Currently duplicated data are detected using `jubaql_timestamp` but it is better if other keys can be used.