Amazon EC2 Spot Instances

Correct Answer: Run the workers in an Auto Scaling group using Spot Instances with a mixed instances policy (capacity-optimized allocation and Capacity Rebalancing enabled). Handle Spot interruption and rebalance signals with lifecycle hooks to gracefully drain, checkpoint, or re-queue in-flight work to a durable queue (such as Amazon SQS), and store any state externally (for example, Amazon S3 or a database) so replacement instances can resume processing.

Spot Instances can be interrupted, so applications must be designed to tolerate loss of individual instances without losing work. The most reliable, low-cost pattern is to run stateless workers in an Auto Scaling group configured to use Spot capacity with a mixed instances policy, capacity-optimized allocation, and Capacity Rebalancing. This combination reduces the chance of interruption and proactively launches replacement capacity when a rebalance recommendation is issued. Handling Spot interruption notices and rebalance recommendations via Auto Scaling lifecycle hooks or instance scripts allows each instance to gracefully drain, checkpoint, or re-queue in-flight tasks to a durable queue like Amazon SQS. Persisting state in external storage (e.g., S3 or a database) ensures a newly launched instance can continue processing where the interrupted one left off.

Correct Answer: Create a managed AWS Batch environment that launches Spot capacity across multiple instance types using the SPOT_CAPACITY_OPTIMIZED allocation strategy, and add an On-Demand compute environment as a lower-priority fallback so jobs run on Spot by default but spill over to On-Demand if Spot capacity is insufficient to meet the deadline.

Using multiple instance types with the SPOT_CAPACITY_OPTIMIZED allocation strategy diversifies across the most available Spot pools, reducing interruption risk and improving throughput so jobs finish faster at low cost. Adding a lower-priority On-Demand compute environment in the job queue ensures that if Spot capacity is unavailable or interruptions slow progress, remaining tasks can run On-Demand to meet the deadline. This approach minimizes cost by favoring Spot while guaranteeing completion on time. Increasing a Spot "bid" is not required; AWS charges the Spot market price regardless of the maximum price you set.

Correct Answer: Amazon EBS volumes with data checkpointing

Amazon EBS volumes with data checkpointing allow you to persist data even if an instance is terminated. Data checkpointing helps save the state of your data at specific points in time, ensuring that you can resume your work after an instance is terminated.