Migrate a monthly, 30-hour offline batch job that must be restartable if interrupted and minimize cost. What should you do?

For a monthly, 30-hour offline batch job that needs to be restartable and cost-effective, a standard Compute Engine VM that is started and stopped as needed is the most suitable option. This approach ensures the job can run reliably for its duration and be restarted without data loss if interrupted, as standard VMs are not preempted. Stopping the VM when not in use minimizes costs, as you only pay for compute resources while the instance is running. Spot VMs (and Spot nodes in GKE) are significantly cheaper but can be preempted at any time, making them unsuitable for a job that must be restartable and complete within a specific window without guaranteed uptime. While an instance template with Spot VMs and a managed instance group offers some resilience, the core issue of preemption for a critical, restartable job remains.