BigQuery's serverless abstraction hides a distributed compute engine that always consumes slot-time, regardless of pricing model. On-demand billing charges for bytes scanned from columnar storage, not rows returned — meaning a query returning a few hundred rows may scan billions. Flat-rate pricing shifts the concern from per-query cost to slot contention and workload isolation. Slots are scheduler currency, and total_slot_ms is the key metric to monitor. Reservations and assignments let teams enforce SLA boundaries between BI, ETL, and ad-hoc workloads. Batch job priority and flex slots help manage peak demand without over-provisioning. Most BigQuery performance problems are ultimately people and architecture problems, not SQL problems.
11 Impressions