Atlas Telemetry

The existing stack wrote every GPS ping straight into the primary database. Dashboards timed out, and the team's answer had been to keep adding read replicas.

I introduced a partitioned time-series store with rollups at one-minute and one-hour grain, fronted by a Next.js dashboard that queries the rollup appropriate to the zoom level. Raw pings now expire on a retention policy instead of accumulating forever.

Operations went from a map that took eleven seconds to paint to one that updates live, on the same hardware budget.