Redis集群生产环境高可用方案实战过程
布署方案说明
1、sentinel负责对redis集群中的主从服务监控、提醒和自动故障转移
2、redis集群负责对外提供相关服务
Sentinel原理介绍
原理：
sentinel是一个分布式系统，可以在一个架构中运行多个sentinel进程，这些进程使用流言协议（gossip protocols）来接收关于rdis主服务器是否下线的信息，并使用投票协议（agreement protocols）来决定是否执行自动故障迁移，以及选举哪个从服务器成为新的主服务器。
流言协议：
sentinel服务通过ping命令来确认监控的服务器是否正常，当足够多数量的sentinel都确认监控的同一服务器停止服务了（主观下线），则判定此服务器停止服务。
投票协议：
其实就选举，sentinel集群根据一定的规则从redis群中选择一个新的服务器成为主服务器，并使其它的服务器做为新的从服务器，并修改自身的配置文件。
服务器布署规划
实验环境采用两台服务器模拟集群环境
服务器系统环境
Centos 6.6 x86_64
Master服务器 10.0.0.3/24
Redis-Mster 10.0.0.3:6379
Redis-Slave1 10.0.0.3:63791
Redis-Slave2 10.0.0.3:63792
Sentinel服务
s 10.0.0.3:26379
s1 10.0.0.3:26378
Slave服务器 10.0.0.4/24
Redis-Slave3 10.0.0.4:63793
Redis-Slave4 10.0.0.4:63794
Sentinel服务
s2 10.0.0.4:26379
s3 10.0.0.4:26378
故障切换前后逻辑图
Redis-sentinel服务配置
主服务器上安装布署过程
安装redis服务
mkdir /usr/local/redis/data
cd /usr/local/src
wget http://download.redis.io/releases/redis-2.8.9.tar.gz
tar zxf redis-2.8.9.tar.gz
cd redis-2.8.9
make && make install
复制配置文件
cp redis.conf /usr/local/bin/
cd /usr/local/bin
cp redis.conf redis-slave1
cp redis.conf redis-slave2
修改配置文件
[root@master bin]#vi redis.conf
daemonize yes
#开启后台运行模式
pidfile /var/run/redis.pid
bind 10.0.0.3
dbfilename dump.rdb
dir /usr/local/redis/data
port 6379
[root@master bin]#vi redis-slave1
daemonize yes
pidfile /var/run/redis-slave1.pid
port 63791
bind 10.0.0.3
dbfilename dump-slave1.rdb
dir /usr/local/redis/data
slaveof 10.0.0.3 6379
slave-read-only yes
[root@master bin]#vi redis-slave2
daemonize yes
pidfile /var/run/redis-slave2.pid
port 63792
bind 10.0.0.3
dbfilename dump-slave2.rdb
dir /usr/local/redis/data
slaveof 10.0.0.3 6379
配置redis-sentinel服务
mkdir /var/log/redis -p
cp /usr/local/src/redis-2.8.9/src/redis-sentinel /usr/bin/
cp /usr/local/src/redis-2.8.9/src/sentinel.conf /usr/local/bin/
cd /usr/local/bin
cp sentinel.conf sentinel-s1.conf
修改配置文件
[root@master bin]# egrep -v "^#|^$" sentinel.conf
port 26379
daemonize yes
logfile /var/log/redis/sentinel.log
sentinel monitor mymaster 10.0.0.3 6379 2
sentinel down-after-milliseconds mymaster 30000
sentinel parallel-syncs mymaster 1
sentinel failover-timeout mymaster 180000
[root@master bin]# egrep -v "^#|^$" sentinel-s1.conf
port 26378
daemonize yes
logfile /var/log/redis/sentinel-s1.log
sentinel monitor mymaster 10.0.0.3 6379 2
sentinel down-after-milliseconds mymaster 30000
sentinel parallel-syncs mymaster 1
sentinel failover-timeout mymaster 180000
以上配置从服务器操作过程同上
启动服务
启动redis服务
[root@master bin]# redis-server redis.conf
[root@master bin]# redis-server redis-slave1
[root@master bin]# redis-server redis-slave2
[root@master bin]# ps -ef|grep redis
root 2579 1 0 23:55 ? 00:00:00 redis-server 10.0.0.3:6379
root 2585 1 0 23:55 ? 00:00:00 redis-server 10.0.0.3:63792
root 2590 1 0 23:55 ? 00:00:00 redis-server 10.0.0.3:63791
root 2597 2479 0 23:56 pts/0 00:00:00 grep --color=auto redis
[root@slave bin]# redis-server redis-slave3
[root@slave bin]# redis-server redis-slave4
[root@slave bin]# ps -ef|grep redis
root 2576 1 0 23:56 ? 00:00:00 redis-server 10.0.0.4:63793
root 2580 1 0 23:56 ? 00:00:00 redis-server 10.0.0.4:63794
root 2584 2502 0 23:56 00:00:00 grep --color=auto redis
启动redis-sentinel服务
[root@master bin]# redis-sentinel sentinel.conf
[root@master bin]# redis-sentinel sentinel-s1.conf
[root@master bin]# ps -ef|grep redis-sentinel
root 2638 1 0 01:05 ? 00:00:04 redis-sentinel *:26379
root 2646 1 0 01:13 ? 00:00:00 redis-sentinel *:26378
root 2650 2479 0 01:13 00:00:00 grep --color=auto redis
[root@slave bin]# redis-sentinel sentinel-s2.conf
[root@slave bin]# redis-sentinel sentinel-s3.conf
[root@slave bin]# ps -ef|grep redis-sentinel
root 2644 1 1 01:14 ? 00:00:00 redis-sentinel *:26378
root 2649 1 0 01:14 ? 00:00:00 redis-sentinel *:26379
root 2653 2502 0 01:15 00:00:00 grep --color=auto redis-sentinel
查看日志观察启动过程
[root@master bin]# tail -f /var/log/redis/sentinel.log
`-.__.-'
[2664] 12 May 01:20:11.036 # Sentinel runid is c327be464ef36e670566a0d76c9dc85bac7f33b1
[2664] 12 May 01:20:11.036 # +monitor master mymaster 10.0.0.3 6379 quorum 2
[2664] 12 May 01:20:11.123 * -dup-sentinel master mymaster 10.0.0.3 6379 #duplicate of 10.0.0.3:26378 or fb1fbe73b51a0a6e71a8ceae57d34ef773d086e3
[2664] 12 May 01:20:11.123 * +sentinel sentinel 10.0.0.3:26378 10.0.0.3 26378 @ mymaster 10.0.0.3 6379
[2664] 12 May 01:20:21.410 * -dup-sentinel master mymaster 10.0.0.3 6379 #duplicate of 10.0.0.4:26379 or 3d43ddea4d4ba8de7dd30e2d332723508f6d4c19
[2664] 12 May 01:20:21.410 * +sentinel sentinel 10.0.0.4:26379 10.0.0.4 26379 @ mymaster 10.0.0.3 6379
[2664] 12 May 01:20:25.103 * -dup-sentinel master mymaster 10.0.0.3 6379 #duplicate of 10.0.0.4:26378 or 6d134d9a3e53c0cb70de842281de8aaf17a84c00
[2664] 12 May 01:20:25.103 * +sentinel sentinel 10.0.0.4:26378 10.0.0.4 26378 @ mymaster 10.0.0.3 6379
可以看出有其它监控服务器加入到集群中来
查看配置文件是否有变化
root@master bin]# egrep -v "^#|^$" sentinel-s1.conf
port 26378
daemonize yes
logfile "/var/log/redis/sentinel-s1.log"
sentinel monitor mymaster 10.0.0.3 6379 2
sentinel config-epoch mymaster 0
sentinel leader-epoch mymaster 0
sentinel known-slave mymaster 10.0.0.3 63792
dir "/usr/local/bin"
sentinel known-slave mymaster 10.0.0.4 63793
sentinel known-slave mymaster 10.0.0.4 63794
sentinel known-slave mymaster 10.0.0.3 63791
sentinel known-sentinel mymaster 10.0.0.3 26379 c327be464ef36e670566a0d76c9dc85bac7f33b1
sentinel known-sentinel mymaster 10.0.0.4 26379 3d43ddea4d4ba8de7dd30e2d332723508f6d4c19
sentinel known-sentinel mymaster 10.0.0.4 26378 6d134d9a3e53c0cb70de842281de8aaf17a84c00
sentinel current-epoch 0
通过日志观察故障切换过程
模拟主服务器故障并查看故障切换
[root@master bin]# redis-cli -h 10.0.0.3 -p 6379 shutdown
[root@master bin]# ps -ef|grep redis
root 2585 1 0 May11 ? 00:00:07 redis-server 10.0.0.3:63792
root 2590 1 0 May11 ? 00:00:07 redis-server 10.0.0.3:63791
root 2660 1 0 01:20 ? 00:00:02 redis-sentinel *:26378
root 2664 1 0 01:20 ? 00:00:02 redis-sentinel *:26379
root 2676 2479 0 01:30 00:00:00 grep --color=auto redis
此时发现主服务器进程不存在，说明服务有故障
清空原来的日志并查看故障切换过程
[root@slave bin]# > /var/log/redis/sentinel-s3.log
[root@slave bin]# tail -f /var/log/redis/sentinel-s3.log
[2669] 12 May 01:30:55.203 # +sdown master mymaster 10.0.0.3 6379
[2669] 12 May 01:30:55.276 # +new-epoch 1
[2669] 12 May 01:30:55.280 # +vote-for-leader c327be464ef36e670566a0d76c9dc85bac7f33b1 1
[2669] 12 May 01:30:56.329 # +odown master mymaster 10.0.0.3 6379 #quorum 4/2
[2669] 12 May 01:30:57.547 # +switch-master mymaster 10.0.0.3 6379 10.0.0.3 63792
[2669] 12 May 01:30:57.548 * +slave slave 10.0.0.4:63794 10.0.0.4 63794 @ mymaster 10.0.0.3 63792
[2669] 12 May 01:30:57.553 * +slave slave 10.0.0.4:63793 10.0.0.4 63793 @ mymaster 10.0.0.3 63792
[2669] 12 May 01:30:57.556 * +slave slave 10.0.0.3:63791 10.0.0.3 63791 @ mymaster 10.0.0.3 63792
[2669] 12 May 01:30:57.561 * +slave slave 10.0.0.3:6379 10.0.0.3 6379 @ mymaster 10.0.0.3 63792
[2669] 12 May 01:31:27.620 # +sdown slave 10.0.0.3:6379 10.0.0.3 6379 @ mymaster 10.0.0.3 63792
可以看出判定master主观下线（+sdown），sentinel选举10.0.0.3 63792为新的主服务器，其它slave自动执行slaveof ，故障转移成功
恢复原主服务器
[root@master bin]# redis-server redis.conf
[root@master bin]# ps -ef|grep redis
root 2585 1 0 May11 ? 00:00:08 redis-server 10.0.0.3:63792
root 2590 1 0 May11 ? 00:00:08 redis-server 10.0.0.3:63791
root 2660 1 0 01:20 ? 00:00:05 redis-sentinel *:26378
root 2664 1 0 01:20 ? 00:00:05 redis-sentinel *:26379
root 2683 1 0 01:36 ? 00:00:00 redis-server 10.0.0.3:6379
root 2689 2479 0 01:36 00:00:00 grep --color=auto redis
[root@slave bin]# tail -f /var/log/redis/sentinel-s3.log
[2673] 12 May 01:36:21.925 # -sdown slave 10.0.0.3:6379 10.0.0.3 6379 @ mymaster 10.0.0.3 63792
当原来主服务器故障恢复后，自动以从角色加入到集群，并不会抢占主服务器的角色
测试读写分离
[root@master bin]# redis-cli -h 10.0.0.3 -p 63792
10.0.0.3:63792> get key
"test"
10.0.0.3:63792> set key file
OK
10.0.0.3:63792> get key
"file"
[root@master bin]# redis-cli -h 10.0.0.3 -p 6379
10.0.0.3:6379> get key
"file"
10.0.0.3:6379> set key file1
(error) READONLY You can't write against a read only slave.
说明新主是提升成功的，原来的主故障恢复后已是从服务器，而且也是只读状态，没有破坏之前的主写从读的状态
至此整个布署过程结束，实现了集群监控与自动故障切换、读写分离的功能