Redis "slaveof no one" command can't work exactly when this slave instance has been auto-discovered by sentinel?

There is a master instance (A) with a slave (B), and the master instance is monitored by a sentinel. I connected to the slave instance by redis-cli and sent the command "slaveof no one" to it, it became a master instance, but in a short while it would become a slave of A instance again. Is this the expected behavior? If so, how can I let B instance no longer be a slave?
Redis version: 2.8.13

Comment From: therealbill

Yes, this is expected behavior. Sentinel is meant to manage the instances involved in the "pod"[1] and will attempt to ensure things are "as they should be".

There are two ways to handle this scenario: 1) Remove and re-add the master 2) Issue "sentinel reset " on the sentinels to reset their configuration. At that point sentinel will rediscover the slaves. Does that help?

That said a way to remove slaves from a configuration is needed.

Cheers, Bill 1. Cluster isn't quite right for the many replication scenarios for Redis and conflicts with "Redis Cluster" so I just use the term "pod" to represent them.