Redis/Sentinel 集群故障转移导致“故障转移-中止-未选举主服务器”

Question

我有一个 2 节点 Redis 集群设置。

[主站] 192.168.56.102：Redis主站（：6379），Redis从站（：6380），哨兵（：26379），哨兵＃2（：26380）

[救援]192.168.56.103：Redis Master（：6379），Redis Slave（：6380），Sentinel（：26379）

每个从属实例都是同一机器上主实例的从属实例。每个哨兵实例监控两个主实例。

我将上述内容与 twemproxy （与这个问题无关）和 client-reconfig-script 结合使用来更新 twemproxy 配置，以便应用程序继续工作。

我正在停止服务器以查看发生了什么以及一切是否正常工作。

[master] stop redis master：法定人数能够成功选举新的master。登录下面。

==> /tmp/sentinel.log <==
[14701] 29 Dec 18:16:55.096 # +sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:16:55.096 # +sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14701] 29 Dec 18:18:04.187 # -sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:18:04.236 # -sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:18:14.160 * +convert-to-slave slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:47:47.151 # +sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14701] 29 Dec 18:47:47.170 # +sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14701] 29 Dec 18:47:57.650 * +reboot slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:47:57.652 * +reboot slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:47:57.715 # -sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14701] 29 Dec 18:47:57.738 # -sdown slave 192.168.56.102:6379 192.168.56.102 6379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:48:08.088 # +sdown master master 192.168.56.102 6380
[14701] 29 Dec 18:48:08.180 # +sdown master master 192.168.56.102 6380
[14701] 29 Dec 18:48:08.280 # +odown master master 192.168.56.102 6380 #quorum 2/2
[14701] 29 Dec 18:48:08.280 # +new-epoch 73
[14701] 29 Dec 18:48:08.280 # +try-failover master master 192.168.56.102 6380
[14701] 29 Dec 18:48:08.471 # +vote-for-leader a664b9f61df2b10bbbb5d865b01c599ddd36183c 73
[14701] 29 Dec 18:48:08.472 # 192.168.56.103:26379 voted for 491b32b95c547a8266faf9b04ce6b0c18486236b 73
[14705] 29 Dec 18:48:08.473 # +new-epoch 73
[14705] 29 Dec 18:48:08.475 # +vote-for-leader 491b32b95c547a8266faf9b04ce6b0c18486236b 73
[14701] 29 Dec 18:48:08.475 # 192.168.56.102:26380 voted for 491b32b95c547a8266faf9b04ce6b0c18486236b 73
[14701] 29 Dec 18:48:08.835 # +config-update-from sentinel 192.168.56.103:26379 192.168.56.103 26379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:48:08.835 # +config-update-from sentinel 192.168.56.103:26379 192.168.56.103 26379 @ master 192.168.56.102 6380
[14705] 29 Dec 18:48:08.836 # +switch-master master 192.168.56.102 6380 192.168.56.102 6379
[14701] 29 Dec 18:48:08.836 # +switch-master master 192.168.56.102 6380 192.168.56.102 6379
[14701] 29 Dec 18:48:08.836 * +slave slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14705] 29 Dec 18:48:08.836 * +slave slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379

==> /tmp/sent.log <==
Failing master: 192.168.56.102:6380
New Master: 192.168.56.102:6379

Failing master: 192.168.56.102:6380
New Master: 192.168.56.102:6379


==> /tmp/sentinel.log <==
[14705] 29 Dec 18:48:11.855 # +sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14701] 29 Dec 18:48:11.903 # +sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379

您还可以看到我的重新配置脚本被成功调用并在屏幕上输出一些详细信息。

问题是当我试图停止[救援]机器上的主实例时。哨兵不断报“-failover-abort-not-elected master resque 192.168.56.103 6379”

==> /tmp/sentinel.log <==
[14705] 29 Dec 18:48:11.855 # +sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14701] 29 Dec 18:48:11.903 # +sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14705] 29 Dec 18:48:43.401 # -sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14701] 29 Dec 18:48:43.433 # -sdown slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14705] 29 Dec 18:48:53.344 * +convert-to-slave slave 192.168.56.102:6380 192.168.56.102 6380 @ master 192.168.56.102 6379
[14705] 29 Dec 18:49:23.617 # +sdown master resque 192.168.56.103 6379
[14701] 29 Dec 18:49:23.625 # +sdown master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:23.674 # +odown master resque 192.168.56.103 6379 #quorum 2/2
[14705] 29 Dec 18:49:23.674 # +new-epoch 74
[14705] 29 Dec 18:49:23.674 # +try-failover master resque 192.168.56.103 6379
[14701] 29 Dec 18:49:23.727 # +odown master resque 192.168.56.103 6379 #quorum 3/2
[14701] 29 Dec 18:49:23.727 # +new-epoch 74
[14701] 29 Dec 18:49:23.727 # +try-failover master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:23.886 # +vote-for-leader b608fcab7a201799826f4d9ee839aed3cf556fdf 74
[14701] 29 Dec 18:49:23.889 # +vote-for-leader a664b9f61df2b10bbbb5d865b01c599ddd36183c 74
[14701] 29 Dec 18:49:23.890 # 192.168.56.102:26380 voted for b608fcab7a201799826f4d9ee839aed3cf556fdf 74
[14705] 29 Dec 18:49:23.890 # 192.168.56.102:26379 voted for a664b9f61df2b10bbbb5d865b01c599ddd36183c 74
[14705] 29 Dec 18:49:23.893 # 192.168.56.103:26379 voted for b608fcab7a201799826f4d9ee839aed3cf556fdf 74
[14701] 29 Dec 18:49:23.893 # 192.168.56.103:26379 voted for b608fcab7a201799826f4d9ee839aed3cf556fdf 74
[14705] 29 Dec 18:49:23.980 # +elected-leader master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:23.980 # +failover-state-select-slave master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:24.064 # -failover-abort-no-good-slave master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:24.117 # Next failover delay: I will not start a failover before Mon Dec 29 18:49:32 2014
[14701] 29 Dec 18:49:28.417 # -failover-abort-not-elected master resque 192.168.56.103 6379
[14701] 29 Dec 18:49:28.489 # Next failover delay: I will not start a failover before Mon Dec 29 18:49:32 2014
[14705] 29 Dec 18:49:32.217 # +new-epoch 75
[14705] 29 Dec 18:49:32.217 # +try-failover master resque 192.168.56.103 6379
[14701] 29 Dec 18:49:32.423 # +new-epoch 75
[14701] 29 Dec 18:49:32.424 # +try-failover master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:32.433 # +vote-for-leader b608fcab7a201799826f4d9ee839aed3cf556fdf 75
[14705] 29 Dec 18:49:32.435 # 192.168.56.103:26379 voted for 491b32b95c547a8266faf9b04ce6b0c18486236b 75
[14701] 29 Dec 18:49:32.437 # +vote-for-leader a664b9f61df2b10bbbb5d865b01c599ddd36183c 75
[14701] 29 Dec 18:49:32.438 # 192.168.56.102:26380 voted for b608fcab7a201799826f4d9ee839aed3cf556fdf 75
[14705] 29 Dec 18:49:32.438 # 192.168.56.102:26379 voted for a664b9f61df2b10bbbb5d865b01c599ddd36183c 75
[14701] 29 Dec 18:49:32.438 # 192.168.56.103:26379 voted for 491b32b95c547a8266faf9b04ce6b0c18486236b 75
[14705] 29 Dec 18:49:36.636 # -failover-abort-not-elected master resque 192.168.56.103 6379
[14705] 29 Dec 18:49:36.691 # Next failover delay: I will not start a failover before Mon Dec 29 18:49:40 2014
[14701] 29 Dec 18:49:36.843 # -failover-abort-not-elected master resque 192.168.56.103 6379
[14701] 29 Dec 18:49:36.905 # Next failover delay: I will not start a failover before Mon Dec 29 18:49:40 2014

不会选举新的主服务器，也不会调用重新配置脚本。

据我所知，要选出新的主人，法定人数（n/2+1）必须同意。这就是为什么我使用 3 个哨兵来测试。

我不明白为什么上面的例子没有以选举结束（比如[master]）。

我正在使用 Redis 服务器 v=2.8.19 sha=0a21368c:1 malloc=jemalloc-3.6.0 位=64 build=e570b291804f6e35

感谢您的帮助，请不要介意“救援”一词的严重拼写错误！

- 编辑

[master] redis slave config:

daemonize yes
pidfile "/var/run/redis/redis-server-slave.pid"
port 6380
tcp-backlog 511
bind 192.168.56.102
timeout 0
tcp-keepalive 0
loglevel notice
logfile "/var/log/redis/redis-server.log"
databases 16
save 900 1
save 300 10
save 60 10000
stop-writes-on-bgsave-error yes
rdbcompression yes
rdbchecksum yes
dbfilename "dump.rdb"
dir "/var/lib/redis"
slave-serve-stale-data yes
slave-read-only yes
repl-diskless-sync no
repl-diskless-sync-delay 5
repl-disable-tcp-nodelay no
slave-priority 100
appendonly no
appendfilename "appendonly.aof"
appendfsync everysec
no-appendfsync-on-rewrite no
auto-aof-rewrite-percentage 100
auto-aof-rewrite-min-size 64mb
aof-load-truncated yes
lua-time-limit 5000
slowlog-log-slower-than 10000
slowlog-max-len 128
latency-monitor-threshold 0
notify-keyspace-events ""
hash-max-ziplist-entries 512
hash-max-ziplist-value 64
list-max-ziplist-entries 512
list-max-ziplist-value 64
set-max-intset-entries 512
zset-max-ziplist-entries 128
zset-max-ziplist-value 64
hll-sparse-max-bytes 3000
activerehashing yes
client-output-buffer-limit normal 0 0 0
client-output-buffer-limit slave 256mb 64mb 60
client-output-buffer-limit pubsub 32mb 8mb 60
hz 10
aof-rewrite-incremental-fsync yes
maxclients 4064
# Generated by CONFIG REWRITE

slaveof 192.168.56.102 6379

[master] 主实例 run_id: 5c1ffb7742ad78cde12dbe4858747a314adaebe9

[主]从实例 run_id: 5772f437519bb38782d38f6675ae5d9157be2419

[救援]主实例run_id：5c1ffb7742ad78cde12dbe4858747a314adaebe9

[救援]从属实例run_id：5772f437519bb38782d38f6675ae5d9157be2419

--

[主] Sentinel (:26379) run_id: a664b9f​​61df2b10bbbb5d865b01c599ddd36183c

[主] Sentinel (:26380) run_id: b608fcab7a201799826f4d9ee839aed3cf556fdf

[从机] Sentinel (:26379) run_id: a664b9f​​61df2b10bbbb5d865b01c599ddd36183c

哨兵信息

# Sentinel
sentinel_masters:2
sentinel_tilt:0
sentinel_running_scripts:0
sentinel_scripts_queue_length:0
master0:name=resque,status=ok,address=192.168.56.103:6379,slaves=0,sentinels=3
master1:name=master,status=ok,address=192.168.56.102:6379,slaves=1,sentinels=3

192.168.56.103 的哨兵配置

port 26379
logfile "/tmp/sentinel.log"
dir "/tmp"

sentinel monitor resque 192.168.56.103 6379 2
sentinel down-after-milliseconds resque 3000
sentinel failover-timeout resque 4000
sentinel client-reconfig-script resque /home/sm0ke/Projects/git/thrace/extra/sentinel-failover.py

sentinel config-epoch resque 0
sentinel leader-epoch resque 76
sentinel known-sentinel resque 192.168.56.102 26379 a664b9f61df2b10bbbb5d865b01c599ddd36183c
sentinel known-sentinel resque 192.168.56.102 26380 b608fcab7a201799826f4d9ee839aed3cf556fdf

maxclients 4064
daemonize yes
# Generated by CONFIG REWRITE
sentinel monitor master 192.168.56.102 6379 2
sentinel down-after-milliseconds master 3000
sentinel failover-timeout master 4000
sentinel client-reconfig-script master /home/sm0ke/Projects/git/thrace/extra/sentinel-failover.py
sentinel config-epoch master 73
sentinel leader-epoch master 73
sentinel known-slave master 192.168.56.102 6380
sentinel known-sentinel master 192.168.56.102 26380 b608fcab7a201799826f4d9ee839aed3cf556fdf
sentinel known-sentinel master 192.168.56.102 26379 a664b9f61df2b10bbbb5d865b01c599ddd36183c
sentinel current-epoch 76

192.168.56.103 上 Sentinel 的 redis-cli 信息

# Server
redis_version:2.8.19
redis_git_sha1:0a21368c
redis_git_dirty:1
redis_build_id:e570b291804f6e35
redis_mode:sentinel
os:Linux 3.2.0-4-amd64 x86_64
arch_bits:64
multiplexing_api:epoll
gcc_version:4.7.2
process_id:8345
run_id:491b32b95c547a8266faf9b04ce6b0c18486236b
tcp_port:26379
uptime_in_seconds:8054
uptime_in_days:0
hz:17
lru_clock:10594345
config_file:/etc/redis/sentinel.conf

# Sentinel
sentinel_masters:2
sentinel_tilt:0
sentinel_running_scripts:16
sentinel_scripts_queue_length:4
master0:name=resque,status=ok,address=192.168.56.103:6379,slaves=0,sentinels=3
master1:name=master,status=ok,address=192.168.56.102:6379,slaves=1,sentinels=3

Answer 1

忽略我关于实例 ID 混乱的问题，那是我的错 - 我忘记了哨兵是在哨兵中投票给领导者，而不是在候选人中投票给新的主人，所以三个不同的 ID 是有意义的。

所以，这就是真正的问题：

master0:name=resque,status=ok,address=192.168.56.103:6379,slaves=0,sentinels=3

那里没有哨兵注意到的奴隶，所以他们在投票期间没有一个好的节点可以运行。

检查实例slaveof上的配置192.168.56.103:6380，确保该实例正在运行，然后连接到它以检查它info以验证它是否正在运行。一旦将192.168.56.103:6379其视为从站（在其info命令中），哨兵就会将其视为已知从站并能够故障转移到它。