信号量,主要工作原理是:acquire时创建一个临时顺序节点,如果创建成功且临时节点数小于等于maxLeases则说明信号量获取成功,否则wait等待,等待目录发生变化或计数改变时唤醒。和Semaphore的功能类似.
分布式信号量的使用。我们需要了解以下三个类。
- InterProcessSemaphoreV2:信号量实现类
- Lease:租约(单个信号)
- SharedCountReader:计数器,用于计算最大租约数量
InterProcessSemaphoreV2 api 介绍
public class InterProcessSemaphoreV2 {
/**
* 构造函数
* @param client CuratorFramework
* @param path 节点路径
* @param maxLeases 允许此实例的最大租约数
*/
public InterProcessSemaphoreV2(CuratorFramework client, String path, int maxLeases);
/**
* 构造函数
* @param client CuratorFramework
* @param path 节点路径
* @param count 用于最大租约的共享计数
*/
public InterProcessSemaphoreV2(CuratorFramework client, String path, SharedCountReader count);
/**
* 此信号量创建的节点放置的数据,必须在调用其中一个acquire()方法之前调用它
*/
public void setNodeData(byte[] nodeData);
/**
* 返回参与信号量的所有当前节点的列表
*/
public Collection getParticipantNodes() throws Exception;
/**
* 关闭给定租约集合中的所有租约
*/
public void returnAll(Collection leases);
/**
* 关闭租约
*/
public void returnLease(Lease lease);
/**
* 获取租约,如果没有租约获取会一直阻塞直到获取到租约
*/
public Lease acquire() throws Exception;
public Lease acquire(long time, TimeUnit unit) throws Exception
/**
* 获取指定数量的租约,如果没有获取到制定数量租约会一直阻塞
*/
public Collection acquire(int qty) throws Exception;
public Collection acquire(int qty, long time, TimeUnit unit) throws Exception;
}
2.3.5 InterProcessMultiLock(多共享锁对象
它可以把多个锁包含起来像一个锁一样进行操作,简单来说就是对多个锁进行一组操作。当acquire的时候就获得多个锁资源,否则失败。同样调用release时所有的锁都被release(失败被忽略)。
InterProcessMultiLock api 介绍
public class InterProcessMultiLock implements InterProcessLock {
/**
* 构造函数
*
* @param client CuratorFramework
* @param paths 节点列表对应的路径(多个路径就是多个锁)
*/
public InterProcessMultiLock(CuratorFramework client, List paths);
/**
* 构造函数
*/
public InterProcessMultiLock(List locks);
/**
* 请求锁
*/
@Override
public void acquire() throws Exception;
@Override
public boolean acquire(long time, TimeUnit unit) throws Exception;
/**
* 释放锁
*/
@Override
public synchronized void release() throws Exception;
/**
* 如果此JVM中的线程获取了所有的锁,则返回true
*/
@Override
public synchronized boolean isAcquiredInThisProcess();
}
2.4 分布式计数器
计数器是用来计数的,利用ZooKeeper可以实现一个分布式计数器。只要使用相同的path就可以得到最新的计数器值,这是由ZooKeeper的一致性保证的。Curator有两个计数器,一个是用int来计数(SharedCount),一个用long来计数(DistributedAtomicLong)。
2.4.1 SharedCount(int计数器)
SharedCount使用int类型来计数。相当于多个zookeeper客户端公用一个计算器。
- SharedCount:计数器的具体实现。
- SharedCountListener:监听数据的改变。
SharedCount api 介绍
public class SharedCount implements Closeable, SharedCountReader, Listenable {
/**
* 构造函数
* @param client CuratorFramework
* @param path 计数器依赖的节点
* @param seedValue 如果当前节点对应的计数器没有值,就会用该值
*/
public SharedCount(CuratorFramework client, String path, int seedValue);
protected SharedCount(CuratorFramework client, String path, SharedValue sv);
/**
* 获取当前计数
*/
@Override
public int getCount();
/**
* 获取当前节点对应的版本信息
*/
@Override
public VersionedValue getVersionedValue();
/**
* 设置计数器的值
*/
public void setCount(int newCount) throws Exception;
/**
* 设置计数器的值,这里要注意如果当前版本的值在这个时刻有改变则设置不成功。CAS操作
*/
public boolean trySetCount(VersionedValue previous, int newCount) throws Exception;
/**
* 添加监听器
*/
@Override
public void addListener(SharedCountListener listener);
@Override
public void addListener(final SharedCountListener listener, Executor executor);
/**
* 移除监听器
*/
@Override
public void removeListener(SharedCountListener listener);
/**
* 启动
*/
public void start() throws Exception;
/**
* 结束
*/
@Override
public void close() throws IOException;
}
SharedCount使用实例。模拟了10个zookeeper客户端。每个客户端都加5次。最终结果50就对了。
public class SharedCountTest {
private static final String PATH_COUNTER = "/int/counter";
class CounterThread extends Thread {
private final CountDownLatch countDownLatch;
private final int threadIndex;
private final SharedCount counter;
CounterThread(SharedCount counter, int index, CountDownLatch countDownLatch) {
this.counter = counter;
this.threadIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
for (int index = 0; index < 5; index++) {
while (true) {
Uninterruptibles.sleepUninterruptibly(5, TimeUnit.SECONDS);
boolean success = counter.trySetCount(counter.getVersionedValue(), counter.getCount() + 1);
if (success) {
break;
}
}
}
} catch (Exception e) {
e.printStackTrace();
} finally {
try {
System.out.println("当前值为:" + counter.getCount());
counter.close();
} catch (Exception e) {
//ignore
}
countDownLatch.countDown();
}
}
}
@Test
public void sharedCount() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 如果节点存在,我们就删除节点
zookeeperClientList.get(0).delete().forPath(PATH_COUNTER);
for (int index = 0; index < zookeeperClientList.size(); index++) {
SharedCount sharedCount = new SharedCount(zookeeperClientList.get(index), PATH_COUNTER, 0);
sharedCount.addListener(new SharedCountListener() {
@Override
public void countHasChanged(SharedCountReader sharedCount, int newCount) throws Exception {
System.out.println("计数器值改变,现在的值为:" + newCount);
}
@Override
public void stateChanged(CuratorFramework client, ConnectionState newState) {
// 连接状态改变
}
});
sharedCount.start();
new CounterThread(sharedCount, index, countDownLatch).start();
}
countDownLatch.await();
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
}
2.4.2 DistributedAtomicLong(long计数器)
DistributedAtomicLong使用Long类型来计数。
DistributedAtomicLong api 介绍
public class DistributedAtomicLong implements DistributedAtomicNumber {
/**
* 构造函数(乐观锁模式)
*
* @param client CuratorFramework
* @param counterPath 节点路径
* @param retryPolicy 重试策略 -- 乐观加锁
*/
public DistributedAtomicLong(CuratorFramework client, String counterPath, RetryPolicy retryPolicy);
/**
* 构造函数,retryPolicy(乐观加锁)还没成功,则进行promotedToLock的方式以互斥的方式加锁重试 (排他锁模式)
*
* @param client CuratorFramework
* @param counterPath 节点路径
* @param retryPolicy 重试策略 -- 乐观加锁
* @param promotedToLock 排他锁策略
*/
public DistributedAtomicLong(CuratorFramework client, String counterPath, RetryPolicy retryPolicy, PromotedToLock promotedToLock);
/**
* 获取当前值
*/
@Override
public AtomicValue get() throws Exception
/**
* 强制设置计数值
*/
@Override
public void forceSet(Long newValue) throws Exception;
/**
* CAS更新(乐观锁模式更新)
*/
@Override
public AtomicValue compareAndSet(Long expectedValue, Long newValue) throws Exception;
/**
* 设置值
*/
@Override
public AtomicValue trySet(Long newValue) throws Exception;
/**
* 如果之前没有初始值,则把初始值设置进去
*/
@Override
public boolean initialize(Long initialize) throws Exception;
/**
* +1
*/
@Override
public AtomicValue increment() throws Exception;
/**
* -1
*/
@Override
public AtomicValue decrement() throws Exception;
/**
* 加一个指定的值
*/
@Override
public AtomicValue add(Long delta) throws Exception;
/**
* 键一个指定的值
*/
@Override
public AtomicValue subtract(Long delta) throws Exception;
}
DistributedAtomicLong怎么使用,直接给实例。也是模拟10个客户端,每个客户端增加5次。最终结果得到50就对了。
public class DistributedAtomicLongTest {
private static final String PATH_COUNTER = "/long/counter";
class CounterThread extends Thread {
private final CountDownLatch countDownLatch;
private final int threadIndex;
private final DistributedAtomicLong counter;
CounterThread(DistributedAtomicLong counter, int index, CountDownLatch countDownLatch) {
this.counter = counter;
this.threadIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
for (int index = 0; index < 5; index++) {
// 保证成功
while (true) {
AtomicValue value = counter.increment();
if (value.succeeded()) {
System.out.println("succeed: " + value.succeeded() + " value:" + value.postValue());
break;
}
Uninterruptibles.sleepUninterruptibly(3, TimeUnit.SECONDS);
}
}
} catch (Exception e) {
e.printStackTrace();
} finally {
countDownLatch.countDown();
}
}
}
@Test
public void distributedAtomicLong() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 如果节点存在,我们就删除节点
if (zookeeperClientList.get(0).checkExists().forPath(PATH_COUNTER) != null) {
zookeeperClientList.get(0).delete().forPath(PATH_COUNTER);
}
for (int index = 0; index < zookeeperClientList.size(); index++) {
// 乐观锁模式
DistributedAtomicLong count = new DistributedAtomicLong(zookeeperClientList.get(index), PATH_COUNTER, new RetryNTimes(10, 10));
boolean initializeSuccess = count.initialize(0L);
if (initializeSuccess) {
System.out.println("初始化成功");
} else {
System.out.println("初始化失败");
}
new CounterThread(count, index, countDownLatch).start();
}
countDownLatch.await();
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
}
2.5 分布式队列
2.5.4 简单队列 - SimpleDistributedQueue
SimpleDistributedQueue是一种简单队列,和jdk中队列类似,拥有offer()、take()方法。
SimpleDistributedQueue的使用是很简单的,所以我们就直接给出SimpleDistributedQueue的使用实例了。
public class SimpleDistributedQueueTest {
private static final String SIMPLE_DISTRIBUTED_QUEUE_PATH = "/SimpleDistributedQueue";
class QueueActionThread extends Thread {
private final SimpleDistributedQueue queue;
private final CountDownLatch countDownLatch;
private final int queueIndex;
QueueActionThread(SimpleDistributedQueue queue, int index, CountDownLatch countDownLatch) {
this.queue = queue;
this.queueIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
for (int index = 0; index < 5; index++) {
String message = "我是队列:" + queueIndex + " 的第-" + index + "-条消息";
this.queue.offer(message.getBytes());
}
Uninterruptibles.sleepUninterruptibly(5, TimeUnit.SECONDS);
for (int index = 0; index < 5; index++) {
byte[] queueItem = queue.take();
System.out.println("我是队列:" + queueIndex + " 我收到了:" + new String(queueItem));
}
} catch (Exception e) {
e.printStackTrace();
} finally {
countDownLatch.countDown();
}
}
}
@Test
public void simpleDistributedQueue() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
for (int index = 0; index < zookeeperClientList.size(); index++) {
SimpleDistributedQueue queue = new SimpleDistributedQueue(zookeeperClientList.get(index), SIMPLE_DISTRIBUTED_QUEUE_PATH);
new QueueActionThread(queue, index, countDownLatch).start();
}
countDownLatch.await();
// 关闭客户端
zookeeperClientList.forEach(CuratorFramework::close);
}
}
2.5.2 普通队列 - DistributedQueue
DistributedQueue是一种非常谱图的队列,没啥骚操作。
DistributedQueue的使用也是非常简单的,我们也直接给出DistributedQueue的使用实例。
public class DistributedQueueTest {
private static final String DISTRIBUTED_QUEUE_PATH = "/queue/distributedQueue";
class QueueActionThread extends Thread {
private final DistributedQueue queue;
private final CountDownLatch countDownLatch;
private final int queueIndex;
QueueActionThread(DistributedQueue queue, int index, CountDownLatch countDownLatch) {
this.queue = queue;
this.queueIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
this.queue.start();
for (int index = 0; index < 5; index++) {
queue.put("队列 " + queueIndex + " 发来的消息:" + index);
Uninterruptibles.sleepUninterruptibly(5, TimeUnit.SECONDS);
}
} catch (Exception e) {
e.printStackTrace();
} finally {
countDownLatch.countDown();
}
}
}
@Test
public void distributedQueue() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
for (int index = 0; index < zookeeperClientList.size(); index++) {
QueueBuilder queueBuild = QueueBuilder.builder(zookeeperClientList.get(index), index % 2 == 0 ? new ConsumerImp(index + "") : null, createQueueSerializer(), DISTRIBUTED_QUEUE_PATH);
DistributedQueue queue = queueBuild.buildQueue();
new QueueActionThread(queue, index, countDownLatch).start();
}
countDownLatch.await();
// 关闭客户端
zookeeperClientList.forEach(CuratorFramework::close);
}
/**
* 队列消息序列化实现类
*/
private static QueueSerializer createQueueSerializer() {
return new QueueSerializer() {
@Override
public byte[] serialize(String item) {
return item.getBytes();
}
@Override
public String deserialize(byte[] bytes) {
return new String(bytes);
}
};
}
private class ConsumerImp implements QueueConsumer{
private String consumerName;
public ConsumerImp(String consumerName) {
this.consumerName = consumerName;
}
@Override
public void consumeMessage(String message) throws Exception {
System.out.println(consumerName + " 收到消息: " + message);
}
@Override
public void stateChanged(CuratorFramework client, ConnectionState newState) {
}
}
}
2.5.3 带id的队列 - DistributedIdQueue
DistributedIdQueue相对于DistributedQueue来说就是队列里面的每个id都带有一个id。所以DistributedIdQueue可以根据id删除队列里面的数据。其他部分和DistributedQueue一样。实例我们就不写了。
2.5.4 优先级队列 - DistributedPriorityQueue
DistributedPriorityQueue是带有优先级的队列,优先级别高的先消费。使用和DistributedQueue是差不多的。实例我们就不写了。
2.5.4 延迟队列 - DistributedDelayQueue
DistributedDelayQueue是带有延时功能的队列。消息入队的时候可以指定延时时间。让该消息延时一段时间之后才可以被消费。用法和DistributedQueue差不多。就不写具体的实例代码了。
2.6 分布式屏障 - Barrier
分布式Barrier是这样一个功能:它会阻塞所有节点上的等待进程,直到某一个被满足, 然后所有的节点继续进行。
2.6.1 DistributedBarrier
DistributedBarrier允许多个分布式线程任务等待放行。直到有地方说放行则这些分布式线程进入执行任务。
DistributedBarrier api 介绍。
public class DistributedBarrier {
/**
* @param client CuratorFramework
* @param barrierPath barrier路径节点
*/
public DistributedBarrier(CuratorFramework client, String barrierPath);
/**
* 设置栅栏,它将阻塞在它上面等待的线程:
*/
public synchronized void setBarrier() throws Exception;
/**
* 设置栅栏
*/
public synchronized void removeBarrier() throws Exception;
/**
* 等待放行条件
*/
public synchronized void waitOnBarrier() throws Exception
public synchronized boolean waitOnBarrier(long maxWait, TimeUnit unit) throws Exception;
}
DistributedBarrier的使用。比如这里我们模拟了10个zookeeper客户端。等待放行。
public class DistributedBarrierTest {
private static final String BARRIER_PATH_COUNTER = "/barrier";
class LogicThread extends Thread {
private final CountDownLatch countDownLatch;
private final int threadIndex;
private final DistributedBarrier barrier;
LogicThread(DistributedBarrier barrier, int index, CountDownLatch countDownLatch) {
this.barrier = barrier;
this.threadIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
System.out.println("线程: " + threadIndex + " 请求进入");
// 阻塞等待
barrier.waitOnBarrier();
System.out.println("线程: " + threadIndex + " 成功进入");
} catch (Exception e) {
e.printStackTrace();
} finally {
countDownLatch.countDown();
}
}
}
@Test
public void distributedBarrier() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 如果节点存在,我们就删除节点
if (zookeeperClientList.get(0).checkExists().forPath(BARRIER_PATH_COUNTER) != null) {
zookeeperClientList.get(0).delete().forPath(BARRIER_PATH_COUNTER);
}
DistributedBarrier controlBarrier = new DistributedBarrier(zookeeperClientList.get(0), BARRIER_PATH_COUNTER);
controlBarrier.setBarrier();
for (int index = 0; index < zookeeperClientList.size(); index++) {
DistributedBarrier barrier = new DistributedBarrier(zookeeperClientList.get(index), BARRIER_PATH_COUNTER);
new LogicThread(barrier, index, countDownLatch).start();
}
Uninterruptibles.sleepUninterruptibly(30, TimeUnit.SECONDS);
controlBarrier.removeBarrier();
countDownLatch.await();
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
}
2.6.2 DistributedDoubleBarrier
DistributedDoubleBarrier:类似CyclicBarrier 。允许多个分布式线程等待,等线程个数达到了指定数量的时候,就可以同时执行或者同时退出了。
DistributedDoubleBarrier api 的使用
public class DistributedDoubleBarrier {
/**
* 构造函数,
* memberQty是成员数量,当enter()方法被调用时,成员被阻塞,直到所有的成员都调用了enter()
* 当leave()方法被调用时,它也阻塞调用线程,直到所有的成员都调用了leave()
*/
public DistributedDoubleBarrier(CuratorFramework client, String barrierPath, int memberQty);
/**
* 进入栅栏并且阻塞,直到所有的成员都进入
*/
public void enter() throws Exception;
public boolean enter(long maxWait, TimeUnit unit) throws Exception;
/**
* 退出栅栏并且阻塞,知道所有的成员都退出
*/
public synchronized void leave() throws Exception;
public synchronized boolean leave(long maxWait, TimeUnit unit) throws Exception;
}
DistributedDoubleBarrier的简单使用,我们模拟10个zookeeper客户端。当有五个说要执行或者退出的时候。我们就执行或者退出。
public class DistributedDoubleBarrierTest {
private static final String BARRIER_PATH_COUNTER = "/barrier";
class LogicThread extends Thread {
private final CountDownLatch countDownLatch;
private final int threadIndex;
private final DistributedDoubleBarrier barrier;
LogicThread(DistributedDoubleBarrier barrier, int index, CountDownLatch countDownLatch) {
this.barrier = barrier;
this.threadIndex = index;
this.countDownLatch = countDownLatch;
}
@Override
public void run() {
try {
Uninterruptibles.sleepUninterruptibly(5 * threadIndex, TimeUnit.SECONDS);
System.out.println("线程:" + threadIndex + " 请求进入");
barrier.enter();
System.out.println("线程:" + threadIndex + " 成功进入");
System.out.println("线程:" + threadIndex + " 请求离开");
barrier.leave();
System.out.println("线程:" + threadIndex + " 成功离开");
} catch (Exception e) {
e.printStackTrace();
} finally {
countDownLatch.countDown();
}
}
}
@Test
public void distributedDoubleBarrier() throws Exception {
CountDownLatch countDownLatch = new CountDownLatch(10);
List zookeeperClientList = Lists.newArrayList();
// 启动10个zookeeper客户端
for (int index = 0; index < 10; index++) {
RetryPolicy retryPolicy = new ExponentialBackoffRetry(1000, 3);
CuratorFramework client = CuratorFrameworkFactory.builder()
.connectString("127.0.0.1:2181")
.retryPolicy(retryPolicy)
.sessionTimeoutMs(6000)
.connectionTimeoutMs(6000)
.build();
// 启动客户端
client.start();
zookeeperClientList.add(client);
}
// 如果节点存在,我们就删除节点
if (zookeeperClientList.get(0).checkExists().forPath(BARRIER_PATH_COUNTER) != null) {
zookeeperClientList.get(0).delete().forPath(BARRIER_PATH_COUNTER);
}
for (int index = 0; index < zookeeperClientList.size(); index++) {
DistributedDoubleBarrier barrier = new DistributedDoubleBarrier(zookeeperClientList.get(index), BARRIER_PATH_COUNTER, 5);
new LogicThread(barrier, index, countDownLatch).start();
}
countDownLatch.await();
zookeeperClientList.forEach(curatorFramework -> {
// 关闭客户端
curatorFramework.close();
});
}
}
三 Spring Boot使用Curator
Spring Boot中使用Curator,我们要想办法创建一个zookeeper客户端.然后把这个客户端对象添加到Spring容器中去.这样我们就可以在各个地方拿到这个zookeeper客户端对象.
说先我们创建一个ZkClient类.这个ZkClient类就代码我们一个zookeeper客户端.
public class ZkClient {
private final Logger logger = LoggerFactory.getLogger(this.getClass());
/**
* zookeeper客户端实例
*/
private CuratorFramework client;
/**
* 服务器列表,格式host1:port1,host2:port2,...
*/
private String zookeeperServer;
/**
* 会话超时时间,单位毫秒,默认60000ms
*/
private int sessionTimeoutMs;
/**
* 连接创建超时时间,单位毫秒,默认60000ms
*/
private int connectionTimeoutMs;
/**
* 重试之间等待的初始时间
*/
private int baseSleepTimeMs;
/**
* 当连接异常时的重试次数
*/
private int maxRetries;
/**
* 为了实现不同的Zookeeper业务之间的隔离,有的时候需要为每个业务分配一个独立的命名空间
*/
private String namespace;
public void setZookeeperServer(String zookeeperServer) {
this.zookeeperServer = zookeeperServer;
}
public void setSessionTimeoutMs(int sessionTimeoutMs) {
this.sessionTimeoutMs = sessionTimeoutMs;
}
public void setConnectionTimeoutMs(int connectionTimeoutMs) {
this.connectionTimeoutMs = connectionTimeoutMs;
}
public void setBaseSleepTimeMs(int baseSleepTimeMs) {
this.baseSleepTimeMs = baseSleepTimeMs;
}
public void setMaxRetries(int maxRetries) {
this.maxRetries = maxRetries;
}
public void setNamespace(String namespace) {
this.namespace = namespace;
}
/**
* spring 自动调用,不需要我们主动调用
*/
public void init() {
// 创建客户端
// 重连规则
RetryPolicy retryPolicy = new ExponentialBackoffRetry(baseSleepTimeMs, maxRetries);
client = CuratorFrameworkFactory.builder()
.connectString(zookeeperServer)
.retryPolicy(retryPolicy)
.sessionTimeoutMs(sessionTimeoutMs)
.connectionTimeoutMs(connectionTimeoutMs)
.namespace(namespace)
.build();
// 启动客户端,连接服务器
client.start();
}
/**
* spring 自动调用,不需要我们主动调用
*/
public void stop() {
// 关闭客户端
client.close();
}
/**
* 获取 zookeeper 客户端对象
*
* @return CuratorFramework
*/
public CuratorFramework getClient() {
return client;
}
}
接下来我们把ZkClient添加到Srping容器里面去.而且这里我们把一些动态配置信息都放到了application.yml文件里面去了.
@Configuration
public class ZkConfiguration {
/**
* 服务器列表,格式host1:port1,host2:port2,...
*/
@Value("${zookeeper.server}")
private String zookeeperServer;
/**
* 会话超时时间,单位毫秒,默认60000ms
*/
@Value(("${zookeeper.sessionTimeoutMs}"))
private int sessionTimeoutMs;
/**
* 连接创建超时时间,单位毫秒,默认60000ms
*/
@Value("${zookeeper.connectionTimeoutMs}")
private int connectionTimeoutMs;
/**
* 当连接异常时的重试次数
*/
@Value("${zookeeper.maxRetries}")
private int maxRetries;
/**
* 重试之间等待的初始时间
*/
@Value("${zookeeper.baseSleepTimeMs}")
private int baseSleepTimeMs;
/**
* 为了实现不同的Zookeeper业务之间的隔离,有的时候需要为每个业务分配一个独立的命名空间
*/
@Value("${zookeeper.namespace}")
private String namespace;
@Bean(initMethod = "init", destroyMethod = "stop")
public ZkClient zkClient() {
ZkClient zkClient = new ZkClient();
zkClient.setZookeeperServer(zookeeperServer);
zkClient.setSessionTimeoutMs(sessionTimeoutMs);
zkClient.setConnectionTimeoutMs(connectionTimeoutMs);
zkClient.setMaxRetries(maxRetries);
zkClient.setBaseSleepTimeMs(baseSleepTimeMs);
zkClient.setNamespace(namespace);
return zkClient;
}
}
application.yml文件增加配置信息
# zeekeeper配置
zookeeper:
server: 127.0.0.1:2181 # 服务器列表,格式host1:port1,host2:port2,...
sessionTimeoutMs: 6000 # 会话超时时间,单位毫秒,默认60000ms
connectionTimeoutMs: 6000 # 连接创建超时时间,单位毫秒,默认60000ms
maxRetries: 3 # 当连接异常时的重试次数
baseSleepTimeMs: 1000 # 重试之间等待的初始时间
namespace: lock # 为了实现不同的Zookeeper业务之间的隔离,有的时候需要为每个业务分配一个独立的命名空间,不需要的时候可以去掉
这样我们就可以在我们项目里面的任何地方得到ZkClient对象了.我们可以在zookeeper客户端为所欲为了.
到此关于java zookeeper客户端Curator的使用部分就讲完了.文章中设计到的所有实例代码在 https://github.com/tuacy/java-study工程目录的zookeeper文件下面可以找到.