If you do need to process the records in each file, it's a little more complicated. You would need to use an Expression Evaluator to set the roll header attribute when record:id() ends in ::0, then configure the HDFS destination to use the roll attribute.