Redis-ha(sentinel)搭建

服务器描述：本次搭建是用来测试，所以是在一台服务器上搭建三个redis服务（一主两从）

服务角色	端口	Redis.conf名称	sentinel配置文件名称	sentinel端口	redis日志路径	sentinel路劲
主(master)	6379	redis.conf	sentinel.conf	26379	/home/zhangxs/data/redislog/redis_server/master.log	/home/zhangxs/data/redislog/sentinel/sentinel6379.log
从(slave)	6380	redis_slave6380.conf	Sentinel6380.conf	26380	/home/zhangxs/data/redislog/redis_server/slave6380.log	/home/zhangxs/data/redislog/sentinel/sentinel6380.log
从(slave)	6381	redis_slave6381.conf	Sentinel6381.conf	26381	/home/zhangxs/data/redislog/redis_server/slave6381.log	/home/zhangxs/data/redislog/sentinel/sentinel6381.log

修改配置文件

1： redis.conf

修改redis服务日志路径：logfile "/home/zhangxs/data/redislog/redis_server/master.log"

其他没有修改，使用的是默认配置

2：redis_slave6380.conf 和redis_slave6381.conf (copy redis_slave.conf)

设置他们指向的master服务的ip和端口：slaveof 127.0.0.1 6379(两个文件都配置)
修改redis_slave6380.conf 和redis_slave6381.conf 端口:redis_slave6380.conf 端口为26380 ; redis_slave6381.conf 端口为26381
设置slave服务日志路径： logfile /home/zhangxs/data/redislog/redis_server/slave6380.log 和 logfile /home/zhangxs/data/redislog/redis_server/slave6381.log

3：sentinel.conf

修改日志路径：logfile "/home/zhangxs/data/redislog/sentinel/sentinel6379.log"

4:Sentinel6380.conf 和 Sentinel6381.conf (copy sentinel.conf)

修改端口号：Sentinel6380.conf 改为 26380; Sentinel6381.conf 改为 26381;
修改日志路径：logfile "/home/zhangxs/data/redislog/sentinel/sentinel6380.log" 和 logfile "/home/zhangxs/data/redislog/sentinel/sentinel6381.log"

上面配置好后，启动redis服务

1：启动master

src/redis-server redis.conf&

2:启动从服务(slave6380)

src/redis-server redis_slave6380.conf &

查看主服务日志，会发现多出来一段

2755:M 29 Jul 00:13:25.469 * Starting BGSAVE for SYNC with target: disk
2755:M 29 Jul 00:13:25.502 * Background saving started by pid 2784
2784:C 29 Jul 00:13:25.608 * DB saved on disk
2784:C 29 Jul 00:13:25.608 * RDB: 6 MB of memory used by copy-on-write
2755:M 29 Jul 00:13:25.677 * Background saving terminated with success
2755:M 29 Jul 00:13:25.677 * Synchronization with slave 127.0.0.1:6380 succeeded   //同步到从服务127.0.0.1:6380成功

查看从服务日志slave6380.log

2780:S 29 Jul 00:13:25.436 * DB loaded from disk: 0.007 seconds
2780:S 29 Jul 00:13:25.436 * Before turning into a slave, using my master parameters to synthesize a cached master: I may be able to synchronize with the new master with just a partial transfer.
2780:S 29 Jul 00:13:25.436 * Ready to accept connections
2780:S 29 Jul 00:13:25.436 * Connecting to MASTER 127.0.0.1:6379 //正在连接master服务
2780:S 29 Jul 00:13:25.436 * MASTER <-> SLAVE sync started //master到slave 启动同步
2780:S 29 Jul 00:13:25.436 * Non blocking connect for SYNC fired the event.
2780:S 29 Jul 00:13:25.436 * Master replied to PING, replication can continue...
2780:S 29 Jul 00:13:25.436 * Trying a partial resynchronization (request a26696cff5d35f38896c4eb068f71adbb7cfc421:474104).
2780:S 29 Jul 00:13:25.577 * Full resync from master: 541cd938f43b4f144e647881af409fa1884ea5a4:0 //从master全量同步
2780:S 29 Jul 00:13:25.577 * Discarding previously cached master state.//丢弃之前缓存的master状态
2780:S 29 Jul 00:13:25.677 * MASTER <-> SLAVE sync: receiving 250 bytes from master //slave从master 同步250个字节
2780:S 29 Jul 00:13:25.677 * MASTER <-> SLAVE sync: Flushing old data
2780:S 29 Jul 00:13:25.677 * MASTER <-> SLAVE sync: Loading DB in memory
2780:S 29 Jul 00:13:25.678 * MASTER <-> SLAVE sync: Finished with success

3:启动从服务(slave6381)

src/redis-server redis_slave6381.conf &

查看主服务日志，会发现多出来一段

//从服务127.0.0.1:6381发送同步请求
2755:M 29 Jul 00:26:46.531 * Slave 127.0.0.1:6381 asks for synchronization
//接受127.0.1:6381的部分重同步请求。从偏移量1开始发送积压的1106字节
2755:M 29 Jul 00:26:46.531 * Partial resynchronization request from 127.0.0.1:6381 accepted. Sending 1106 bytes of backlog starting from offset 1.

查看从服务日志slave6381.log

2809:S 29 Jul 00:26:46.531 * DB loaded from disk: 0.000 seconds
2809:S 29 Jul 00:26:46.531 * Before turning into a slave, using my master parameters to synthesize a cached master: I may be able to synchronize with the new master with just a partial transfer.
2809:S 29 Jul 00:26:46.531 * Ready to accept connections
2809:S 29 Jul 00:26:46.531 * Connecting to MASTER 127.0.0.1:6379
2809:S 29 Jul 00:26:46.531 * MASTER <-> SLAVE sync started
2809:S 29 Jul 00:26:46.531 * Non blocking connect for SYNC fired the event.
2809:S 29 Jul 00:26:46.531 * Master replied to PING, replication can continue...
2809:S 29 Jul 00:26:46.531 * Trying a partial resynchronization (request 541cd938f43b4f144e647881af409fa1884ea5a4:1).
2809:S 29 Jul 00:26:46.531 * Successful partial resynchronization with master.
2809:S 29 Jul 00:26:46.532 * MASTER <-> SLAVE sync: Master accepted a Partial Resynchronization

4：测试数据同步功能（使用redis-cli 连接服务器）

1：连接master

[root@vm1 src]# redis-cli -h 127.0.0.1 -p 6379
127.0.0.1:6379> set name fj

2:连接slave6380

[root@vm1 src]# redis-cli -h 127.0.0.1 -p 6380
127.0.0.1:6380> get name
"fj"
127.0.0.1:6380>

3:连接slave6381

[root@vm1 src]# redis-cli -h 127.0.0.1 -p 6381
127.0.0.1:6381> get name
"fj"
127.0.0.1:6381>

Ok 主从同步没有问题。

默认情况下从服务是不允许set数据的，测试下

127.0.0.1:6380> set name hello
(error) READONLY You can't write against a read only slave.
127.0.0.1:6380>

127.0.0.1:6381> set name hello
(error) READONLY You can't write against a read only slave.
127.0.0.1:6381>

启动各个服务的sentinel

启动sentinel6379

src/redis-sentinel sentinel.conf &

查看Sentinel6379.log

2908:X 29 Jul 01:01:32.838 # oO0OoO0OoO0Oo Redis is starting oO0OoO0OoO0Oo
2908:X 29 Jul 01:01:32.839 # Redis version=4.0.10, bits=64, commit=00000000, modified=0, pid=2908, just started
2908:X 29 Jul 01:01:32.839 # Configuration loaded
2908:X 29 Jul 01:01:32.839 * Increased maximum number of open files to 10032 (it was originally set to 1024).
2908:X 29 Jul 01:01:32.840 * Running mode=sentinel, port=26379.
2908:X 29 Jul 01:01:32.840 # WARNING: The TCP backlog setting of 511 cannot be enforced because /proc/sys/net/core/somaxconn is set to the lower value of 128.
2908:X 29 Jul 01:01:32.855 # Sentinel ID is 1a77392638e41bb0ea0a865ffc93b8de6335227f
2908:X 29 Jul 01:01:32.855 # +monitor master mymaster 127.0.0.1 6379 quorum 2
//一个新的从服务器已经被sentinel识别并关联
2908:X 29 Jul 01:01:32.856 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
2908:X 29 Jul 01:01:32.858 * +slave slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379

启动sentinel6380

 src/redis-sentinel sentinel6380.conf &

查看 Sentinel6380.log

2937:X 29 Jul 01:08:14.325 # Redis version=4.0.10, bits=64, commit=00000000, modified=0, pid=2937, just started
2937:X 29 Jul 01:08:14.325 # Configuration loaded
2937:X 29 Jul 01:08:14.327 * Increased maximum number of open files to 10032 (it was originally set to 1024).
2937:X 29 Jul 01:08:14.377 * Running mode=sentinel, port=26380.
2937:X 29 Jul 01:08:14.377 # WARNING: The TCP backlog setting of 511 cannot be enforced because /proc/sys/net/core/somaxconn is set to the lower value of 128.
2937:X 29 Jul 01:08:14.379 # Sentinel ID is 4a6aebffdd1301bf054e722c34e8a6611418ba8a
2937:X 29 Jul 01:08:14.379 # +monitor master mymaster 127.0.0.1 6379 quorum 2
2937:X 29 Jul 01:08:14.380 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
2937:X 29 Jul 01:08:14.381 * +slave slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
//一个新的sentinel（监控）已经被识别并关联
2937:X 29 Jul 01:08:14.919 * +sentinel sentinel 1a77392638e41bb0ea0a865ffc93b8de6335227f 127.0.0.1 26379 @ mymaster 127.0.0.1 6379

Sentinel6380启动后会发现，Sentinel6379.log 加了一段日志

//一个新的sentinel（监控）已经被识别并关联
2908:X 29 Jul 01:08:16.367 * +sentinel sentinel 4a6aebffdd1301bf054e722c34e8a6611418ba8a 127.0.0.1 26380 @ mymaster 127.0.0.1 6379

新启动一个sentinel，会通过发布订阅功能自动发现监控相同master下的其他sentinel。这一功能是通过向频道 sentinel:hello 发送信息来实现的。

启动sentinel6381

[root@vm1 redis-4.0.10]# src/redis-sentinel sentinel6381.conf &

查看 Sentinel6381.log

2961:X 29 Jul 01:11:09.823 # Configuration loaded
2961:X 29 Jul 01:11:09.823 * Increased maximum number of open files to 10032 (it was originally set to 1024).
2961:X 29 Jul 01:11:09.852 * Running mode=sentinel, port=26381.
2961:X 29 Jul 01:11:09.852 # WARNING: The TCP backlog setting of 511 cannot be enforced because /proc/sys/net/core/somaxconn is set to the lower value of 128.
2961:X 29 Jul 01:11:09.853 # Sentinel ID is 1db1a4dcdf0ecca00b64d9362c2a2dd338da0030
2961:X 29 Jul 01:11:09.853 # +monitor master mymaster 127.0.0.1 6379 quorum 2
2961:X 29 Jul 01:11:09.853 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
2961:X 29 Jul 01:11:09.855 * +slave slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
2961:X 29 Jul 01:11:10.334 * +sentinel sentinel 1a77392638e41bb0ea0a865ffc93b8de6335227f 127.0.0.1 26379 @ mymaster 127.0.0.1 6379
2961:X 29 Jul 01:11:11.446 * +sentinel sentinel 4a6aebffdd1301bf054e722c34e8a6611418ba8a 127.0.0.1 26380 @ mymaster 127.0.0.1 6379

新加入的sentinel6381， sentinel6379和sentinel6380 都会收到通知(//一个新的sentinel（监控）已经被识别并关联)

2908:X 29 Jul 01:11:11.880 * +sentinel sentinel 1db1a4dcdf0ecca00b64d9362c2a2dd338da0030 127.0.0.1 26381 @ mymaster 127.0.0.1 6379

2937:X 29 Jul 01:11:11.878 * +sentinel sentinel 1db1a4dcdf0ecca00b64d9362c2a2dd338da0030 127.0.0.1 26381 @ mymaster 127.0.0.1 6379

sentinel的状态都会记录到sentinel.conf文件中，用于启动后恢复状态，查看下各个sentinel.conf 文件变动后的部分

sentinel.conf

启动前：无启动后：sentinel myid 1a77392638e41bb0ea0a865ffc93b8de6335227f //自己的sentinel myid

启动前：无启动后：

# Generated by CONFIG REWRITE

#master下得两个从服务

sentinel known-slave mymaster 127.0.0.1 6380

sentinel known-slave mymaster 127.0.0.1 6381

#master下其他两个sentinel

sentinel known-sentinel mymaster 127.0.0.1 26380 4a6aebffdd1301bf054e722c34e8a6611418ba8a

sentinel known-sentinel mymaster 127.0.0.1 26381 1db1a4dcdf0ecca00b64d9362c2a2dd338da0030

sentinel current-epoch 0

Sentinel6380.conf和Sentinel6381.conf改动和sentinel.conf 基本一样。不一样的就是记录自己sentinel myid和master下其他两个sentinel不一样，大同小异。

测试故障迁移

Sentinel 故障迁移我使用的是默认配置(不需要再配置，可以自定义修改)

//判断master失效，至少有两个sentinel同意才会执行故障迁移
sentinel monitor mymaster 127.0.0.1 6379 2 
//如果在10秒内sentinel 都收到master的一次有效回复，就认为该master主观下线
sentinel down-after-milliseconds mymaster 60000 
sentinel failover-timeout mymaster 180000
//在执行故障转移时，同时只有一个slave能对新的master进行数据同步
sentinel parallel-syncs mymaster 1 
sentinel monitor resque 192.168.1.3 6380 4 
sentinel down-after-milliseconds resque 10000 
sentinel failover-timeout resque 180000 
sentinel parallel-syncs resque 5

1：查看redis的相关服务

root       2755   2551  0 00:11 pts/2    00:00:06 src/redis-server 127.0.0.1:6379
root       2780   2551  0 00:13 pts/2    00:00:06 src/redis-server 127.0.0.1:6380
root       2809   2551  0 00:26 pts/2    00:00:05 src/redis-server 127.0.0.1:6381
root       2816   2529  0 00:30 pts/1    00:00:00 redis-cli -h 127.0.0.1 -p 6379
root       2822   2530  0 00:33 pts/0    00:00:00 redis-cli -h 127.0.0.1 -p 6380
root       2841   2823  0 00:34 pts/6    00:00:00 redis-cli -h 127.0.0.1 -p 6381
root       2908   2551  0 01:01 pts/2    00:00:07 src/redis-sentinel *:26379 [sentinel]
root       2937   2551  0 01:08 pts/2    00:00:06 src/redis-sentinel *:26380 [sentinel]
root       2961   2551  0 01:11 pts/2    00:00:06 src/redis-sentinel *:26381 [sentinel]
root       3000   2551  0 01:50 pts/2    00:00:00 grep --color=auto redis

2：查看整个备份状态

127.0.0.1:6379> info

# Server
# Clients
# Memory
# Persistence
# Stats
# Replication
role:master
connected_slaves:2
slave0:ip=127.0.0.1,port=6380,state=online,offset=544591,lag=0
slave1:ip=127.0.0.1,port=6381,state=online,offset=544591,lag=0
master_replid:541cd938f43b4f144e647881af409fa1884ea5a4
master_replid2:0000000000000000000000000000000000000000
master_repl_offset:544857
second_repl_offset:-1
repl_backlog_active:1
repl_backlog_size:1048576
repl_backlog_first_byte_offset:1
repl_backlog_histlen:544857
# CPU

# Cluster

# Keyspace

其他信息我都删掉了，只留下【Replication】的信息，其他信息可以在redis-cli 命令行中使用【info】命令查看

可以看到，6379是master角色，master下有两个从服务port=6380，port=6381

3: Kill 掉 master，观察日志

kill -9 2755

master被干掉了，所以master.log 没有日志，看其他两个从服务日志（截取部分）

redis_slave6380.log

///////////////////////////一分钟内
2780:S 29 Jul 01:58:53.414 # Connection with master lost.
2780:S 29 Jul 01:58:53.414 * Caching the disconnected master state.
2780:S 29 Jul 01:58:54.163 * Connecting to MASTER 127.0.0.1:6379
2780:S 29 Jul 01:58:54.163 * MASTER <-> SLAVE sync started
2780:S 29 Jul 01:58:54.164 # Error condition on socket for SYNC: Connection refused
2780:S 29 Jul 01:58:55.168 * Connecting to MASTER 127.0.0.1:6379
2780:S 29 Jul 01:58:55.169 * MASTER <-> SLAVE sync started
....
...
...
2780:S 29 Jul 01:59:22.381 # Error condition on socket for SYNC: Connection refused
2780:S 29 Jul 01:59:23.389 * Connecting to MASTER 127.0.0.1:6379
2780:S 29 Jul 01:59:23.389 * MASTER <-> SLAVE sync started
2780:S 29 Jul 01:59:23.389 # Error condition on socket for SYNC: Connection refused
///////////////////////////一分钟后
2780:S 29 Jul 01:59:24.321 * SLAVE OF 127.0.0.1:6381 enabled (user request from 'id=8 addr=127.0.0.1:52556 fd=11 name=sentinel-4a6aebff-cmd age=3070 idle=0 flags=x db=0 sub=0 psub=0 multi=3 qbuf=133 qbuf-free=32635 obl=36 oll=0 omem=0 events=r cmd=exec')
2780:S 29 Jul 01:59:24.323 # CONFIG REWRITE executed with suc
2780:S 29 Jul 01:59:24.399 * Connecting to MASTER 127.0.0.1:6381
2780:S 29 Jul 01:59:24.399 * MASTER <-> SLAVE sync started
2780:S 29 Jul 01:59:24.399 * Non blocking connect for SYNC fired the event.
2780:S 29 Jul 01:59:24.399 * Master replied to PING, replication can continue...
2780:S 29 Jul 01:59:24.399 * Trying a partial resynchronization (request 541cd938f43b4f144e647881af409fa1884ea5a4:617714).
2780:S 29 Jul 01:59:24.400 * Successful partial resynchronization with master.
2780:S 29 Jul 01:59:24.400 # Master replication ID changed to 514edab0972b4b6e5388edc4f14fbdb4d223d39e
2780:S 29 Jul 01:59:24.400 * MASTER <-> SLAVE sync: Master accepted a Partial Resynchronization.

从01:58:53.414 到01:59:23.389 这一分钟内一直在尝试连接master，一分钟内没有连接成功后，sentinel 就会master判断为主观下线，看日志

Sentinel6379.log

//判定master 主观下线
2908:X 29 Jul 01:59:23.492 # +sdown master mymaster 127.0.0.1 6379
//当前的纪元（epoch）已经被更新。
2908:X 29 Jul 01:59:23.546 # +new-epoch 1
 //开始给sentinel6380投票,谁来主导这次故障转移
2908:X 29 Jul 01:59:23.549 # +vote-for-leader 4a6aebffdd1301bf054e722c34e8a6611418ba8a 1
//判定master 客观观下线,已经有2个sentinel同意
2908:X 29 Jul 01:59:23.569 # +odown master mymaster 127.0.0.1 6379 #quorum 3/2
2908:X 29 Jul 01:59:23.569 # Next failover delay: I will not start a failover before Sun Jul 29 02:05:23 2018
2908:X 29 Jul 01:59:24.328 # +config-update-from sentinel 4a6aebffdd1301bf054e722c34e8a6611418ba8a 127.0.0.1 26380 @ mymaster 127.0.0.1 6379
2908:X 29 Jul 01:59:24.328 # +switch-master mymaster 127.0.0.1 6379 127.0.0.1 6381
2908:X 29 Jul 01:59:24.328 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6381
2908:X 29 Jul 01:59:24.328 * +slave slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381
2908:X 29 Jul 01:59:54.333 # +sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

Sentinel6380.log

//判定master 主观下线
2937:X 29 Jul 01:59:23.459 # +sdown master mymaster 127.0.0.1 6379
//判定master 客观观下线,已经有2个sentinel同意
2937:X 29 Jul 01:59:23.536 # +odown master mymaster 127.0.0.1 6379 #quorum 2/2
2937:X 29 Jul 01:59:23.537 # +new-epoch 1
//尝试故障转移master
2937:X 29 Jul 01:59:23.537 # +try-failover master mymaster 127.0.0.1 6379
 //开始给sentinel6380投票,谁来主导这次故障转移
2937:X 29 Jul 01:59:23.540 # +vote-for-leader 4a6aebffdd1301bf054e722c34e8a6611418ba8a 1
//其他两个sentinel 都投票给4a6aebffdd1301bf054e722c34e8a6611418ba8a 【6380sentinel】
2937:X 29 Jul 01:59:23.549 # 1db1a4dcdf0ecca00b64d9362c2a2dd338da0030 voted for 4a6aebffdd1301bf054e722c34e8a6611418ba8a 1
2937:X 29 Jul 01:59:23.549 # 1a77392638e41bb0ea0a865ffc93b8de6335227f voted for 4a6aebffdd1301bf054e722c34e8a6611418ba8a 1
//6379这个服务赢得选举可以进行故障转移
2937:X 29 Jul 01:59:23.619 # +elected-leader master mymaster 127.0.0.1 6379
//发现6379这个服务是故障转移状态，就开始选择master下得从服务
2937:X 29 Jul 01:59:23.619 # +failover-state-select-slave master mymaster 127.0.0.1 6379
//故障转移操作现在处于 select-slave 状态 —— Sentinel 正在寻找可以升级为主服务器的从服务器。(选择mymaster 127.0.0.1 6379 下 6381 的从服务)
2937:X 29 Jul 01:59:23.710 # +selected-slave slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
//Sentinel 正在将6379下的从服务器6381升级为主服务器，等待升级功能完成。
2937:X 29 Jul 01:59:23.710 * +failover-state-send-slaveof-noone slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
//master下的从服务 6381  等待升级
2937:X 29 Jul 01:59:23.769 * +failover-state-wait-promotion slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
//升级master下从服务6381
2937:X 29 Jul 01:59:24.251 # +promoted-slave slave 127.0.0.1:6381 127.0.0.1 6381 @ mymaster 127.0.0.1 6379
//故障转移状态切换到了 reconf-slaves 状态。（再次确认从服务器转为主服务器）
2937:X 29 Jul 01:59:24.251 # +failover-state-reconf-slaves master mymaster 127.0.0.1 6379
//牵头的sentinel 向6380从服务器发送slaveof 指令，将它设置为新的master
2937:X 29 Jul 01:59:24.321 * +slave-reconf-sent slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
//6379不再处于客观下线状态，客观下线状态只用于master服务，6379已经不是master了
2937:X 29 Jul 01:59:24.653 # -odown master mymaster 127.0.0.1 6379
//6380服务正在将自己设置为6381主服务的从服务器，还未完成
2937:X 29 Jul 01:59:25.270 * +slave-reconf-inprog slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
//从服务器6380已经完成对新master服务的同步
2937:X 29 Jul 01:59:25.271 * +slave-reconf-done slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6379
//master6379 故障转移结束，所有的从服务器开始同步新的master
2937:X 29 Jul 01:59:25.347 # +failover-end master mymaster 127.0.0.1 6379
//配置变更主服务器的ip地址已经改变， 选择master 为6381
2937:X 29 Jul 01:59:25.347 # +switch-master mymaster 127.0.0.1 6379 127.0.0.1 6381
//6381下的两个从服务（新的从服务被识别并关联）
2937:X 29 Jul 01:59:25.347 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6381
2937:X 29 Jul 01:59:25.347 * +slave slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381
//添加master下从服务6379 为客观下线
2937:X 29 Jul 01:59:55.401 # +sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

Sentinel6381.log

//判定master 主观下线
2961:X 29 Jul 01:59:23.459 # +sdown master mymaster 127.0.0.1 6379 
2961:X 29 Jul 01:59:23.545 # +new-epoch 1
//开始给6380投票
2961:X 29 Jul 01:59:23.548 # +vote-for-leader 4a6aebffdd1301bf054e722c34e8a6611418ba8a 1 
2961:X 29 Jul 01:59:24.325 # +config-update-from sentinel 4a6aebffdd1301bf054e722c34e8a6611418ba8a 127.0.0.1 26380 @ mymaster 127.0.0.1 6379
2961:X 29 Jul 01:59:24.325 # +switch-master mymaster 127.0.0.1 6379 127.0.0.1 6381
2961:X 29 Jul 01:59:24.325 * +slave slave 127.0.0.1:6380 127.0.0.1 6380 @ mymaster 127.0.0.1 6381
2961:X 29 Jul 01:59:24.325 * +slave slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381
2961:X 29 Jul 01:59:54.348 # +sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

可以看到在 01:59:23秒也就是一分钟之后，三个监控master的sentinel 都判定了master为主观下线(sdown)，我们配置的至少有2个sentinel 同意master 主观下线，master就会被切换到客观下线(odown) 【+odown master mymaster 127.0.0.1 6379 #quorum 2/2】。当判断master为客观下线后，sentinel 就开始选举出新的master，可以看到Sentinel6380.log 日志要比其他的sentinel.log多，因为整个选举的过程是Sentinel6380 在牵头执行。

4：在6381下查看整个服务的备份状态

# Replication
role:master
connected_slaves:1
slave0:ip=127.0.0.1,port=6380,state=online,offset=1125775,lag=0
master_replid:514edab0972b4b6e5388edc4f14fbdb4d223d39e
master_replid2:541cd938f43b4f144e647881af409fa1884ea5a4
master_repl_offset:1125775
second_repl_offset:617714
repl_backlog_active:1
repl_backlog_size:1048576
repl_backlog_first_byte_offset:77200
repl_backlog_histlen:1048576

可以看到6381的角色成为了master，只有一个slave，因为另一个挂了。

5：再次启动6379服务

查看6379服务的日志

3037:S 29 Jul 02:46:47.753 # CONFIG REWRITE executed with success.
3037:S 29 Jul 02:46:48.359 * Connecting to MASTER 127.0.0.1:6381
3037:S 29 Jul 02:46:48.360 * MASTER <-> SLAVE sync started
3037:S 29 Jul 02:46:48.360 * Non blocking connect for SYNC fired the event.
3037:S 29 Jul 02:46:48.361 * Master replied to PING, replication can continue...
3037:S 29 Jul 02:46:48.362 * Trying a partial resynchronization (request 7b0dc6ac9c2188e3c92eb29eea200ea6c572619c:1).
3037:S 29 Jul 02:46:48.608 * Full resync from master: 514edab0972b4b6e5388edc4f14fbdb4d223d39e:1178142
3037:S 29 Jul 02:46:48.608 * Discarding previously cached master state.
3037:S 29 Jul 02:46:48.708 * MASTER <-> SLAVE sync: receiving 253 bytes from master
3037:S 29 Jul 02:46:48.708 * MASTER <-> SLAVE sync: Flushing old data
3037:S 29 Jul 02:46:48.708 * MASTER <-> SLAVE sync: Loading DB in memory
3037:S 29 Jul 02:46:48.708 * MASTER <-> SLAVE sync: Finished with success
3037:S 29 Jul 02:46:48.709 * Background append only file rewriting started by pid 3042
3037:S 29 Jul 02:46:48.750 * AOF rewrite child asks to stop sending diffs.
3042:C 29 Jul 02:46:48.750 * Parent agreed to stop sending diffs. Finalizing AOF...
3042:C 29 Jul 02:46:48.750 * Concatenating 0.00 MB of AOF diff received from parent.
3042:C 29 Jul 02:46:48.750 * SYNC append only file rewrite performed
3042:C 29 Jul 02:46:48.750 * AOF rewrite: 6 MB of memory used by copy-on-write
3037:S 29 Jul 02:46:48.781 * Background AOF rewrite terminated with success
3037:S 29 Jul 02:46:48.782 * Residual parent diff successfully flushed to the rewritten AOF (0.00 MB)
3037:S 29 Jul 02:46:48.782 * Background AOF rewrite finished successfully

看到启动后重写配置文件，然后自动连接6381这个新的master服务，开始从master 上全量同步数据

查看6381这个新master日志

//响应6379的同步请求
2809:M 29 Jul 02:46:48.362 * Slave 127.0.0.1:6379 asks for synchronization 
//不接受同步部分数据请求
2809:M 29 Jul 02:46:48.362 * Partial resynchronization not accepted: Replication ID mismatch (Slave asked for '7b0dc6ac9c2188e3c92eb29eea200ea6c572619c', my replication IDs are '514edab0972b4b6e5388edc4f14fbdb4d223d39e' and '541cd938f43b4f144e647881af409fa1884ea5a4')
//开始同步
2809:M 29 Jul 02:46:48.362 * Starting BGSAVE for SYNC with target: disk
2809:M 29 Jul 02:46:48.607 * Background saving started by pid 3041
3041:C 29 Jul 02:46:48.607 * DB saved on disk
#6m内存用于写复制
3041:C 29 Jul 02:46:48.608 * RDB: 6 MB of memory used by copy-on-write
//后台保存成功
2809:M 29 Jul 02:46:48.708 * Background saving terminated with success
2809:M 29 Jul 02:46:48.708 * Synchronization with slave 127.0.0.1:6379 succeeded

查看sentinel日志

6379sentinel.log

//减去6379服务的主观下线状态
2908:X 29 Jul 02:46:37.610 # -sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381
//转换为master6381 下的从服务
2908:X 29 Jul 02:46:47.628 * +convert-to-slave slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

6380sentinel.log

2937:X 29 Jul 02:46:37.767 # -sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

6381sentinel.log

2961:X 29 Jul 02:46:38.023 # -sdown slave 127.0.0.1:6379 127.0.0.1 6379 @ mymaster 127.0.0.1 6381

再看下6381这个新master的整个备份信息

# Replication
role:master
connected_slaves:2
slave0:ip=127.0.0.1,port=6380,state=online,offset=1348132,lag=0
slave1:ip=127.0.0.1,port=6379,state=online,offset=1348132,lag=0
master_replid:514edab0972b4b6e5388edc4f14fbdb4d223d39e
master_replid2:541cd938f43b4f144e647881af409fa1884ea5a4
master_repl_offset:1348132
second_repl_offset:617714
repl_backlog_active:1
repl_backlog_size:1048576
repl_backlog_first_byte_offset:299557
repl_backlog_histlen:1048576

新的master 增加了一个slave6379 从服务

我们再搭建前做的redis配置，当故障转移成功后，这些配置会被重写，重写的内容基本都在配置文件的最后

Redis.conf配置文件，多了

# Generated by CONFIG REWRITE
slaveof 127.0.0.1 6381

Redis6381.conf配置文件没有变

# Master-Slave replication. Use slaveof to make a Redis instance a copy of
# slaveof <masterip> <masterport>
slaveof 127.0.0.1 6381

新的master配置，也就是redis_slave6381.conf 已经没有了slaveof 配置

# Master-Slave replication. Use slaveof to make a Redis instance a copy of
# slaveof <masterip> <masterport>

Sentinel.conf 也会发生变化，可以自己去看看

6：再次测试故障转移后的同步功能

之前的master已经不再支持set

127.0.0.1:6379> set name zhangxs
(error) READONLY You can't write against a read only slave.

新master set成功

127.0.0.1:6381> set name zhangxs
OK

127.0.0.1:6379> get name
"zhangxs"

127.0.0.1:6380> get name
"zhangxs"

同步没问题。

转移后的服务器变成了

服务角色	端口	Redis.conf名称	sentinel配置文件名称	sentinel端口	redis日志路径	sentinel路劲
从(master)	6379	redis.conf	sentinel.conf	26379	/home/zhangxs/data/redislog/redis_server/master.log	/home/zhangxs/data/redislog/sentinel/sentinel6379.log
从(slave)	6380	redis_slave6380.conf	Sentinel6380.conf	26380	/home/zhangxs/data/redislog/redis_server/slave6380.log	/home/zhangxs/data/redislog/sentinel/sentinel6380.log
主(slave)	6381	redis_slave6381.conf	Sentinel6381.conf	26381	/home/zhangxs/data/redislog/redis_server/slave6381.log	/home/zhangxs/data/redislog/sentinel/sentinel6381.log

参考文档：http://www.redis.cn/topics

Redis-ha(sentinel)搭建

猜你喜欢