An Amazon EMR cluster (using EMRFS) runs critical batch work every day starting at 1:00 AM and takes about 6 hours. The cluster currently uses On-Demand instances for primary, core, and task nodes, and completion time is not urgent. The architecture should be reviewed to reduce compute costs. Which change should the solutions architect recommend?

The correct option balances cost savings with reliability for critical workloads. Primary and core nodes are essential for EMR cluster stability and data persistence (HDFS), so using On-Demand Instances ensures uninterrupted operation. Task nodes are more fault-tolerant and can be interrupted without losing significant progress, making Spot Instances ideal for cost optimization. Terminating only task nodes when processing completes allows the core cluster to remain available for potential future use or debugging, while still saving costs. Purchasing Compute Savings Plans further reduces the cost of the predictable On-Demand usage. Running all nodes on Spot Instances (first option) is too risky for critical workloads due to potential interruptions. Keeping all nodes on On-Demand (third option) misses significant cost-saving opportunities. Terminating the entire cluster (second option) is less efficient than only terminating task nodes if the core cluster could be reused.