GCP Professional Cloud Architect Practice Question 8: BigQuery Capacity Cost Optimization
Google Cloud Professional Cloud Architect · Test 1 · question 8 of 10
Finance reports that BigQuery spend at a media analytics firm swings by a factor of four from month to month and is impossible to forecast. Investigation shows a steady, always-on set of hourly ELT jobs that consume a fairly constant amount of compute, plus an ad-hoc analyst workload that spikes hard around monthly reporting. Two goals come out of the review: the bill must become predictable and cheaper for the steady portion, and analyst spikes must stop slowing the production ELT jobs.
What should the architect do?
Show answer and explanation
Answer:
This is the workload-management design BigQuery capacity pricing exists for, and it answers both goals at once. Slot commitments discount the steady-state portion and turn it into a known monthly number, which is what makes the bill predictable. Separate reservations are the isolation mechanism: assigning the ELT projects to one reservation and the analyst projects to another means a reporting-week spike consumes the analysts' slots rather than competing with production. Setting the commitment at the baseline and letting reservations autoscale above it keeps the spikes served without paying peak prices year round.
Option A leaves the fundamental problem untouched. On-demand billing is per byte scanned, so cost still tracks unpredictable query behaviour, and custom quotas control spend only by making queries fail once the cap is hit, a poor experience that still provides no workload isolation.
Option B fixes the bill but at the wrong number. Sizing a single reservation to the combined peak means paying for the reporting-week capacity every day of the month, and putting all workloads in one reservation means the analysts and the ELT jobs still share the same slot pool, the contention the review asked to remove.
Option C is genuinely good engineering and will reduce bytes scanned, but it is orthogonal to the question. Query tuning does not create a predictable bill under on-demand pricing and does nothing to isolate the two workloads from each other; it is a complement to option D, not a replacement.
Reference: Introduction to BigQuery reservations