redis哨兵模式

我会带着你远行 2022-12-21 10:52 133阅读 0赞

备注:测试版本 redis 4.0.9

### 文章目录 ###

*  一.redis哨兵模式介绍
 *  二.redis哨兵模式搭建
 *   *  2.1 修改配置文件并启动哨兵模式
     *  2.2 模拟故障
 *  三.Sentinel 总结
 *   *  3.1 Sentinel的作用：
     *  3.2 Sentinel的工作方式:
     *  3.3 Sentinel参数介绍
 *  参考

# 一.redis哨兵模式介绍 #

Sentinel(哨兵)是用于监控redis集群中Master状态的工具，是Redis 的高可用性解决方案，sentinel哨兵模式已经被集成在redis2.4之后的版本中。sentinel是redis高可用的解决方案，sentinel系统可以监视一个或者多个redis master服务，以及这些master服务的所有从服务；当某个master服务下线时，自动将该master下的某个从服务升级为master服务替代已下线的master服务继续处理请求。

sentinel可以让redis实现主从复制，当一个集群中的master失效之后，sentinel可以选举出一个新的master用于自动接替master的工作，集群中的其他redis服务器自动指向新的master同步数据。一般建议sentinel采取奇数台，防止某一台sentinel无法连接到master导致误切换。

# 二.redis哨兵模式搭建 #

环境要求:

<table> 
 <thead> 
  <tr> 
   <th>服务器类别</th> 
   <th>IP</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td>主</td> 
   <td>10.31.1.123</td> 
  </tr> 
  <tr> 
   <td>从</td> 
   <td>10.31.1.124</td> 
  </tr> 
 </tbody> 
</table>

123、124两台服务器已经成功搭建了主从

## 2.1 修改配置文件并启动哨兵模式 ##

123主库执行如下操作:

-- 拷贝哨兵模式的配置文件
    [root@10-31-1-123 redis]# find / -name sentinel.conf                    
    /usr/local/src/redis-4.0.9/sentinel.conf
    [root@10-31-1-123 redis]# 
    [root@10-31-1-123 redis]# cp /usr/local/src/redis-4.0.9/sentinel.conf /usr/local/redis/
    [root@10-31-1-123 redis]# 
    
    -- 修改哨兵模式的参数文件
    vi  sentinel.conf
    bind 10.31.1.123 127.0.0.1
    sentinel monitor mymaster 10.31.1.123 6379 1
    sentinel auth-pass mymaster 123456
    logfile "/tmp/26379.log"
    -- 启动哨兵
     redis-sentinel sentinel.conf &
    -- 查看日志
    [root@10-31-1-123 tmp]# tail -100f 26379.log 
    22387:X 03 Nov 18:17:19.068 # oO0OoO0OoO0Oo Redis is starting oO0OoO0OoO0Oo
    22387:X 03 Nov 18:17:19.068 # Redis version=4.0.9, bits=64, commit=00000000, modified=0, pid=22387, just started
    22387:X 03 Nov 18:17:19.068 # Configuration loaded
    22387:X 03 Nov 18:17:19.069 * Increased maximum number of open files to 10032 (it was originally set to 1024).
    22387:X 03 Nov 18:17:19.072 * Running mode=sentinel, port=26379.
    22387:X 03 Nov 18:17:19.072 # WARNING: The TCP backlog setting of 511 cannot be enforced because /proc/sys/net/core/somaxconn is set to the lower value of 128.
    22387:X 03 Nov 18:17:19.073 # Sentinel ID is b07e0bdbc836f6cb29f943bb8cf8d983d68b6e7a
    22387:X 03 Nov 18:17:19.073 # +monitor master mymaster 10.31.1.123 6379 quorum 2
    22387:X 03 Nov 18:17:19.074 * +slave slave 10.31.1.124:6379 10.31.1.124 6379 @ mymaster 10.31.1.123 6379

主库123 redis.conf文件要添加密码认证，不然切换的时候没有密码认证，连接不上新的主库

masterauth 123456

## 2.2 模拟故障 ##

在125上远程访问主库123上的redis

关闭主库123的redis服务

[root@10-31-1-123 redis]# redis-cli -a 123456 shutdown
    [root@10-31-1-123 redis]#

此时125上显示not connect

[root@10-31-1-125 ~]# redis-cli -h 10.31.1.123 -p 6379 -a 123456 
    10.31.1.123:6379> 
    10.31.1.123:6379> keys *
    1) "my_db3"
    2) "my_db1"
    3) "my_db4"
    4) "my_db2"
    5) "user_id"
    6) "my_db5"
    10.31.1.123:6379> keys *
    Could not connect to Redis at 10.31.1.123:6379: Connection refused
    not connected>

重新启动123上的redis，查看123 redis服务日志  
发现此时123上的主库已经变为从库了，从新主库124上同步数据

3020:S 04 Nov 09:48:48.871 * Connecting to MASTER 10.31.1.124:6379
    3020:S 04 Nov 09:48:48.871 * MASTER <-> SLAVE sync started
    3020:S 04 Nov 09:48:48.871 * Non blocking connect for SYNC fired the event.
    3020:S 04 Nov 09:48:48.872 * Master replied to PING, replication can continue...
    3020:S 04 Nov 09:48:48.872 * Partial resynchronization not possible (no cached master)
    3020:S 04 Nov 09:48:48.875 * Full resync from master: e82e16eb8645f4dced3d99e64c15da8710297cb5:3623877
    3020:S 04 Nov 09:48:48.926 * MASTER <-> SLAVE sync: receiving 283 bytes from master
    3020:S 04 Nov 09:48:48.927 * MASTER <-> SLAVE sync: Flushing old data
    3020:S 04 Nov 09:48:48.927 * MASTER <-> SLAVE sync: Loading DB in memory
    3020:S 04 Nov 09:48:48.927 * MASTER <-> SLAVE sync: Finished with success
    3020:S 04 Nov 09:48:48.928 * Background append only file rewriting started by pid 3024
    3020:S 04 Nov 09:48:48.953 * AOF rewrite child asks to stop sending diffs.
    3024:C 04 Nov 09:48:48.953 * Parent agreed to stop sending diffs. Finalizing AOF...
    3024:C 04 Nov 09:48:48.954 * Concatenating 0.00 MB of AOF diff received from parent.
    3024:C 04 Nov 09:48:48.954 * SYNC append only file rewrite performed
    3024:C 04 Nov 09:48:48.954 * AOF rewrite: 6 MB of memory used by copy-on-write
    3020:S 04 Nov 09:48:48.971 * Background AOF rewrite terminated with success
    3020:S 04 Nov 09:48:48.972 * Residual parent diff successfully flushed to the rewritten AOF (0.00 MB)
    3020:S 04 Nov 09:48:48.972 * Background AOF rewrite finished successfully

查看123的哨兵日志  
发现启动了failover

1262:X 04 Nov 09:16:33.137 # +sdown master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.137 # +odown master mymaster 10.31.1.123 6379 #quorum 1/1
    1262:X 04 Nov 09:16:33.137 # +new-epoch 1
    1262:X 04 Nov 09:16:33.137 # +try-failover master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.138 # +vote-for-leader b07e0bdbc836f6cb29f943bb8cf8d983d68b6e7a 1
    1262:X 04 Nov 09:16:33.139 # +elected-leader master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.139 # +failover-state-select-slave master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.215 # +selected-slave slave 10.31.1.124:6379 10.31.1.124 6379 @ mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.215 * +failover-state-send-slaveof-noone slave 10.31.1.124:6379 10.31.1.124 6379 @ mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.316 * +failover-state-wait-promotion slave 10.31.1.124:6379 10.31.1.124 6379 @ mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.742 # +promoted-slave slave 10.31.1.124:6379 10.31.1.124 6379 @ mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.742 # +failover-state-reconf-slaves master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.806 # +failover-end master mymaster 10.31.1.123 6379
    1262:X 04 Nov 09:16:33.807 # +switch-master mymaster 10.31.1.123 6379 10.31.1.124 6379
    1262:X 04 Nov 09:16:33.807 * +slave slave 10.31.1.123:6379 10.31.1.123 6379 @ mymaster 10.31.1.124 6379
    1262:X 04 Nov 09:16:46.013 * +convert-to-slave slave 10.31.1.123:6379 10.31.1.123 6379 @ mymaster 10.31.1.124 6379

查看 124 redis服务日志  
刚开始的时候是连接不到主库，后面提升为主库

12344:S 04 Nov 09:16:28.562 * Connecting to MASTER 10.31.1.123:6379
    12344:S 04 Nov 09:16:28.563 * MASTER <-> SLAVE sync started
    12344:S 04 Nov 09:16:28.563 # Error condition on socket for SYNC: Connection refused
    12344:S 04 Nov 09:16:29.565 * Connecting to MASTER 10.31.1.123:6379
    12344:S 04 Nov 09:16:29.566 * MASTER <-> SLAVE sync started
    12344:S 04 Nov 09:16:29.566 # Error condition on socket for SYNC: Connection refused
    12344:S 04 Nov 09:16:30.569 * Connecting to MASTER 10.31.1.123:6379
    12344:S 04 Nov 09:16:30.569 * MASTER <-> SLAVE sync started
    12344:S 04 Nov 09:16:30.570 # Error condition on socket for SYNC: Connection refused
    12344:S 04 Nov 09:16:31.572 * Connecting to MASTER 10.31.1.123:6379
    12344:S 04 Nov 09:16:31.572 * MASTER <-> SLAVE sync started
    12344:S 04 Nov 09:16:31.572 # Error condition on socket for SYNC: Connection refused
    12344:S 04 Nov 09:16:32.573 * Connecting to MASTER 10.31.1.123:6379
    12344:S 04 Nov 09:16:32.573 * MASTER <-> SLAVE sync started
    12344:S 04 Nov 09:16:32.574 # Error condition on socket for SYNC: Connection refused
    12344:M 04 Nov 09:16:33.316 # Setting secondary replication ID to 21f6348d7db2446e2369e1b6243a7d4fc5fe3a42, valid up to offset: 3494901. New replication ID is e82e16eb8645f4dced3d99e64c15da8710297cb5
    12344:M 04 Nov 09:16:33.316 * Discarding previously cached master state.
    12344:M 04 Nov 09:16:33.316 * MASTER MODE enabled (user request from 'id=52920 addr=10.31.1.123:55852 fd=10 name=sentinel-b07e0bdb-cmd age=205 idle=0 flags=x db=0 sub=0 psub=0 multi=3 qbuf=0 qbuf-free=32768 obl=36 oll=0 omem=0 events=r cmd=exec')
    12344:M 04 Nov 09:16:33.319 # CONFIG REWRITE executed with success.
    12344:M 04 Nov 09:19:25.098 * 1 changes in 900 seconds. Saving...
    12344:M 04 Nov 09:19:25.099 * Background saving started by pid 21765
    21765:C 04 Nov 09:19:25.100 * DB saved on disk
    21765:C 04 Nov 09:19:25.103 * RDB: 6 MB of memory used by copy-on-write
    12344:M 04 Nov 09:19:25.200 * Background saving terminated with success

查看123主库的redis.conf  
可以看到已经成为新的主库124的备库，从124上同步数据

# Generated by CONFIG REWRITE
    slaveof 10.31.1.124 6379

查看123主库的sentinel.conf  
可以看到，主从库已经互换了

sentinel monitor mymaster 10.31.1.124 6379 1
    # Generated by CONFIG REWRITE
    sentinel known-slave mymaster 10.31.1.123 6379
    sentinel current-epoch 1

# 三.Sentinel 总结 #

## 3.1 Sentinel的作用： ##

A、Master 状态监测  
B、如果Master 异常，则会进行Master-slave 转换，将其中一个Slave作为Master，将之前的Master作为Slave  
C、Master-Slave切换后，master\_redis.conf、slave\_redis.conf和sentinel.conf的内容都会发生改变，即master\_redis.conf中会多一行slaveof的配置，sentinel.conf的监控目标会随之调换

## 3.2 Sentinel的工作方式: ##

1)：每个Sentinel以每秒钟一次的频率向它所知的Master，Slave以及其他Sentinel实例发送一个PING命令  
2)：如果一个实例（instance）距离最后一次有效回复 PING 命令的时间超过 down-after-milliseconds 选项所指定的值，则这个实例会被 Sentinel 标记为主观下线。  
3)：如果一个Master被标记为主观下线，则正在监视这个Master的所有 Sentinel 要以每秒一次的频率确认Master的确进入了主观下线状态。  
4)：当有足够数量的 Sentinel（大于等于配置文件指定的值）在指定的时间范围内确认Master的确进入了主观下线状态， 则Master会被标记为客观下线  
5)：在一般情况下， 每个 Sentinel 会以每10 秒一次的频率向它已知的所有Master，Slave发送 INFO 命令  
6)：当Master被 Sentinel 标记为客观下线时，Sentinel 向下线的 Master 的所有 Slave 发送 INFO 命令的频率会从 10 秒一次改为每秒一次  
7)：若没有足够数量的 Sentinel 同意Master 已经下线， Master 的客观下线状态就会被移除。  
若 Master 重新向 Sentinel 的 PING 命令返回有效回复， Master 的主观下线状态就会被移除。

## 3.3 Sentinel参数介绍 ##

# 禁止保护模式
    protected-mode no
    # 配置监听的主服务器，这里sentinel monitor代表监控，mymaster代表服务器的名称，可以自定义，192.168.11.128代表监控的主服务器，6379代表端口，2代表只有两个或两个以上的哨兵认为主服务器不可用的时候，才会进行failover操作。
    sentinel monitor mymaster 192.168.11.128 6379 2
    # sentinel author-pass定义服务的密码，mymaster是服务名称，123456是Redis服务器密码
    # sentinel auth-pass <master-name> <password>
    sentinel auth-pass mymaster 123456

# 参考 #

1.https://www.cnblogs.com/kevingrace/p/9004460.html  
2.https://blog.csdn.net/yaooch/article/details/80167571