A Dataflow streaming job calls an external enrichment API for every element. Worker CPU stays near 15 percent while the job falls further behind its input. Which change most directly addresses the cause?
Show answer and explanation
Answer: D
Low CPU combined with a growing lag means the workers are idle much of the time, waiting on a slow or quota-limited external dependency. Batching requests reduces round trips, and tuning concurrency to the quota keeps calls flowing without triggering throttling. Bigger disks, a single large machine, or dropping the enrichment do not address that waiting time, and removing enrichment would break the required output.
Why the other options are wrong
- A. Persistent disk size affects storage capacity, not the time workers spend waiting on network calls, so the lag would not shrink.
- B. A single large machine removes the horizontal parallelism Dataflow relies on and still leaves the workers waiting on the external service.
- C. Dropping the enrichment changes the output the business needs and only moves the wait to a later job, so it is not a valid fix.