A logistics firm scores about 100 million shipment rows that live in a warehouse table. A TensorFlow model needs roughly 45 minutes to run on that volume, and the output must be ready before the morning reports. Which option keeps spend lowest while finishing on time?
Show answer and explanation
Answer: B
Managed batch prediction is built for large offline scoring. It reads the source table, provisions workers for the job, and tears them down afterward, so you pay only for the run. An always-on endpoint sits idle between batches and makes one request per row, which is slow and costly at this volume. Moving data through a separate pipeline adds work without benefit when the source is already a table the batch job can read.
Why the other options are wrong
- A. A continuously running endpoint bills for idle hours, and handling 100 million single-row calls is slow and expensive compared with one bulk job, so it fails the cost goal.
- C. Exporting data and running a streaming pipeline adds data movement and always-on workers for a job that runs once a day. It is more complex and not cheaper than reading the table in place.
- D. A TensorFlow network generally cannot be turned into warehouse SQL functions, so this option is not feasible. The model would have to be re-implemented and retrained, which the scenario does not allow.