maintenance with a disk swap (reboot of one node in the scratch cluster)
files may be unavailable for a little moment.