Otherwise we pick up previous cluster instances (in EC2, stopped instances hang around for a while - maybe 30 minutes?)