Usually they eventually go away, though that does take a while thru the cluster reconciliation process. The other way I’ve done that is by restarting the etcd/controlplane nodes.