A company periodically retrains ML models using new streaming device data. They need a high-throughput, durable, scalable ingestion system that can tolerate up to 5 minutes of ingestion latency and apply basic transformations during ingestion. Which approach provides the most operational efficiency?

This option is the most operationally efficient because it leverages Kinesis Data Streams for high-throughput, durable ingestion and Kinesis Data Firehose for scalable delivery to S3 with built-in transformation capabilities via Lambda. Kinesis Data Streams handles the initial ingestion from devices, and Firehose automatically scales to consume from the stream and deliver to S3, minimizing operational overhead. The 5-minute latency tolerance aligns well with Firehose's buffering capabilities. Incorrect options: Writing directly to S3 with Lambda for transformation and then to Kinesis Data Streams and Firehose introduces unnecessary complexity and latency, as data is moved between services multiple times. Using AWS Glue jobs triggered by S3 events for streaming data is less efficient for real-time or near real-time ingestion and transformation compared to Kinesis services, which are designed for streaming workloads. Glue jobs are better suited for batch processing of larger datasets. Connecting an AWS Glue job directly to a Kinesis Data Firehose delivery stream for transformation is not a standard or efficient pattern. Firehose's primary function is delivery, and its built-in Lambda transformation is more suitable for this use case.