If the worker dies, we need to push all worker services to it again when it comes back up. Ideally we'd also kill all known existing processes.