- The architecture uses two Cloud Run services, a partitioned BigQuery store, and a transparent JSON cache to keep costs low.
- The collector operates as a small script embedded via an iframe, probing browser capabilities and submitting data as GET requests to avoid CORS preflight.
- BigQuery stores data in day-partitioned tables with 120-day expiration, using the Storage Write API to cut costs.
- The reporting service runs many small aggregation queries in JavaScript, assembled client-side, with a transparent cache layer in Cloud Storage.
- The cache serves stale data with background revalidation, reducing BigQuery queries to about 15 refreshes per active aggregate per month.
- Total cost is around $3 per month, with Cloud Run being the main expense at $2.30.
- The author would improve schema management and invalidation, but the low cost design is effective.