首先我们来尝试下搭建一个Redis Cluster集群
前置准备
Redis Cluster需要Redis 3.0及以上版本才支持,此文发布的时候,Redis的最高版本为3.0.5。
wget http://download.redis.io/releases/redis-3.0.5.tar.gz
tar -xvf redis-3.0.5.tar.gz
cd redis-3.0.5
make
编译完Redis,生成的可执行文件在redis-3.0.5/src
之中,为了方便使用,我们把可执行文件的目录加入PATH之中。
vim ~/.bashrc
# 增加以下内容
REDIS_HOME=/home/xiezefan/sofeware/redis-3.0.5/src
PATH=$REDIS_HOME:$PATH
export PATH
# 保存后让修改生效
source ~/.bashrc
要创建Redis Cluster,我们还需要安装Ruby以及RubyGems。
yum install ruby
yum install gcc g++ make automake autoconf curl-devel openssl-devel zlib-devel httpd-devel apr-devel apr-util-devel sqlite-devel
yum install ruby-rdoc ruby-devel
yum install rubygems
gem install redis
创建集群
此次此时我们需要创建8个节点,端口号7000~7007
复制redis-3.0.5/redis.conf
,修改一下内容
mkdir cluster
cp ~/sofeware/redis-3.0.5/redis.conf cluster/
vim cluster/redis.conf
# 修改以下内容
port 7000
cluster-enabled yes
cluster-config-file nodes.conf
cluster-node-timeout 5000
appendonly yes
批量复制7份,修改配置文件的端口号。修改完毕后,分别进入各个节点的目录中启动redis
cd 7000
redis-server redis.conf
cd 7001
redis-server redis.conf
# 依次启动7000-7007
...
至此,8个Redis全都启动完毕,但是他们还处于彼此互相不知道彼此的阶段。
xiezefan@ubuntu:~$ ps -ef | grep redis
xiezefan 13372 1 0 20:09 ? 00:00:08 redis-server *:7000 [cluster]
xiezefan 13376 1 0 20:09 ? 00:00:08 redis-server *:7001 [cluster]
xiezefan 13380 1 0 20:09 ? 00:00:08 redis-server *:7002 [cluster]
xiezefan 13382 1 0 20:09 ? 00:00:08 redis-server *:7003 [cluster]
xiezefan 13386 1 0 20:09 ? 00:00:08 redis-server *:7004 [cluster]
xiezefan 13390 1 0 20:09 ? 00:00:08 redis-server *:7005 [cluster]
xiezefan 13394 1 0 20:09 ? 00:00:08 redis-server *:7006 [cluster]
xiezefan 13400 1 0 20:09 ? 00:00:08 redis-server *:7007 [cluster]
下一步,我们要将7000-7005这六个节点连接成一个集群。
redis-trib.rb create --replicas 1 10.211.55.4:7000 10.211.55.4:7001 10.211.55.4:7002 10.211.55.4:7003 10.211.55.4:7004 10.211.55.4:7005
该命令表示,将7000-7006节点创建一个集群,冗余为1,就是3主3从。切记,此处指定的IP会在client发送move命令的时候返回,所以一定要指定为客户端能访问到的IP,例如下面这种IP是不可行的,client拿到的IP就位是127.0.0.1导致一直重定向失败。
redis-trib.rb create --replicas 1 127.0.0.1:7000 127.0.0.1:7001 127.0.0.1:7002 127.0.0.1:7003 127.0.0.1:7004 127.0.0.1:7005
输入后,redis-trib自动分配给出一个slot的分配方案
>>> Performing hash slots allocation on 6 nodes...
Using 3 masters:
10.211.55.4:7000
10.211.55.4:7001
10.211.55.4:7002
Adding replica 10.211.55.4:7003 to 10.211.55.4:7000
Adding replica 10.211.55.4:7004 to 10.211.55.4:7001
Adding replica 10.211.55.4:7005 to 10.211.55.4:7002
M: 9dfef549f7917794cbabaf96781ed0e19957c1f3 10.211.55.4:7000
slots:0-5460 (5461 slots) master
M: c14485e8d7f1f3ec5c505b41fe727b657c951d8d 10.211.55.4:7001
slots:5461-10922 (5462 slots) master
M: 8b308093e99f4299b8c18ab1dd81c5a83a3528c6 10.211.55.4:7002
slots:10923-16383 (5461 slots) master
S: 632409883570eb5cecf6089583fba64a41d1154f 10.211.55.4:7003
replicates 9dfef549f7917794cbabaf96781ed0e19957c1f3
S: edd31196e2980360b0738c57af95e1a69b0f9c9b 10.211.55.4:7004
replicates c14485e8d7f1f3ec5c505b41fe727b657c951d8d
S: 95063a99c5cf2cc4abc51ca1e8ff0f3b8d60271c 10.211.55.4:7005
replicates 8b308093e99f4299b8c18ab1dd81c5a83a3528c6
Can I set the above configuration? (type 'yes' to accept):
输入yes后,集群自动创建,创建完毕后,通过redis-cli进入任一节点,使用cluster nodes
命令可以查看各个节点的状态
xiezefan@ubuntu:~$ redis-cli -p 7000
127.0.0.1:7000>cluster nodes
c14485e8d7f1f3ec5c505b41fe727b657c951d8d 10.211.55.4:7001 master - 0 1449060968280 2 connected 5461-10922
95063a99c5cf2cc4abc51ca1e8ff0f3b8d60271c 10.211.55.4:7005 slave 8b308093e99f4299b8c18ab1dd81c5a83a3528c6 0 1449060968280 6 connected
edd31196e2980360b0738c57af95e1a69b0f9c9b 10.211.55.4:7004 slave c14485e8d7f1f3ec5c505b41fe727b657c951d8d 0 1449060967274 5 connected
8b308093e99f4299b8c18ab1dd81c5a83a3528c6 10.211.55.4:7002 master - 0 1449060966769 3 connected 10923-16383
632409883570eb5cecf6089583fba64a41d1154f 10.211.55.4:7003 slave 9dfef549f7917794cbabaf96781ed0e19957c1f3 0 1449060967274 4 connected
9dfef549f7917794cbabaf96781ed0e19957c1f3 10.211.55.4:7000 myself,master - 0 0 1 connected 0-5460
随便输入一个查询指令get user.1
127.0.0.1:7000> get user.1
(error) MOVED 9645 10.211.55.4:7001
因为user.1所在的solt-9645在7001节点上,cluster给你发送MOVED指令让你去7001节点查找数据,连接redis-cli的时候,使用-c
参数可以指定查询时接收到MOVED指令自动跳转
xiezefan@ubuntu:~$ redis-cli -c -p 7000
127.0.0.1:7000> get user.1
-> Redirected to slot [9645] located at 10.211.55.4:7001
(nil)
集群扩容
现在我们已经有一个包含6个节点的集群,我写了段代码,往集群写入10W条测试数据。
现在模拟机器扩容场景,为集群加入一个master节点7006和一个slave节点7007。
redis-trib.rb add-node 10.211.55.4:7006 10.211.55.4:7000
以上命令将7006节点接入7000所在的集群。接下来,我们为7006增加一个slave节点。
redis-trib.rb add-node --slave 10.211.55.4:7007 10.211.55.4:7000
以上命令表示增加slave节点,将7006的节点加入7000节点所在的集群中作为slave节点,随机依附现有的master节点中slave最少的节点,如果需要再指定特别的master节点,使用
redis-trib.rb add-node --slave --master-id 23b412673af0506df6382353e3a65960d5b7e66d 10.211.55.4:7007 10.211.55.4:7000
其中的23b412673af0506df6382353e3a65960d5b7e66d
是7006节点的id,我们可以通过cluster nodes
命令查看节点的id。
接下来我们用坐负载均衡,Slot是Redis Cluster数据承载的最小单位,我们可以指定将一定范围的Slot转移到新的节点来实现负载均衡。
Redis Cluster转移一个Slot的步骤是:
- 在目标节点上声明将从源节点上迁入Slot
CLUSTER SETSLOT <slot> IMPORTING <source_node_id>
- 在源节点上声明将往目标节点迁出Slot
CLUSTER SETSLOT <slot> IMPORTING <source_node_id>
- 批量从源节点获取KEY
CLUSTER GETKEYSINSLOT <slot> <count>
- 将获取的Key迁移到目标节点
MIGRATE <target_ip> <target_port> <key_name> 0 <timeout>
- 重复步骤3,4直到所有数据迁移完毕
- 分别向双方节点发送
CLUSTER SETSLOT <slot> NODE <target_node_id>
,该命令将会广播给集群其他节点,已经将Slot转移到目标节点。 - 等待集群状态变为OK
CLUSTER INFO
中的cluster_state = ok
我编写了一个脚本来批量迁移Slot
#!/bin/bash
source_host=$1 # 源节点HOST
source_port=$2 # 源节点端口
target_host=$3 # 目标节点HOST
target_port=$4 # 目标节点端口
start_slot=$5 # 迁移节点的其实范围
end_slot=$6 # 迁移节点的结束范围
for slot in `seq ${start_slot} ${end_slot}`
do
redis-cli -c -h ${target_host} -p ${target_port} cluster setslot ${slot} importing `redis-cli -c -h ${source_host} -p ${source_port} cluster nodes | grep ${source_port} | awk '{print $1}'`
echo "Setslot importing ${slot} to ${target_host}:${target_port} success"
redis-cli -c -h ${source_host} -p ${source_port} cluster setslot ${slot} migrating `redis-cli -c -h ${target_host} -p ${target_port} cluster nodes | grep ${target_port} | awk '{print $1}'`
echo "Setslot migrating ${slot} from ${source_host}:${source_port} success"
while [ 1 -eq 1 ]
do
allkeys=`redis-cli -c -h ${source_host} -p ${source_port} cluster getkeysinslot ${slot} 10`
if [ -z "${allkeys}" ]
then
redis-cli -c -h ${source_host} -p ${source_port} cluster setslot ${slot} node `redis-cli -c -h ${target_host} -p ${target_port} cluster nodes | grep ${target_port} | awk '{print $1}'`
redis-cli -c -h ${target_host} -p ${target_port} cluster setslot ${slot} node `redis-cli -c -h ${source_host} -p ${target_port} cluster nodes | grep ${target_port} | awk '{print $1}'`
echo "Migrate slot ${slot} finish"
break
else
for key in ${allkeys}
do
redis-cli -c -h ${source_host} -p ${source_port} migrate ${target_host} ${target_port} ${key} 0 5000
echo "Migrate slot ${slot} key ${key} success"
done
fi
done
done
执行命令 bash rebalance-cluster.sh 10.211.55.4 7000 10.211.55.4 7006 0 1000
将7000节点上0-1000这个范围内的Slot转移到7006节点,通过cluster nodes
命令我们可以看到0-1000这个区间是slot已经从7000转移到7006
xiezefan@ubuntu:~/sheel$ redis-cli -c -p 7000 cluster nodes
23b412673af0506df6382353e3a65960d5b7e66d 10.211.55.4:7006 master - 0 1449064402389 7 connected 0-1000
0c2954d21d7bcdae333f4fdecf468ce05aa25544 10.211.55.4:7001 master - 0 1449064400372 2 connected 5461-10922
384a3bb5bd9ecb2fc7db75c866abc715d7966f82 10.211.55.4:7002 master - 0 1449064401381 3 connected 10923-16383
c3e09d286ef2dce49843268b20832d65a5d516a1 10.211.55.4:7004 slave 0c2954d21d7bcdae333f4fdecf468ce05aa25544 0 1449064401885 5 connected
50737b4a91443ab1a34eec4ef99d4f6fe5d358f4 10.211.55.4:7005 slave 384a3bb5bd9ecb2fc7db75c866abc715d7966f82 0 1449064402389 6 connected
3c62cc6664bba378cceb8ae8e02f5d727deafe9d 10.211.55.4:7007 slave 23b412673af0506df6382353e3a65960d5b7e66d 0 1449064400878 7 connected
d6441916dcd89cbf431465d92dfc0eb3dd235295 10.211.55.4:7003 slave 6ee21c5d93a6d2f293a2df1b37e8c9c27cb55ad8 0 1449064402389 4 connected
6ee21c5d93a6d2f293a2df1b37e8c9c27cb55ad8 10.211.55.4:7000 myself,master - 0 0 1 connected 1001-5460
参考文章
- 全面剖析Redis Cluster原理和应用
- Redis集群功能预览
- Redis-Cluster实战--8.Redis-Cluster水平扩容(redis-cli实现版)
- 谈谈陌陌争霸在数据库方面踩过的坑( Redis 篇)
本文首发: http://xiezefan.me/2015/12/03/redis_cluster_research_1/