Currently KafkaRoller evaluates each node based on the role defined in NodeRef. However, both KRaft and Zookeeper broker would have broker role defined in its NodeRef, therefore KafkaRoller cannot tell which mode it is running in. KafkaRoller needs to identify the current active controller and depending on which mode the cluster is in, it needs to make a different API call (e.g. describeCluster against broker nodes or describeQuorum against controller nodes). Since it cannot tell which mode the cluster is running on, when running in KRaft mode, it makes both API calls and results in inconsistent and duplicated logs reporting the active controller. In KRaft, describeCluster returns a random broker and will mean this broker is deferred.

Created by Strimzi#9373.