An application calls DynamoDB using the low-level BatchGetItem operation and often receives responses that include entries in UnprocessedKeys. Which actions will make the application more resilient when UnprocessedKeys are returned? (Choose two.)

When BatchGetItem returns UnprocessedKeys, it indicates that DynamoDB could not process all requested items in the batch, often due to throttling or transient issues. Retrying the batch operation using exponential backoff with a randomized delay is crucial for resilience. This strategy prevents overwhelming DynamoDB with immediate retries, allowing the service to recover, and the randomized delay helps distribute retries over time, avoiding a "thundering herd" problem. Increasing the provisioned read capacity of the affected DynamoDB tables directly addresses the root cause of throttling if the read capacity is insufficient for the workload. This ensures DynamoDB has enough resources to handle the read requests. Modifying the application to use an AWS SDK is generally good practice but doesn't directly solve the UnprocessedKeys issue if the underlying capacity or retry logic is flawed. Increasing write capacity is irrelevant for BatchGetItem, which is a read operation.