redis sentinel 2节点，停机自动切换

我们知道，redis sentinel可以自动切换master和slave。在实验的时候，只用了两个节点。模拟master机器down机。这样，只剩下了一个sentinel。

但我们知道，sentinel选择leader的时候，要大部分sentinel同意才认为他是leader。当前情形下，就要求2个节点都同意。但是，我们只剩下了一个sentinel，就永远不会选择出leader，也就无法进行failover操作。

为什么有一个sentinel down了，还会有2个节点呢？他是如何判断有几个sentinel的呢？只要改变它是不是就可以了？

照这个思路，看sentinel的配置文件，有下面一行：

sentinel known-sentinel mymaster 10.133.17.203 26380 d4737f3a088135cf04d9a2efae7cda756ae9a053

应该就是靠这个算的，删掉，重启，查看sentinel数量：

src/redis-cli -p 26380 -a qwe123 info Sentinel

果然数量只剩下一个了。

看日志：

这是刚开始down机时，slave这台机器的sentinel状态：

20139:X 13 Dec 02:35:21.035 # +new-epoch 4
20139:X 13 Dec 02:35:21.035 # +try-failover master mymaster 10.133.17.203 6379
20139:X 13 Dec 02:35:21.080 # +vote-for-leader dd533ec477f3e68d22fbb0203700e3993669c76a 4
20139:X 13 Dec 02:35:31.590 # -failover-abort-not-elected master mymaster 10.133.17.203 6379

提示找不到leader。

修改配置重启后：

20172:X 13 Dec 02:37:03.743 # +new-epoch 7
20172:X 13 Dec 02:37:03.743 # +try-failover master mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:03.788 # +vote-for-leader d4737f3a088135cf04d9a2efae7cda756ae9a053 7
20172:X 13 Dec 02:37:03.788 # +elected-leader master mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:03.788 # +failover-state-select-slave master mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:03.854 # +selected-slave slave 10.133.17.203:6380 10.133.17.203 6380 @ mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:03.854 * +failover-state-send-slaveof-noone slave 10.133.17.203:6380 10.133.17.203 6380 @ mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:03.938 * +failover-state-wait-promotion slave 10.133.17.203:6380 10.133.17.203 6380 @ mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:04.247 # +promoted-slave slave 10.133.17.203:6380 10.133.17.203 6380 @ mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:04.247 # +failover-state-reconf-slaves master mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:04.305 # +failover-end master mymaster 10.133.17.203 6379
20172:X 13 Dec 02:37:04.305 # +switch-master mymaster 10.133.17.203 6379 10.133.17.203 6380
20172:X 13 Dec 02:37:04.306 * +slave slave 10.133.17.203:6379 10.133.17.203 6379 @ mymaster 10.133.17.203 6380
20172:X 13 Dec 02:37:10.314 # +sdown slave 10.133.17.203:6379 10.133.17.203 6379 @ mymaster 10.133.17.203 6380

可见正确切换了。

等master机器回复后，启动他的sentinel：

20191:X 13 Dec 02:48:27.759 # Sentinel runid is 925ee651aac9fb852a75f33ad9c5364cad2268d9
20191:X 13 Dec 02:48:27.759 # +monitor master mymaster 10.133.17.203 6379 quorum 1
20191:X 13 Dec 02:48:29.190 * -dup-sentinel master mymaster 10.133.17.203 6379 #duplicate of 10.133.17.203:26380 or d4737f3a088135cf04d9a2efae7cda756ae9a053
20191:X 13 Dec 02:48:29.190 * +sentinel sentinel 10.133.17.203:26380 10.133.17.203 26380 @ mymaster 10.133.17.203 6379
20191:X 13 Dec 02:48:29.250 # +new-epoch 7
20191:X 13 Dec 02:48:29.250 # +config-update-from sentinel 10.133.17.203:26380 10.133.17.203 26380 @ mymaster 10.133.17.203 6379
20191:X 13 Dec 02:48:29.250 # +switch-master mymaster 10.133.17.203 6379 10.133.17.203 6380
20191:X 13 Dec 02:48:29.250 * +slave slave 10.133.17.203:6379 10.133.17.203 6379 @ mymaster 10.133.17.203 6380

他发现配置中sentinel重复了，因此自动更改。然后更新配置文件。这样，就恢复同步了。