Zookeeper

简单介绍

Apache Zookeeper是开发和维护开源服务器的服务，它能够实现高度可靠的分布式协调。

安装Zookeeper(无需安装)

wget http://mirror.bit.edu.cn/apache/zookeeper/zookeeper-3.4.10/zookeeper-3.4.10.tar.gz
tar zxvf zookeeper-3.4.10.tar.gz

安装Zookeeper C扩展支持

cd zookeeper-3.4.8/src/c
./configure --prefix=/usr/
make
make install

安装php的zookeeper扩展

wget https://pecl.php.net/get/zookeeper-0.6.2.tgz
tar zxvf zookeeper-0.6.2.tgz
cd zookeeper-0.6.2
phpize
./configure --with-libzookeeper-dir=/usr/
$ make
$ make install

启动zookeeper server

# {zookeeperserver}代表你zookeeper解压的目录mkdir -p /project/zookeeper/demo/data#zoo.conf 是默认启动所加载的配置文件
cp /{zookeeperserver}/conf/zoo_sample.cfg /{zookeeperserver}/conf/zoo.cfg/{zookeeperserver}/bin/zkServer start

zookeeper client 测试

/{zookeeperserver}/bin/zkCli.sh -server 127.0.0.1:2181# [zk: 127.0.0.1:2181(CONNECTED) x] 客户端连接信息shell
[zk: 127.0.0.1:2181(CONNECTED) 3] ls /
[zookeeper]
[zk: 127.0.0.1:2181(CONNECTED) 4] create /test qkl001
Created /test
[zk: 127.0.0.1:2181(CONNECTED) 5] create /zgq zgq002
Created /zgq
[zk: 127.0.0.1:2181(CONNECTED) 6] ls /
[zookeeper, test, zgq]
[zk: 127.0.0.1:2181(CONNECTED) 5] delete /zgq
Created /zgq
[zk: 127.0.0.1:2181(CONNECTED) 7] quit

PHP下实践

zookeeper类信息

Zookeeper — Zookeeper类-- Zookeeper::addAuth — 指定应用程序凭据-- Zookeeper::connect — 创建与Zookeeper沟通的句柄-- Zookeeper::__construct — 创建与Zookeeper沟通的句柄-- Zookeeper::create — 同步创建节点-- Zookeeper::delete — 同步删除Zookeeper中的一个节点-- Zookeeper::exists — 同步检查Zookeeper节点的存在性-- Zookeeper::get — 同步获取与节点关联的数据。-- Zookeeper::getAcl — 同步地获取与节点关联的ACL。-- Zookeeper::getChildren — 同步列出节点的子节点-- Zookeeper::getClientId — 返回客户端会话ID，仅在连接当前连接时才有效（即最后观察者状态为ZooOnCeleDelphi状态）-- Zookeeper::getRecvTimeout — 返回此会话的超时，如果连接当前连接（只有上次观察者状态为ZooOnCeleTytStand状态）才有效。此值可能在服务器重新连接后发生更改。-- Zookeeper::getState — 获取Zookeeper连接的状态-- Zookeeper::isRecoverable — 检查当前的Zookeeper连接状态是否可以恢复-- Zookeeper::set — 设置与节点关联的数据-- Zookeeper::setAcl — 同步设置与节点关联的ACL-- Zookeeper::setDebugLevel — 设置库的调试级别-- Zookeeper::setDeterministicConnOrder — 启用/禁用仲裁端点顺序随机化-- Zookeeper::setLogStream — 设置库用于日志记录的流-- Zookeeper::setWatcher — 设置观察函数

客户端操作

zkCli.cmd# output 表示连接成功
WatchedEvent state:SyncConnected type:None path:null
[zk: localhost:2181(CONNECTED) 0]# 以下是操作命令
ls /
create /ztest 1

获取节点信息(Zookeeper::get)

# get.php$zoo = new Zookeeper('192.168.1.45:2181');
$r = $zoo->get( '/ztest');
var_dump($r);

获取节点信息(Zookeeper::get)并设置watcher

# get.php
class ZookeeperDemo extends Zookeeper
{public function watcher($type, $state, $key){var_dump($type);var_dump($state);var_dump($key);if ($type == 3) {var_dump($this->get('/zgetw'));// Watcher gets consumed so we need to set a new one$this->get('/zgetw', array($this, 'watcher'));}}}$zoo = new ZookeeperDemo('192.168.1.45:2181');
$zoo->get('/zgetw', [$zoo, 'watcher']);while (true) {echo '.';sleep(1);
}

客户端操作

set /ztest 2
set /ztest 3
set /ztest 5

获取节点信息(Zookeeper::get) 结果

php get.php#output
......int(3)
int(3)
string(6) "/ztest"
string(1) "2"
..int(3)
int(3)
string(6) "/ztest"
string(1) "3"
..int(3)
int(3)
string(6) "/ztest"
string(1) "4"
.int(3)
int(3)
string(6) "/ztest"
string(1) "5"

创建子节点

create /ztest/child1 c1
# output: Created /ztest/child1create /ztest/child2 c2
# output: Created /ztest/child2ls /ztest
# output: [child2, child1]

php下获取子节点

# getChild.php
$zookeeper = new Zookeeper('192.168.1.45:2181');
$path = '/ztest';
$r = $zookeeper->getchildren($path);
if (!empty($r)) {foreach ($r as $c){var_dump($c);}
}# php getChild.php
string(6) "child2"
string(6) "child1"

创建顺序节点

[zk: localhost:2181(CONNECTED) 22] create /stest 1
Created /stest
[zk: localhost:2181(CONNECTED) 23] create -s /stest/seq 1
Created /stest/seq0000000000
[zk: localhost:2181(CONNECTED) 24] create -s /stest/seq 1
Created /stest/seq0000000001
[zk: localhost:2181(CONNECTED) 25] create -s /stest/seq 1
Created /stest/seq0000000002
[zk: localhost:2181(CONNECTED) 26] create -s /stest/seq 1
Created /stest/seq0000000003
[zk: localhost:2181(CONNECTED) 27] create -s /stest/seq 1
Created /stest/seq0000000004
[zk: localhost:2181(CONNECTED) 28] create -s /stest/seq 1
Created /stest/seq0000000005
[zk: localhost:2181(CONNECTED) 29] create -s /stest/seq 1
Created /stest/seq0000000006
[zk: localhost:2181(CONNECTED) 30] create -s /stest/seq 1
Created /stest/seq0000000007
[zk: localhost:2181(CONNECTED) 31] create -s /stest/seq 2
Created /stest/seq0000000008
[zk: localhost:2181(CONNECTED) 32] create -s /stest/seq 2
Created /stest/seq0000000009
[zk: localhost:2181(CONNECTED) 33] create -s /stest/seq 2
Created /stest/seq0000000010
[zk: localhost:2181(CONNECTED) 34] create -s /stest/seq 2
Created /stest/seq0000000011
[zk: localhost:2181(CONNECTED) 35] create -s /stest/seq 2
Created /stest/seq0000000012
[zk: localhost:2181(CONNECTED) 36] create -s /stest/seq 2
Created /stest/seq0000000013
[zk: localhost:2181(CONNECTED) 37] create -s /stest/seq 2
Created /stest/seq0000000014

代码

$zookeeper = new Zookeeper('192.168.1.45:2181');
$aclArray = array(array('perms'  => Zookeeper::PERM_ALL,'scheme' => 'world','id'     => 'anyone',)
);
$path = '/t3';
//ZOO_EPHEMERAL = 1
//ZOO_SEQUENCE  = 2
//这里这里的flag=NULL,flag=0 表示创建永久节点，=1创建临时节点，=2创建seq 顺序节点
$realPath = $zookeeper->create($path, null, $aclArray, 1);
var_dump($realPath);

worker

一个利用顺序临时节点的leader迁移实现

# worker.php
<?php/*** Created by PhpStorm.* User: qkl* Date: 2018/8/27* Time: 14:25*/
class Worker
{const CONTAINER = '/worker_test';protected $acl = [['perms' => Zookeeper::PERM_ALL,'scheme' => 'world','id' => 'anyone']];private $isLeader = false;private $znode = '';private $prevNode = '';public function __construct($host = '', $watcher_cb = null, $recv_timeout = 10000){$this->zk = new Zookeeper($host, $watcher_cb, $recv_timeout);}public function register(){if (!$this->zk->exists(self::CONTAINER)) {$this->zk->create(self::CONTAINER, null, $this->acl);}# 创建一个临时的顺序节点$this->znode = $this->zk->create(self::CONTAINER . '/w-',null,$this->acl,Zookeeper::EPHEMERAL | Zookeeper::SEQUENCE);//获取顺序节点名，去除父路径$this->znode = str_replace(self::CONTAINER . '/', '', $this->znode);printf("我创建了节点: %s\n", self::CONTAINER . '/' . $this->znode);$this->prevNode = $this->getPrev();if (is_null($this->prevNode)) {$this->setLeader(true);} else {$this->zk->get(self::CONTAINER . "/" . $this->prevNode, [$this, 'watcher']);}}public function getPrev(){$workers = $this->zk->getChildren(self::CONTAINER);sort($workers);$size = count($workers);for ($i = 0; $i < $size; $i++) {if ($this->znode == $workers[$i] && $i > 0) {return $workers[$i - 1];}}return null;}public function watcher($type, $state, $key){$prevNode = $this->prevNode;printf("watchNode-getPrevious：%s\n", self::CONTAINER . "/" . $prevNode);if (!is_null($prevNode)) {if ($type == 2) {printf($prevNode . " had deleted\n");}if ($type == 3) {printf("我重新监控节点：%s\n", self::CONTAINER . "/" . $prevNode);$this->zk->get(self::CONTAINER . "/" . $prevNode, [$this, 'watcher']);}}}public function isLeader(){return $this->isLeader;}public function setLeader($flag){$this->isLeader = $flag;}public function run(){$this->register();while (true) {if ($this->isLeader()) {$this->leaderJob();} else {$this->watcherJob();}sleep(2);}}public function leaderJob(){echo "现在我是Leader\n";}public function watcherJob(){echo "我在监控：".(self::CONTAINER . "/" . $this->prevNode)."\n";}}$worker = new Worker('192.168.1.45:2181');
$worker->run();

尝试

# 终端1
php worker.php
# 终端2
php worker.php
# 终端3
php worker.php# 此处运行不会被节点变化不会被监控到

再次尝试

# zkCli
create /worker_test/w-
/worker_test/0000000020php worker.php# output
我创建了节点: /worker_test/w-0000000022
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
watchNode-getPrevious：/worker_test/w-0000000020
我重新监控节点：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
watchNode-getPrevious：/worker_test/w-0000000020
我重新监控节点：/worker_test/w-0000000020
2018-08-28 02:11:46,684:2486(0x7f28ed0a1700):ZOO_WARN@zookeeper_interest@1570: Exceeded deadline by 15ms
我在监控：/worker_test/w-0000000020
我在监控：/worker_test/w-0000000020
watchNode-getPrevious：/worker_test/w-0000000020
w-0000000020 had deleted
我在监控：/worker_test/w-0000000020

这边发现一个类似bug问题

如果我们直接运行worker.php
在worker.php运行创建的临时顺序节点是不会被watcher到的
我们必须先首先创建好相关的节点再启动监控，不知道这里是不是php版本的zookeeper的bug
有了解的小伙伴可以告之下

Watcher通知状态与事件类型一览

ZOO_CREATED_EVENT(value=1)：节点创建事件，需要watch一个不存在的节点，当节点被创建时触发，此watch通过zoo_exists()设置
ZOO_DELETED_EVENT(value=2)：节点删除事件，此watch通过zoo_exists()或zoo_get()设置
ZOO_CHANGED_EVENT(value=3)：节点数据改变事件，此watch通过zoo_exists()或zoo_get()设置
ZOO_CHILD_EVENT(value=4)：子节点列表改变事件，此watch通过zoo_get_children()或zoo_get_children2()设置
ZOO_SESSION_EVENT(value=-1)：会话事件，客户端与服务端断开或重连时触发
ZOO_NOTWATCHING_EVENT(value=-2)：watch移除事件，服务端出于某些原因不再为客户端watch节点时触发