Redis READONLY You can't write against a read only replica:故障切换后老连接不会断,降级前写进旧主的 100 个 key 同步后全丢,replica-read-only no 不是修法

  从库角色:slave,replica-read-only = yes
  > SET x 1                                                      (error) READONLY You can't write against a read only replica.
  > INCR cnt                                                     (error) READONLY You can't write against a read only replica.
  > DEL k                                                        (error) READONLY You can't write against a read only replica.
  > EXPIRE k 10                                                  (error) READONLY You can't write against a read only replica.
  > LPUSH l b                                                    (error) READONLY You can't write against a read only replica.
  > FLUSHALL                                                     (error) READONLY You can't write against a read only replica.
  > GET k                                                        "v"
  > LRANGE l 0 -1                                                1) "a"
  > EVAL return redis.call('SET','a','1') 0                      (error) READONLY You can't write against a read only replica. script: 88c923858e73954cede0
  > EVAL return redis.call('GET','k') 0                          "v"
  > EVAL_RO return redis.call('GET','k') 1 k                     "v"
  > PUBLISH ch hi                                                (integer) 0
  > CONFIG SET maxmemory 100mb                                   "OK"
  > BGSAVE                                                       "Background saving started"
  作为对照,同样的写命令发给主库 > SET x 1 -> "OK"

  > REPLICAOF NO ONE
      "OK"
  提升后新主角色:master;旧主这时仍然是:master
  降级之前:客户端仍在往旧主写,100 次写入成功 100 次(这就是脑裂窗口:新主上没有这些 key)
  新主上有多少个 split* key:0
  > REPLICAOF 127.0.0.1 8171
      "OK"
  旧主降级后角色:slave,master_link_status = up
  > SET after 1
      (error) READONLY You can't write against a read only replica.
  > INCR counter
      (error) READONLY You can't write against a read only replica.
  同一条连接是否还活着:PING -> "PONG"(没有被断开,只是写被拒)
  旧主上降级前写入的 100 个 split* key 现在还剩:0 个(同步后以新主的数据为准,它们丢了)
  新主从旧主那里收到的 key:before 存在 = 1

  > CONFIG SET replica-read-only no (在旧主/现从库 8170 上) -> "OK"
  > SET local_only written-on-replica
      "OK"
  现在 8170 上 GET local_only = "written-on-replica";新主 8171 上 = (nil)
  > REPLICAOF NO ONE
      "OK"
  > REPLICAOF 127.0.0.1 8171
      "OK"
  重新全量同步后 8170 上 GET local_only = (nil)

  从库 127.0.0.1:8193(主库 8190)
  > GET key2
      (error) MOVED 4998 127.0.0.1:8190
  > READONLY
      "OK"
  > GET key2
      "v"
  > SET key2 w
      (error) MOVED 4998 127.0.0.1:8190
  > READWRITE
      "OK"
  > GET key2
      (error) MOVED 4998 127.0.0.1:8190