apparently it is required to call .compute() after calling to_textfiles. I felt this behavior is weird comparing it to spark for example. It would be nice to not need to call .compute() after calling to_textfiles.