All stacks · Warehouses & analytics
Serverless analytics on Google Cloud.
BigQueryBigQuery charges by bytes scanned. Learn how partitioning and clustering cut scan costs by 90%+, and the query patterns that silently bypass them.
BigQueryWhat BigQuery reads before it reads your data: partition metadata, Capacitor block statistics, and the specific query shapes that silently turn a pruned scan into a full one.
BigQueryHow I take a runaway BigQuery invoice apart: find the top jobs, fix pruning, cap bytes billed, use materialized views honestly, and decide when Editions beats on-demand.
ToolsThe pricing models decide the architecture: warehouse-hours versus on-demand scans, plus multi-cloud vs GCP-native gravity.
AI & GenAINative vector search in Snowflake, BigQuery, and pgvector has quietly become good enough for most RAG workloads. Here are the thresholds, the real costs, and when a dedicated store still wins.
SQLWhat actually happens to a CTE in Postgres, Snowflake, and BigQuery, when it blocks predicate pushdown, and when referencing it twice costs you two scans.
BigQuery is the most serverless of the major warehouses: no clusters to size, no warehouses to resume, just SQL against a scanning engine that charges by bytes read or by reserved slots. That model makes the first query effortless and makes cost discipline a schema-design problem.
The tutorials here focus on the BigQuery-specific craft: partitioning and clustering that cut scans by orders of magnitude, on-demand versus capacity pricing decisions, scheduled queries and Dataform, and patterns for streaming inserts and external Iceberg tables.
Set maximum_bytes_billed on queries or as a project default — the query fails fast instead of scanning 40 TB. Partition large tables by date, cluster by common filters, and never SELECT * on wide tables. Preview and dry-run both show estimated bytes before you run.
On-demand ($ per TB scanned) is right until your monthly scan spend is stable and high enough that reserved slots are cheaper — for many teams that crossover sits in the low thousands of dollars per month. Editions autoscaling makes reservations far less rigid than they used to be.
Partition by the column that bounds most queries (almost always event date), then cluster by the next most common filters like customer_id. Partitioning enables hard pruning and expiration; clustering sorts within partitions for finer-grained skipping.
It is tied to GCP for hosting, but Omni queries data in S3 and Azure, external tables and BigLake read open formats, and the Storage Read API feeds external engines. If your organization is multi-cloud, the coupling is looser than it once was.
That query cost depends on columns scanned, not rows returned. A LIMIT 10 on an unpartitioned table still scans everything the query references. Storage is cheap; careless scanning is what shows up on the invoice.
One email a week with our latest tutorials. No spam.
Newsletter signup is not live yet. Use the contact form if you want to be notified.