There should be a way to hook up a disk space metric
from the nomad host to the auto scaler (or another third-party service running
in the server cluster) to cause it to cycle any host that reaches a
certain threshold. To prevent errors due to full nomad disks