What will trig lettuce topology refresh strategy: PERSISTENT_RECONNECTS

[primabrucexu]

I'm trying to figure out how lettuce adaptive refresh topology work.
In this scenario, I shutdown one master in redis cluster and the cluster poll up another slave as new master. But I found still some of commands were sent to the old master. With debug enable, I noticed that lettuce didn't mark the offline node as FAIL in partitions flags. (It will be marked when I enable periodic refresh). I also tried network partition and completely offline lettuce from cluster but commands still sent to the offline node
I changed some source code to make sure all commands can be sent to the offline node. Does it matter?
So I want to know what scenario can trig this strategy.
BTW, I found a similar issus says it's a bug, #1857