redis一主二从加哨兵

redis版本：redis-3.0.6.tar.gz

master:192.168.3.180

slave:192.168.3.184 (机器原因，两从都在这上面)

一.redis安装

cd /root/tools/
tar -zxvf redis-3.0..tar.gz
cd redis-3.0.
make install PREFIX=/usr/local/redis
ln -s /usr/local/redis/bin/redis-cli /usr/local/bin/redis-cli
cp utils/redis_init_script /etc/init.d/redis
mkdir /etc/redis
cp redis.conf /etc/redis/.conf
cp -rp sentinel.conf /etc/redis/sentinel_26379.conf

二.配置

主服务器上redis

vim .conf
daemonize yes
pidfile /var/run/redis_6379.pid
port
bind 0.0.0.0  （很重要，否则主从同步会失败）
logfile "/var/log/redis_6379.log"
dbfilename "dump_6379.rdb"
dir "/opt/redis/6379"
requirepass  （设置redis密码） 

哨兵sentinel (本文只有一个哨兵，哨兵也可以部署多台机器，形成哨兵集群，避免单点的问题，保证系统的高可用。)

#修改或添加以下几项
vim sentinel_26379.conf
port
daemonize yes
logfile "/tmp/sentinel.log"
dir "/opt/redis/redis_sentinel"
sentinel monitor mymaster 192.168.3.180
(最后的数字1指明当有多少个sentinel认为一个master失效时，master才算真正失效)
sentinel parallel-syncs mymaster
(最后的数字2表示有多少个slave)
sentinel auth-pass mymaster password-xxx  （如果redis配有密码，为了使哨兵能登录监控，需要给哨兵也配置密码认证）

从服务器上

#从服务器配置可复制主服务器的，只需要修改相应端口和修改添加以下两行即可
slaveof 192.168.3.180   （指定主redis）
masterauth  "" (指定master密码)

测试：

分别启动master和两个slave

/usr/local/redis/bin/redis-server /etc/redis/.conf

master上：

[root@novel tmp]# redis-cli -p  -a password
127.0.0.1:> info replication
# Replication
role:master
connected_slaves:
slave0:ip=192.168.3.184,port=,state=online,offset=,lag=
slave1:ip=192.168.3.184,port=,state=online,offset=,lag=
master_repl_offset:
repl_backlog_active:
repl_backlog_size:
repl_backlog_first_byte_offset:
repl_backlog_histlen:

slave上：

[root@danny redis]# redis-cli -p  -a password
127.0.0.1:> info replication
# Replication
role:slave
master_host:192.168.3.180
master_port:
master_link_status:up

[root@danny redis]# redis-cli -p  -a password
127.0.0.1:> info replication
# Replication
role:slave
master_host:192.168.3.180
master_port:
master_link_status:up

以上可看到已经完成主从同步，也可以去看看slave的启动日志，查看同步实时信息

[root@danny redis]# tailf /var/log/redis_6390.log
:S  Jan ::29.985 * Connecting to MASTER 192.168.3.180:
:S  Jan ::29.985 * MASTER <-> SLAVE sync started
:S  Jan ::29.986 * Non blocking connect for SYNC fired the event.
:S  Jan ::29.987 * Master replied to PING, replication can continue...
:S  Jan ::29.987 * Partial resynchronization not possible (no cached master)
:S  Jan ::29.989 * Full resync from master: 4da2c58d50928717d9a45216ced5c36a45a3b78c:
:S  Jan ::30.055 * MASTER <-> SLAVE sync: receiving  bytes from master
:S  Jan ::30.055 * MASTER <-> SLAVE sync: Flushing old data
:S  Jan ::30.055 * MASTER <-> SLAVE sync: Loading DB in memory
:S  Jan ::30.055 * MASTER <-> SLAVE sync: Finished with success

三.哨兵启用

进入redis的安装目录可以看到

[root@danny bin]# ls
dump.rdb  redis-benchmark  redis-check-aof  redis-check-dump  redis-cli  redis-sentinel  redis-server
[root@danny bin]# pwd
/usr/local/redis/bin

启动哨兵

./redis-sentinel /etc/redis/sentinel_26379.conf

四.哨兵测试

打开slave机上哨兵日志

关闭master机上redis

日志如下：

:X  Jan ::21.086 # +sdown master mymaster 192.168.3.180
:X  Jan ::21.086 # +odown master mymaster 192.168.3.180  #quorum /
:X  Jan ::21.086 # +new-epoch
:X  Jan ::21.086 # +try-failover master mymaster 192.168.3.180
:X  Jan ::21.132 # +vote-for-leader 911877d1d33938dbdfdd0224ea61a3245df12617
:X  Jan ::21.132 # +elected-leader master mymaster 192.168.3.180
:X  Jan ::21.133 # +failover-state-select-slave master mymaster 192.168.3.180
:X  Jan ::21.224 # +selected-slave slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::21.224 * +failover-state-send-slaveof-noone slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::21.307 * +failover-state-wait-promotion slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::22.189 # +promoted-slave slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::22.189 # +failover-state-reconf-slaves master mymaster 192.168.3.180
:X  Jan ::22.237 * +slave-reconf-sent slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::23.194 * +slave-reconf-inprog slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.180
:X  Jan ::23.194 * +slave-reconf-done slave 192.168.3.184:6391 192.168.3.184 6391 @ mymaster 192.168.3.180
:X  Jan ::23.270 # +failover-end master mymaster 192.168.3.180
:X  Jan ::23.270 # +switch-master mymaster 192.168.3.180 6390 192.168.3.184 6390
:X  Jan ::23.271 * +slave slave 192.168.3.184: 192.168.3.184  @ mymaster 192.168.3.184
:X  Jan ::23.271 * +slave slave 192.168.3.180: 192.168.3.180  @ mymaster 192.168.3.184
:X  Jan ::53.292 # +sdown slave 192.168.3.180: 192.168.3.180  @ mymaster 192.168.3.184
:X  Jan ::40.874 # -sdown slave 192.168.3.180: 192.168.3.180  @ mymaster 192.168.3.184 

可以看到，新的master已经被选举出来了。

备注：

有时候redis的master宕了之后哨兵并没有实现故障转移，哨兵日志打印为：

failover-abort-not-elected master mymaster 192.168.1.88 6379

可能原因之一就是哨兵集群的下列两个初始值没有一致，保证每个哨兵的初始值一样即可。

sentinel config-epoch mymaster
sentinel leader-epoch mymaster