Warning message repeated in Wildfly after scale-down using KUBE_PING
ActiveMQ Artemis only uses JGroups (or any other discovery mechanism) to discover other brokers for the purpose of clustering them together. Once another broker is discovered then they establish TCP connections between themselves after which JGroups doesn't serve any role which means that JGroups seeing the broker leave the cluster is irrelevant.
The fact that the cluster bridge fails is enough to tell ActiveMQ Artemis that the broker has left the cluster. The question at that point is what should the broker do in response to the dead node. By default it will attempt to reconnect indefinitely as it expects the node to come back at some point. This is a reasonable expectation in a traditional use-case but not so much in the cloud. This behavior is controlled by the reconnect-attempts property on the cluster-connection. Set reconnect-attempts to something you think is reasonable (e.g. 10) and you'll see the bridge reconnect give up and stop logging.