What’s the best way to build analytics for up to one million event records?

Make the rollup process idempotent so events can be replayed or corrected safely. You’ll also want a strategy for late-arriving data, such as reopening recent time buckets and recomputing them. Redis can still be useful as a short-lived cache for especially expensive dashboard responses, but it probably shouldn’t hold the whole event history.

A million rows is not especially large, so you probably don’t need an exotic architecture. A conventional SQL database with good indexes and a few materialized views for common dashboard queries would likely be simpler and more flexible than using Firestore as the analytics layer. Keep the raw events available for auditing, and let the database handle grouping and aggregation.

I wouldn’t use Redis as the primary analytical store for hundreds of megabytes of raw events that you recalculate repeatedly. Keep the events in your durable database, then create incremental rollups for the metrics users actually need. For example, upsert hourly buckets grouped by time period, bus, driver, and alert type. The dashboard can read those much smaller tables and only query raw events when someone drills into details.

If the time-based queries become the main challenge, a time-series option such as TimescaleDB is worth considering. It can retain the raw events for audits while supporting time buckets, retention policies, and continuous or incremental aggregates. That gives you analytical queries without loading the entire dataset into application memory.