I think the root of my problem is that the Docker driver is not acknowledging the interrupt given by Nomad when a previous version of a job should be killed.
