先写结论：
通过设置从库切换的优先级，我们可以通过sentinel 来在线切换master到指定的实例，切换时间2s，并且重新建立主从关系，速度很快，从而满足我们迁移、维护硬件等需求。
环境：
Sentinel ： 192.168.164.231 26379
Master : 192.168.164.232 6379
Slave1: 192.168.164.231 6379
Slave2: 192.168.164.243 6379
使用主要命令：
config set slave-priority <parameter>
config get slave-priority <parameter>
Config rewrite
sentinel failover <parameter>
实验步奏：
现在我们需要把master切换到Slave2: 192.168.164.243 6379上：
1、
首先在slave2 ：192.168.164.243 6379
上设置从库切换优先级：
127.0.0.1:6379> config set slave-priority 200
OK
127.0.0.1:6379> config get slave-priority
1) "slave-priority"
2) "200"
2、
在Sentinel ： 192.168.164.231 26379
上执行：192.168.164.231:26379> sentinel failover test
OK
# Sentinel
sentinel_masters:1
sentinel_tilt:0
sentinel_running_scripts:0
sentinel_scripts_queue_length:0
master0:name=test,status=ok,address=192.168.164.232:6379,slaves=2,sentinels=1
192.168.164.231:26379> sentinel failover test
OK
192.168.164.231:26379> info Sentinel
# Sentinel
sentinel_masters:1
sentinel_tilt:0
sentinel_running_scripts:0
sentinel_scripts_queue_length:0
master0:name=test,status=ok,address=192.168.164.231:6379,slaves=2,sentinels=1
这里说明 "slave-priority"越高，该节点的优先级越差
3、
重新登录slave2 ：192.168.164.243 6379
127.0.0.1:6379> config set slave-priority 50
OK
127.0.0.1:6379> config get slave-priority
1) "slave-priority"
2) "50"
在Sentinel ： 192.168.164.231 26379
上执行：192.168.164.231:26379> sentinel failover test
OK
完成了到192.168.164.243:6379的切换
# Sentinel
sentinel_masters:1
sentinel_tilt:0
sentinel_running_scripts:0
sentinel_scripts_queue_length:0
master0:name=test,status=ok,address=192.168.164.243:6379,slaves=2,sentinels=1
从sentinel的log来看：
81889:X 21 Apr 10:19:10.717 # Executing user requested FAILOVER of 'test'
81889:X 21 Apr 10:19:10.717 # +new-epoch 867
81889:X 21 Apr 10:19:10.717 # +try-failover master test 192.168.164.231 6379
81889:X 21 Apr 10:19:10.798 # +vote-for-leader 1262f201dc53f23fd6f0da707785d90a861e7fe0 867
81889:X 21 Apr 10:19:10.798 # +elected-leader master test 192.168.164.231 6379
81889:X 21 Apr 10:19:10.798 # +failover-state-select-slave master test 192.168.164.231 6379
81889:X 21 Apr 10:19:10.853 # +selected-slave slave 192.168.164.243:6379 192.168.164.243 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:10.853 * +failover-state-send-slaveof-noone slave 192.168.164.243:6379 192.168.164.243 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:10.925 * +failover-state-wait-promotion slave 192.168.164.243:6379 192.168.164.243 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:11.878 # +promoted-slave slave 192.168.164.243:6379 192.168.164.243 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:11.878 # +failover-state-reconf-slaves master test 192.168.164.231 6379
81889:X 21 Apr 10:19:11.945 * +slave-reconf-sent slave 192.168.164.232:6379 192.168.164.232 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:12.931 * +slave-reconf-inprog slave 192.168.164.232:6379 192.168.164.232 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:12.931 * +slave-reconf-done slave 192.168.164.232:6379 192.168.164.232 6379 @ test 192.168.164.231 6379
81889:X 21 Apr 10:19:12.980 # +failover-end master test 192.168.164.231 6379
81889:X 21 Apr 10:19:12.981 # +switch-master test 192.168.164.231 6379 192.168.164.243 6379
81889:X 21 Apr 10:19:12.985 * +slave slave 192.168.164.232:6379 192.168.164.232 6379 @ test 192.168.164.243 6379
81889:X 21 Apr 10:19:12.985 * +slave slave 192.168.164.231:6379 192.168.164.231 6379 @ test 192.168.164.243 6379
81889:X 21 Apr 10:19:23.063 * +convert-to-slave slave 192.168.164.231:6379 192.168.164.231 6379 @ test 192.168.164.243 6379
切换主库需要时间2s，完成slave的重建10s（和数据量有关）
完全按照我们设置的优先级优先级来进行了切换。