For scott's recent workflows, it seems the jobs are failing because of incomplete kickstart output. these stage-in jobs have generate_checksum set to true, and are transferring about 4K files per transfer job.
the job itself runs on shock