Yup found it. So to run individual nodes, all input/output datasets must be written to disk right? In memory datasets cannot work in this case.