Sign In
Platform Note 004

Event Projections Architecture: CQRS with Redpanda

The Command/Projection/Query separation using Firebase Callable Functions, Postgres Outbox, and Rust normalizer worker for real-time Firestore projections.

  • Architecture
  • Redpanda
  • Firestore
  • Realtime

What shipped

  • Commands: ingestEvent callable function publishes to app-events topic
  • Reliable Delivery: Transactional Outbox in Django + outbox_relay for publishing
  • Projections: telemetry_worker consumes events and materializes into Firestore
  • Queries: Direct Firestore subscriptions for real-time UI updates

Technical flow

  1. Client submits event → Firebase callable (version, idempotency_key)
  2. Django writes OutboxEvent atomically with main transaction
  3. Outbox relay publishes to Redpanda with Kafka compatibility
  4. Rust normalizer (or Python worker) consumes with idempotency
  5. Projections written to deml Firestore database
  6. UI subscribes via Firestore SDK for real-time updates

Why it mattered

This architecture enables horizontal scaling without state coordination. The platform showcase (Tenant0) processes identically to customer tenants through the same code path. Versioned events provide governance for schema evolution.

Living release record

Blue Notes summarize shipped outcomes. The public repository remains the detailed engineering record.

View source