Redsi自定义客户端分片实现

今天设计一个基于客户端的分片缓存方案...比如如果业务量不是很大的话,这种情况下是可以自己实现分片逻辑的,这样出了问题,也比较好定位嘛,接下来我们就分析并实现一种这样的方案:

主要包括缓存分片和读写分离以及失效转移,三个功能

缓存分片:

& 随机分片

& HASH一致性分片

读写分离:

& 随机分片

& 轮询分片

首先看一下项目整体结构:

代码整体不复杂也不多:看看核心接口

& 分片接口

package com.redis.shard.strategy;

/**
 * @author 18011618
 * 基于分片策略接口
 * 默认实现机制基于关键字的hash取模
 */
public interface ShardingStrategy {
   /**
    * 基于hash算法
    * @param key 输入的字符串
    * @param nodeCount 节点数量
    * @return
    */
   public <T> int shard(T key, int nodeCount);
}

& 读写分离接口

package com.redis.shard.strategy;

/**
 * @author 18011618
 * 用来提供读写分离策略的接口
 * 基于轮询
 * 基于随机
 */
public interface SelectShardStrategy {
   
   /**
    * 读写分离机制
    * @param nodeCount
    * @return
    */
   public int selectShard(int nodeCount);
}

看看具体实现:

package com.redis.shard.strategy.impl;

import com.redis.shard.strategy.ShardingStrategy;

/**
 * @author 18011618
 * 实现具体的hash分片
 */
public class HashShardingStrategy implements ShardingStrategy {
   @Override
   public <T> int shard(T key, int nodeCount) {
      return key.hashCode() % nodeCount;
   }
   
}

随机选择:

package com.redis.shard.strategy.impl;

import java.util.Date;
import java.util.Random;

import com.redis.shard.strategy.SelectShardStrategy;

/**
 * @author 18011618
 * 基于随机实现策略
 */
public class RandomSelectStrategy implements SelectShardStrategy{
   
   private Random random = new Random(new Date().getTime());
   
   @Override
   public int selectShard(int nodeCount) {
      return random.nextInt(nodeCount);
   }
   
}

轮询策略:

 
package com.redis.shard.strategy.impl;

import java.util.concurrent.atomic.AtomicLong;

import com.redis.shard.strategy.SelectShardStrategy;
/**
 * 基于轮询的策略
 * @author 18011618
 */
public class RoundRobinSelectStrategy implements SelectShardStrategy {
   private AtomicLong atomic = new AtomicLong(0);
   @Override
   public int selectShard(int nodeCount) {
      long value = atomic.incrementAndGet();
      if(value == Long.MAX_VALUE){
         atomic.set(0);
      }
      return (int)value % nodeCount;
   }
   
}

接下来看看分片节点信息的封装:

package com.redis.shard;

import java.util.ArrayList;
import java.util.List;

import org.apache.commons.pool2.impl.GenericObjectPoolConfig;

import redis.clients.jedis.JedisPool;

import com.redis.shard.strategy.SelectShardStrategy;
import com.redis.shard.strategy.impl.RandomSelectStrategy;
import com.redis.shard.strategy.impl.RoundRobinSelectStrategy;

/**
 * @author 18011618
 * 封装客户端缓存分片的节点信息
 */
public class RedisShardNode {
   //191.168.1.1:6350,191.168.1.2:6350
   public static final String NODE_SEPARATOR = ",";
   //191.168.1.1:6350
   public static final String HOST_PORT_SEPARATOR = ":";

   //主节点
   private JedisPool master;
   //从节点
   private List<JedisPool> slaves;
   
   //读写分离机制 如果开启了读写分离会启用该策略
   public SelectShardStrategy selectStrategy;
   
   //指定默认读写策略为轮询
   public RedisShardNode( JedisPool master, List<JedisPool> slaves){
      this.master = master;
      this.slaves = slaves;
      this.selectStrategy = new RandomSelectStrategy();
   }
   
   //根据用户传递的读写策略
   public RedisShardNode(JedisPool master,List<JedisPool>slaves,SelectShardStrategy selectShardStrategy){
      this.master = master;
      this.slaves = slaves;
      this.selectStrategy = selectShardStrategy;
   }
   //通过参数实例化+指定默认策略
   public RedisShardNode(String masterConnStr, List<String> slavesConnStrs){
      String[] masterHostPortArray = masterConnStr.split(HOST_PORT_SEPARATOR);
      //设置master
      this.master = new JedisPool(new GenericObjectPoolConfig(),
            masterHostPortArray[0], Integer.valueOf(masterHostPortArray[1]));
      
      //设置slaves
      this.slaves = new ArrayList<JedisPool>();
      
      for (String slaveConnStr : slavesConnStrs) {
         String[] slaveHostPortArray = slaveConnStr
               .split(HOST_PORT_SEPARATOR);
         this.slaves.add(new JedisPool(new GenericObjectPoolConfig(),
               slaveHostPortArray[0], Integer
                     .valueOf(slaveHostPortArray[1])));
      }
      this.selectStrategy = new RoundRobinSelectStrategy();
   }
   //实例化参数+传递策略
   public RedisShardNode(String masterConnStr, List<String> slavesConnStrs,SelectShardStrategy selectShardStrategy){
      this(masterConnStr,slavesConnStrs);
      this.selectStrategy = selectShardStrategy;
   }

   public JedisPool getMaster() {
      return master;
   }

   public void setMaster(JedisPool master) {
      this.master = master;
   }

   public List<JedisPool> getSlaves() {
      return slaves;
   }

   public void setSlaves(List<JedisPool> slaves) {
      this.slaves = slaves;
   }
   //获取一个从节点
   public JedisPool getSlaveRedisShardNode() {
      int nodeIndex = selectStrategy.selectShard(slaves.size());
      return slaves.get(nodeIndex);
   }
}

最后看看客户端分片实现机制:因为要基于JEDISs所以需要基础Jedis

package com.redis.shard;

import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.util.StringUtils;

import redis.clients.jedis.Jedis;

import com.redis.shard.strategy.ShardingStrategy;
import com.redis.shard.strategy.impl.HashShardingStrategy;

/**
 * @author 18011618
 * 封装对外的redis分片客户端使用方法
 */
public class RedisShardClient extends Jedis{
   
   protected final Logger log = LoggerFactory.getLogger(RedisShardClient.class);
   //存储分片节点
   private List<RedisShardNode> redisShardNodes = new ArrayList<RedisShardNode>();
   //分片策略
   private ShardingStrategy shardingStategy = new HashShardingStrategy();
   //是否开启读写
   private boolean readWriteSeparate = false;
   //连接字符串
   private List<String> nodeConnStrs;
   
   //一定要有这个函数 否则反射的时候会报错
   public RedisShardClient(){
      
   }
   
   //传递集群连接字符串
   public RedisShardClient(List<String> nodeConnStrs){
      if (StringUtils.isEmpty(nodeConnStrs)) {
         log.error("The nodeConnStrs {} for Redic is invalid.", nodeConnStrs);
         throw new IllegalArgumentException(
               "The nodeConnStrs for Redic is invalid.");
      }
      this.nodeConnStrs = nodeConnStrs;
      //调用初始化方法
      init();
   }
   
   /**
    * 初始化分片节点
    */
   public void init(){
      for(String nodeStr:nodeConnStrs){
         this.addRedisNode(nodeStr);
      }
   }
   
   private RedisShardClient addRedisNode(String masterConnStr, List<String> slaveConnStrs){
      redisShardNodes.add(new RedisShardNode(masterConnStr, slaveConnStrs));
      return this;
   }
   public RedisShardClient addRedisNode(String nodeConnStr) {
      String[] nodes = nodeConnStr.split(RedisShardNode.NODE_SEPARATOR);
      return addRedisNode(nodes[0], Arrays.asList(Arrays.copyOf(nodes, 1)));
   }
   

   /**
    * read方法 往主节点
    */
   public <T> Jedis getRead(T key){
      int nodeIndex = shardingStategy.shard(key, redisShardNodes.size());
      RedisShardNode node = redisShardNodes.get(nodeIndex);
      //如果开启了读写分离
      if(!readWriteSeparate)
         return node.getMaster().getResource();//从主节点
      //随机选择一个从节点进行读取
      return node.getSlaveRedisShardNode().getResource();
   }
   
   /**
    * write方法
    */
   public <T> Jedis getWrite(T key){
      int nodeIndex = shardingStategy.shard(key, redisShardNodes.size());
      RedisShardNode node = redisShardNodes.get(nodeIndex);
      return node.getMaster().getResource();
   }
   
   /**
    * 重写get方法
    */
   @Override
   public String get(String key) {
      Jedis jedis = getRead(key);
      String result = jedis.get(key);
      jedis.close();
      return result;
   }

   /**
    * 重写set方法
    */
   @Override
   public String set(String key, String value) {
      Jedis jedis =getWrite(key);
      String result = jedis.set(key, value);
      jedis.close();
      return result;
   }

}

最后看一下配置文件:

<?xml version="1.0" encoding="UTF-8"?>
<beans xmlns="http://www.springframework.org/schema/beans"
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
      xsi:schemaLocation="http://www.springframework.org/schema/beans http://www.springframework.org/schema/beans/spring-beans-3.0.xsd">

   <bean id="redisShardClient" class="com.redis.shard.RedisShardClient" init-method="init">
      <!--配置读写分离开关 -->
      <property name="readWriteSeparate" value="${redis.readWrite}"/>
      <property name="nodeConnStrs">
      <!-- 配置多个节点 -->
         <list>
            <value>${reids.node1}</value>
<!--            <value>${reids.node2}</value>
 -->         </list>
      </property>
   </bean>

</beans>

OK，再写一个测试类:

import org.springframework.context.ApplicationContext;
import org.springframework.context.support.ClassPathXmlApplicationContext;

import com.redis.shard.RedisShardClient;
@SuppressWarnings("all")
public class RediShardTest  {
   public static void main(String[] args) {
      String path ="/spring/spring_redis_shard_client_test.xml";
      ApplicationContext context = new ClassPathXmlApplicationContext();
      RedisShardClient redisShardClient = (RedisShardClient)context.getBean("redisShardClient");
      redisShardClient.set("uname", "jhp");
      
      String uname = redisShardClient.get("uname");
      System.out.println("uname:"+uname);
   }
}

对应的配置文件:

具体配置文件内容:

<?xml version="1.0" encoding="UTF-8"?>

<beans xmlns="http://www.springframework.org/schema/beans"
       xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
       xsi:schemaLocation="http://www.springframework.org/schema/beans http://www.springframework.org/schema/beans/spring-beans-3.0.xsd">

<bean
        class="org.springframework.beans.factory.config.PropertyPlaceholderConfigurer">
    <property name="locations" value="classpath:spring/redic.properties"/>
</bean>

<import resource="classpath:spring/spring_redis_client_shard.xml"/>
</beans>

以及属性配置文件:

reids.node1=localhost:6379,localhost:6379
redis.readWrite=true 

到此一个简单的基于客户端分片功能就实现了,大家可以增加更加复杂的策略完成更多的功能,比如HASH一致性,这个在dubbo的

cluster模块中loadbalance是有实现,可以去参考