A financial firm receives thousands of JSON market-data records per second and needs a scalable AWS solution with minimal operational overhead to detect anomalous data points in real time. Which design meets these requirements?

The correct option leverages Kinesis Data Streams for scalable real-time data ingestion and Amazon MSK for Apache Flink with the built-in RANDOMCUTFOREST function. This combination provides a fully managed, real-time anomaly detection solution with minimal operational overhead, as Flink handles stream processing and the RCF algorithm is optimized for this use case. The second option, using SageMaker endpoints with Lambda, introduces more operational overhead due to managing SageMaker endpoints and Lambda invocations for each record, which can be less efficient for high-throughput real-time stream processing compared to Flink. The third option, running Apache Kafka on EC2, significantly increases operational overhead as it requires managing EC2 instances, Kafka clusters, and their scaling. The fourth option, using SQS FIFO and AWS Glue ETL, is designed for batch processing, not real-time anomaly detection, and would introduce significant latency.