kubernetes中的redis-ha无法故障转移回master

Question

我试图创建一个简单的redis高可用性设置与1主，1从和2哨兵。

当从redis-master故障转移到redis-slave时，该设置工作正常。当redis-slave恢复时，它将自己正确地注册为新的redis-slave主服务器的从服务器。

但是，当redis-slave作为主服务器关闭时，redis-master 不能作为主服务器返回。redis-master的日志进入循环，显示：

1:S 12 Dec 11:12:35.073 * MASTER <-> SLAVE sync started
1:S 12 Dec 11:12:35.073 * Non blocking connect for SYNC fired the event.
1:S 12 Dec 11:12:35.074 * Master replied to PING, replication can continue...
1:S 12 Dec 11:12:35.075 * Trying a partial resynchronization (request 684581a36d134a6d50f1cea32820004a5ccf3b2d:285273).
1:S 12 Dec 11:12:35.076 * Master is currently unable to PSYNC but should be in the future: -NOMASTERLINK Can't SYNC while not connected with my master
1:S 12 Dec 11:12:36.081 * Connecting to MASTER 10.102.1.92:6379
1:S 12 Dec 11:12:36.081 * MASTER <-> SLAVE sync started
1:S 12 Dec 11:12:36.082 * Non blocking connect for SYNC fired the event.
1:S 12 Dec 11:12:36.082 * Master replied to PING, replication can continue...
1:S 12 Dec 11:12:36.083 * Trying a partial resynchronization (request 684581a36d134a6d50f1cea32820004a5ccf3b2d:285273).
1:S 12 Dec 11:12:36.084 * Master is currently unable to PSYNC but should be in the future: -NOMASTERLINK Can't SYNC while not connected with my master
1:S 12 Dec 11:12:37.087 * Connecting to MASTER 10.102.1.92:6379
1:S 12 Dec 11:12:37.088 * MASTER <-> SLAVE sync started
...

每个复制文档都声明：

自Redis4.0以来，当一个实例在故障转移后被提升为master时，它仍然能够与旧master的从机执行部分重新同步。

但日志似乎显示了另一种情况。更详细的日志版本显示了第一个redis-master到redis-slave故障转移和随后的redis-slave到redis-master日志。

你知道怎么回事吗？如何才能允许redis-master返回master角色？配置详细信息如下：

NAME             TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)     AGE
redis-master     ClusterIP   10.102.1.92     <none>        6379/TCP    11m
redis-slave      ClusterIP   10.107.0.73     <none>        6379/TCP    11m
redis-sentinel   ClusterIP   10.110.128.95   <none>        26379/TCP   11m

requirepass test1234
masterauth test1234
dir /data

tcp-keepalive 60
maxmemory-policy noeviction
appendonly no
bind 0.0.0.0
save 900 1
save 300 10
save 60 10000

slave-announce-ip redis-master.fp8-cache
slave-announce-port 6379

requirepass test1234
slaveof redis-master.fp8-cache 6379
masterauth test1234
dir /data

tcp-keepalive 60
maxmemory-policy noeviction
appendonly no
bind 0.0.0.0
save 900 1
save 300 10
save 60 10000

slave-announce-ip redis-slave.fp8-cache
slave-announce-port 6379

Answer 1

原来这个问题与使用主机名而不是IP有关：

slaveof redis-master.fp8-cache 6379
...
slave-announce-ip redis-slave.fp8-cache

因此，当master作为slave返回时，sentinel显示现在有2个slave：一个带有ip地址，另一个带有主机名。不确定这两个从条目（指向同一个Redis服务器）是如何导致上面的问题的。现在，我更改了配置，使用IP地址而不是主机名，Redis HA的工作状态完美无缺。