An CPD RPC may contain multiple sub-operations, including 0 ~ N sub-modification. Simply checking the RPC opc (DAOS_OBJ_RPC_CPD) cannot detect whether it contains modifications or not. As so miss modification related check and caused some IO trouble. For example, one of DAOS users reported:

       The problem I encountered :   At the beginning of reintegrate, the modify options with ORF_REINTEGRATING_IO should return DER_UPDATE_AGAIN and retry because pool_discard not complete. But the IO from ds_obj_cpd_handler ignore this check and write the data on nvme success, and then was delete by obj_discard soon。