背景
前段时间看了S4流计算引擎,里面使用到了zookeeper进行集群管理,所以也就花了点时间研究了下zookeeper,不求看懂所有源码,但求了解其实现机制和原理,清楚其基本使用。这也是为后续hadoop,gridgain的分布式计算的产品。
学习
首先就是收集一些前人的一些学习资料和总结内容,方便自己快速入门。
这里罗列了几篇不错的文章:
- http://www.ibm.com/developerworks/cn/opensource/os-cn-zookeeper/ (介绍了zookeeper能用来干嘛)
- http://zookeeper.apache.org/doc/r3.3.2/zookeeperOver.html (官方文档,大致介绍zookeeper)
- 统一命名空间(Name Service)
- 配置推送 (Watch)
- 集群管理(Group membership)
统一命名空间
在zookeeper中实现了一个类似file system系统的数据结构,比如/zookeeper/status。 每个节点都对应于一个znode节点。
znode节点的数据结构模型:
znode的数据结构内容:
-
czxid
The zxid of the change that caused this znode to be created.
-
mzxid
The zxid of the change that last modified this znode.
-
ctime
The time in milliseconds from epoch when this znode was created.
-
mtime
The time in milliseconds from epoch when this znode was last modified.
-
version
The number of changes to the data of this znode.
-
cversion
The number of changes to the children of this znode.
-
aversion
The number of changes to the ACL of this znode.
-
ephemeralOwner
The session id of the owner of this znode if the znode is an ephemeral node. If it is not an ephemeral node, it will be zero.
-
dataLength
The length of the data field of this znode.
-
numChildren
The number of children of this znode.
说明: zxid (ZooKeeper Transaction Id,每次请求对应一个唯一的zxid,如果zxid a < zxid b ,则可以保证a一定发生在b之前)。
针对树状结构的处理,来看一下客户端使用的api :
- String create(String path, byte data[], List<ACL> acl, CreateMode createMode)
- void create(String path, byte data[], List<ACL> acl, CreateMode createMode, StringCallback cb, Object ctx)
- void delete(String path, int version)
- void delete(String path, int version, VoidCallback cb, Object ctx)
- Stat setData(String path, byte data[], int version)
- void setData(String path, byte data[], int version, StatCallback cb, Object ctx)
- Stat setACL(String path, List<ACL> acl, int version)
- void setACL(String path, List<ACL> acl, int version, StatCallback cb, Object ctx)
- Stat exists(String path, Watcher watcher)
- Stat exists(String path, boolean watch)
- void exists(String path, Watcher watcher, StatCallback cb, Object ctx)
- void exists(String path, boolean watch , StatCallback cb, Object ctx)
- byte[] getData(String path, Watcher watcher, Stat stat)
- byte[] getData(String path, boolean watch , Stat stat)
- void getData(String path, Watcher watcher, DataCallback cb, Object ctx)
- void getData(String path, boolean watch , DataCallback cb, Object ctx)
- List<String> getChildren(String path, Watcher watcher)
- List<String> getChildren(String path, boolean watch )
- void getChildren(String path, Watcher watcher, ChildrenCallback cb, Object ctx)
- void getChildren(String path, boolean watch , ChildrenCallback cb, Object ctx)
- List<String> getChildren(String path, Watcher watcher, Stat stat)
- List<String> getChildren(String path, boolean watch , Stat stat)
- void getChildren(String path, Watcher watcher, Children2Callback cb, Object ctx)
- void getChildren(String path, boolean watch , Children2Callback cb, Object ctx)
说明:每一种按同步还是异步,添加指定watcher还是默认watcher又分为4种。默认watcher可以在ZooKeeper zk = new ZooKeeper(serverList, sessionTimeout, watcher)中进行指定。如果包含boolean watch的读方法传入true则将默认watcher注册为所关注事件的watch。如果传入false则不注册任何watch
CreateMode主要有几种:
- PERSISTENT (持续的,相比于EPHEMERAL,不会随着client session的close/expire而消失)
- PERSISTENT_SEQUENTIAL
- EPHEMERAL (短暂的,生命周期依赖于client session,对应session close/expire后其znode也会消失)
- EPHEMERAL_SEQUENTIAL (SEQUENTIAL意为顺序的)
- StringCallback
- VoidCallback
- StatCallback
- DataCallback (getData请求)
- ChildrenCallback
- Children2Callback
配置推送(Watcher)
zookeeper为解决数据的一致性,使用了Watcher的异步回调接口,将服务端znode的变化以事件的形式通知给客户端,主要是一种反向推送的机制,让客户端可以做出及时响应。比如及时更新后端的可用集群服务列表。
这里有篇文章介绍Watcher/Callback比较详细,可以参考下:
- http://luzengyi.blog.163.com/blog/static/529188201064113744373/
- http://luzengyi.blog.163.com/blog/static/529188201061155444869/
如果想更好的理解Watcher的使用场景,可以了解下使用Watcher机制实现分布式的Barrier , Queue , Lock同步。
Barrier例子:
- public class Barrier implements Watcher {
- private static final String addr = "10.20.156.49:2181";
- private ZooKeeper zk = null;
- private Integer mutex;
- private int size = 0;
- private String root;
- public Barrier(String root, int size){
- this.root = root;
- this.size = size;
- try {
- zk = new ZooKeeper(addr, 10 * 1000, this);
- mutex = new Integer(-1);
- Stat s = zk.exists(root, false);
- if (s == null) {
- zk.create(root, new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
- }
- } catch (Exception e) {
- e.printStackTrace();
- }
- }
- public synchronized void process(WatchedEvent event) {
- synchronized (mutex) {
- mutex.notify();
- }
- }
- public boolean enter(String name) throws Exception {
- zk.create(root + "/" + name, new byte[0], Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL);
- while (true) {
- synchronized (mutex) {
- List<String> list = zk.getChildren(root, true);
- if (list.size() < size) {
- mutex.wait();
- } else {
- return true;
- }
- }
- }
- }
- public boolean leave(String name) throws KeeperException, InterruptedException {
- zk.delete(root + "/" + name, 0);
- while (true) {
- synchronized (mutex) {
- List<String> list = zk.getChildren(root, true);
- if (list.size() > 0) {
- mutex.wait();
- } else {
- return true;
- }
- }
- }
- }
- }
测试代码:
- public class BarrierTest {
- public static void main(String args[]) throws Exception {
- for (int i = 0; i < 3; i++) {
- Process p = new Process("Thread-" + i, new Barrier("/test/barrier", 3));
- p.start();
- }
- }
- }
- class Process extends Thread {
- private String name;
- private Barrier barrier;
- public Process(String name, Barrier barrier){
- this.name = name;
- this.barrier = barrier;
- }
- @Override
- public void run() {
- try {
- barrier.enter(name);
- System.out.println(name + " enter");
- Thread.sleep(1000 + new Random().nextInt(2000));
- barrier.leave(name);
- System.out.println(name + " leave");
- } catch (Exception e) {
- e.printStackTrace();
- }
- }
- }
通过该Barrier,可以协调不同任务之间的同步处理,这里主要还是利用了Watcher机制的反向推送,避免客户端的循环polling动作,只要针对有事件的变化做一次响应。
集群管理
我不罗嗦,taobao有几篇文章已经介绍的很详细。
- http://rdc.taobao.com/blog/cs/?p=162 (paxos 实现)
- http://rdc.taobao.com/blog/cs/?p=261 (paxos算法介绍续)
- http://rdc.taobao.com/team/jm/archives/448 (zookeeper代码解析)
- Leader
- Follower
- Obserer
- server.1:localhost:2181:3181:observer
3. 可通过命令行,查看当前server所处的状态
- [ljh@ccbu-156-49 bin]$ echo stat | nc localhost 2181
- Zookeeper version: 3.3.3--1, built on 06/24/2011 13:12 GMT
- Clients:
- /10.16.4.30:34760[1](queued=0,recved=632,sent=632)
- /127.0.0.1:43626[0](queued=0,recved=1,sent=0)
- /10.16.4.30:34797[1](queued=0,recved=2917,sent=2917)
- Latency min/avg/max: 0/0/33
- Received: 3552
- Sent: 3551
- Outstanding: 0
- Zxid: 0x200000003
- Mode: follower ##当前模式
- Node count: 8
使用zookeeper,我们能干些什么?
官方文档中,有举了几个应用场景,就是使用zookeeper提供分布式锁机制,从而实现分布式的一致性处理。
典型的几个场景:
- Barrier
- Queue
- Lock
- 2PC
其他
zookeeper基本是基于API和console进行znode的操作,并没有一个比较方便的操作界面,这里也发现了taobao 伯岩写的一个工具,可以比较方便的查询zookeeper信息。
工具的开发语言主要是node.js(最近比较火),其标榜的是无阻塞的api使用。其原理主要是基于google的V8(chrome的javascript的解析器,C语言编写),node.js本身是基于js语法进行开发,通过V8解析为C语言的执行代码
其标榜的无阻塞I/O实现,那可想而知就是linux系统下的select/poll的I/O模型。有兴趣的可以看下node.js的官网,下载一个玩玩。
文档地址: http://www.blogjava.net/killme2008/archive/2011/06/06/351793.html
代码地址: https://github.com/killme2008/node-zk-browser
通过git下载源码后,需要安装下node.js的几个模块express, express-namespace, zookeeper。 node.js下有个比较方便的模块管理器npm,类似于redhat的rpm,ubuntu的apt-get。
安装模块:
- npm install -g express
几个界面: