Google BigQuery

When to Use

• Serverless Analytics: No infrastructure to manage. Just run SQL.
• Real-time Analytics: High-speed streaming ingestion.
• ML Integration:

  CREATE MODEL

  lets you train ML models using standard SQL (BigQuery ML).

Quick Start

-- Standard SQL
SELECT name, COUNT(*) as count
FROM `bigquery-public-data.usa_names.usa_1910_2013`
GROUP BY name
ORDER BY count DESC
LIMIT 10;

Core Concepts

Slots and Reservations

Columnar Storage (Capacitor)

SELECT *

Partitioning & Clustering

• Partitioning: Splits table by Date/Int (e.g., Daily partitions). Prunes data scanning massive cost savings.
• Clustering: Sorts data within partitions for faster filtering.

Best Practices (2025)

• Partition by Date: Almost mandatory for time-series logs.
• Use BigQuery ML: Train models (Regression, K-Means) directly where data lives.
• Estimate Cost:

  Dry Run

  your query to see how many bytes it will scan before running it.

• Don't run

  SELECT *

  : You pay per column read. Select only what you need.
• Don't treat it like an OLTP: Single row inserts are slow (unless using Streaming API). It is for bulk analytics.

References

• BigQuery Documentation