A data migration must run to successful completion once, with failed Pods retried under controller policy. Which workload resource best expresses that requirement?

1. Deployment
2. Service
3. DaemonSet
4. Job