The KeysCollectionManager checks for refresh failure, but only handles it by doing backoff for the retries. This means that admins are not aware that refreshing has failed, which is important because persistent refresh failures can prevent the cluster from operating correctly. These failures should be logged at the default level (because they should be rare/infrequent, and even though the first retry is after 200ms, this will increase until they are done every 10 mins).