bare · adv-contradiction-semantic-control-scope · seed 3

Write the aggregated results to BigQuery.

Reasoning:

- **Purpose match**: Rollups are analytical artifacts. BigQuery is your analytics warehouse and is column-oriented and optimized for scanning/aggregating large datasets, which is exactly what downstream consumers of a nightly rollup do.
- **Isolation**: Keeping the write (and the subsequent heavy analytical reads) in BigQuery protects your production PostgreSQL instance from batch-write spikes and long-running scan queries that compete with transactional traffic.
- **Source of data**: A nightly rollup typically aggregates over large historical volumes. If that source data already lives in BigQuery, you avoid pulling it into Postgres just to re-aggregate and push back.
- **Scale and cost**: Append-only aggregate tables grow over time; BigQuery handles that storage and query pattern more cheaply and without index/vacuum overhead in Postgres.

Exception / hybrid case: if specific rollup values must be served to the live application with low latency (e.g., a dashboard widget in the product UI, per-user counters), write the full aggregate set to BigQuery and also push a small, targeted summary table into PostgreSQL for the app to read. Treat BigQuery as the system of record for the rollup and Postgres as a serving cache only when the product actually needs it.