The author previous blog introduced the Redis "master-slave replication" to build services and "sentinel mode", but both services there are some issues remain unresolved, such as:
- Single Master write performance bottlenecks
- Data storage can not be distributed, limited stand-alone memory
To solve these problems, Redis also provides a more scalable, better availability of cluster Cluster program.
What Redis Cluster A cluster is?
Redis Cluster distributed server cluster is a cluster of a plurality of nodes in the cluster from the master composition which has a characteristic data replication, and availability of data pieces stored.
Redis Cluster Cluster Sentinel Sentinel does not need to be removed to complete the function node and failover.
Each node needs to be set to cluster mode, this mode is not the center of the cluster nodes, horizontally scalable, the official said the document can be linearly extended to 1000 nodes.
Redis Cluster cluster performance and availability are better than "master-slave replication" and "sentry mode" configuration is also very simple, more suitable for enterprise services.
Cluster Cluster native structures
FIG architecture
shown above, this environment building three Master, 6 sets Slave.
Redis default there are 16,384 slots, these slots must be assigned to the Master of all, otherwise it will lead to some data can not be written.
When data is written, the Redis CRC16 algorithm used to calculate a number Key, a hash code is similar to the value determined with a slot modulo 16384, and writes data to the Master slot the node resides.
After the Master to write data, the data will be synchronized to the corresponding Slave node, i.e. from the master copy.
After a single Master down, will automatically fail the Redis, a Slave to Master promoted under the new Master.
实际应用中Redis节点需要部署到不同的物理机，笔者没有那么多的物理机，遂用单机的不同端口跑多个Redis节点服务。
测试环境：8001、8004、8007为Master。
8002、8003 slaveof 8001
8005、8006 slaveof 8004
8008、8009 slaveof 8007
几个基本命令
- cluster info：查看集群信息
- cluster nodes：查看节点信息
- cluster meet ip port：将节点加入到工作集群
- cluster keyslot key：计算Key的CRC16算法结果值
- cluster addslots slot (槽下标)：设置槽位
- cluster replicate (master node-id)：将节点设为指定Master的Slave
1、创建配置文件
拷贝Redis提供的默认配置文件，做一些常规修改(不赘述)，搭建集群主要修改如下：
# 启用Cluster集群
cluster-enabled yes
# 保存集群信息的文件,每台节点都会同步集群信息并保存到各自的文件中
cluster-config-file nodes-8001.conf
2、依次启动Master节点
redis-server 8001/redis.conf
redis-server 8004/redis.conf
redis-server 8007/redis.conf
查看服务是否启动成功：
注意：这时候节点是无法写入数据的，因为没有设置哈希槽Slot，数据不知道该写入哪个节点，如下图：
3、集群meet
集群Master节点都虽然已经启动，但是目前节点之间互相还没有关联，互相都不认识，还没有形成集群。
查看集群信息，发现只有自身节点。
cluster meet ip port
通过该命令来让节点互相认识，并形成网状结构。
cluster meet 192.168.1.120 8004
cluster meet 192.168.1.120 8007
重新查看集群信息
4、启动Slave节点
除了Master，还得有Slave节点，否则无法实现高可用和数据备份。
启动Slave节点和Master步骤一样，这里不再赘述，启动完成后如下图所示：
5、主从分配
启动的节点若不做配置，默认都是Master，需要进行主从分配。
主从分配之前，需要将新启动的Slave节点meet一下，先让节点加入到集群。
cluster replicate node-id通过该命令来分配主从。
登录到Slave节点执行该命令，node-id为Master节点的id。
查看节点的node-id
将Slave节点分配到Master节点下。(演示8002、8003，其他不赘述)
redis-cli -p 8002 cluster replicate 22bbad3e9cd6065861f13b95cef5d437483c6360
OK
redis-cli -p 8003 cluster replicate 22bbad3e9cd6065861f13b95cef5d437483c6360
OK
全部分配完成后，如下图所示：
6、指派槽位
前面已经说过，没有为Master指定Slot槽位，是无法写入数据的。
cluster keyslot key该命令可以查看指定Key的CRC16计算的值。
Redis集群默认有16384个槽位，现有3台Master节点，平均进行分配。
- Master-8001：0~5461
- Master-8004：5462~10922
- Master-8007：10923~16383
cluster addslots slot [槽位下标]该命令用来设置节点的槽位。
为了快速分配，写了个Shell脚本，如下：
#!/bin/sh
# 读取端口
echo -n 'port:'
read port
#槽位起始下标
echo -n 'start:'
read s
#槽位结束下标
echo -n 'end:'
read e
#循环设置槽位
for i in $(seq $s $e)
do
redis-cli -h 192.168.1.120 -p $port cluster addslots $i
echo redis-cli -h 192.168.1.120 -p $port cluster addslots $i;
done
运行脚本，批量设置槽位，设置完成后查看如下图：
7、集群测试
指派好槽位后，集群的搭建工作就算完成了。
现在可以正常写入数据了:
以集群的模式去连接：redis-cli -c
Jedis集群写入数据
Set<HostAndPort> hostAndPorts = new HashSet<>(3);
hostAndPorts.add(new HostAndPort("192.168.1.120", 8001));
hostAndPorts.add(new HostAndPort("192.168.1.120", 8004));
hostAndPorts.add(new HostAndPort("192.168.1.120", 8007));
JedisCluster jedisCluster = new JedisCluster(hostAndPorts);
for (int i = 0; i < 100; i++) {
jedisCluster.set(String.valueOf(i), String.valueOf(i));
}
jedisCluster.close();
100个数据会被分散到3台Master里分片存储。
redis-cli快速搭建
上面记录的是自己原生搭建Redis Cluster集群的方式，稍微有点麻烦，Redis提供了一种更简单的方式来快速搭建。
redis-cli --cluster help命令会显示一些搭建集群的帮助信息。
快速搭建集群
redis-cli --cluster create host1:port1 ... hostN:portN --cluster-replicas [主从节点的比例]
redis-cli --cluster create 192.168.1.120:8001 192.168.1.120:8002 192.168.1.120:8003 192.168.1.120:8004 192.168.1.120:8005 192.168.1.120:8006 192.168.1.120:8007 192.168.1.120:8008 192.168.1.120:8009 --cluster-replicas 2
查看节点信息
redis-cli工具一条命令即可快速搭建集群，自动分配主从节点，分配槽位。
集群扩容
如果集群的性能达不到预期，可以进行扩容。
Redis Cluster支持线性扩容，官方宣称可以扩容至1000个节点。
扩容命令：redis-cli --cluster add-node new_host:new_port 存在的IP:存在的端口。
节点默认作为Master被添加，如果需要指明为Slave，加参数：--cluster-slave --cluster-master-id Master的node-id。
For example, 7001,7002,7003 added to the cluster.
# 7001Master
redis-cli --cluster add-node 192.168.1.120:7001 192.168.1.120:8001
# 7002作为7001的Slave
redis-cli --cluster add-node 192.168.1.120:7002 192.168.1.120:7001 --cluster-slave --cluster-master-id f8bbd9f97db60f4b00a80a2bdcf555e95559582a
# 7003作为7001的Slave
redis-cli --cluster add-node 192.168.1.120:7003 192.168.1.120:7001 --cluster-slave --cluster-master-id f8bbd9f97db60f4b00a80a2bdcf555e95559582a
After you add a node, you can cluster nodessee the command node information.
Slots and data migration
Master node expansion slot is not no default, that no data is written, the need for slot allocation and migration data.
槽位和Key是绑定的，槽位重新分配意味着数据需要重新迁移。
Command:redis-cli --cluster reshard ip port .
For example: 4096 slots allocated to new entrants in 7001.
redis-cli --cluster reshard 192.168.1.120:8001
How many slots do you want to move (from 1 to 16384)? 4096 #输入分配的槽位数量
What is the receiving node ID? #输入要分配给哪台节点的node-id
Source node #1: all #要从哪些节点取出槽位分配给新节点？all表示从所有节点中平均取
After the assignment finished viewing slot allocation information:
Cluster volume reduction
If the performance of the cluster excess, can also choose to remove some node, Redis are also supported.
Slot and migrate data
注意：移除节点前，需要将该节点的槽位和数据重新分配到其他Master节点！
Failure to do so can result in data loss, as well as some data is written to fail.
7001 and assigned data slots to 8001:
-cluster-from station node from which
to -cluster-to dispensing station node to which
the number of slots -cluster-slots migration
# 移除7001的槽位和数据
redis-cli --cluster reshard 192.168.1.120:8001 --cluster-from f8bbd9f97db60f4b00a80a2bdcf555e95559582a --cluster-to 201ce4d928c795384bbfde39215065d2eb41592e --cluster-slots 4096
After slots and data migration is complete, you can remove the node.
Command: redis-cli --cluster del-node host:port node_id
For example: removing just added 7001 nodes.
# 集群移除节点
redis-cli --cluster del-node 192.168.1.120:8001 f8bbd9f97db60f4b00a80a2bdcf555e95559582a
After the node is removed, Redis will automatically SHUTDOWN.
