A legacy batch processor is being moved to managed compute. Tasks have dependencies, varying resource needs, and can run for hours. Which service should be evaluated rather than forcing every task into a short Lambda execution?

1. S3 object tags as the only compute execution engine.
2. AWS Batch with appropriate compute environments, job definitions, and retry policies.
3. Route 53 weighted records as a dependency-aware job scheduler.
4. One Lambda invocation for every task regardless of documented runtime limits.