Why is running ray on spark restricted to local / databricks clusters

Ray now supports running ray on spark. The documentation for this can be found here :


Does anyone know why the support for this is limited to standalone and databricks clusters ? Is there any reason why this would not work on cloudera / yarn ?