Deployment

Production target architecture for Race Platform. The same source tree that runs in docker-compose locally swaps in managed cloud services through driver environment variables.

Cloud target

Layer	Local	Production
DB	Postgres 16 (docker)	Supabase Postgres + Auth + Realtime
Object storage	MinIO (docker)	Cloudflare R2
KV	Redis (docker)	Cloudflare KV
API	Hono on Node (docker)	Cloudflare Workers
Realtime	ws on Node (docker)	Cloudflare Durable Objects
Queue	(none)	Cloudflare Queues (webhook retries, KPI jobs)
Strategy engine	Python (docker)	Fly.io
Search / AI memory	(none yet)	Cloudflare Vectorize
Edge ingester	Rust binary	Rust binary on pit-box laptop

Everything is provisioned via OpenTofu — see infra/opentofu/ for the modules and .tfvars shape.

Environment-driven driver selection

Adapters live in apps/<svc>/src/adapters/, interfaces in apps/<svc>/src/ports/. Selection happens at runtime via these env vars:

Env var	Choices	Default
STORAGE_DRIVER	minio / r2	minio
KV_DRIVER	redis / cf-kv	redis
DB_DRIVER	postgres / supabase	postgres
REALTIME_DRIVER	ws / cf-do	ws

There is no cloud-provider name in business logic — adding AWS as a future driver target is one new adapter, not a refactor.

Cost target

• $0 during local development
• ~$40/month through alpha pilots (Supabase Free + Cloudflare Workers Free tier + R2's zero-egress pricing)
• Scales linearly with pilot teams added

What you can do today

• Run the full stack locally with the docker-compose path
• Run the OpenTofu plan against a Cloudflare + Supabase project in dry-run mode (tofu plan)
• Build the API for the Workers target (pnpm build)

What's coming

• First-pilot production rollout — wires up Supabase + R2 + Workers end to end
• CI deploy pipeline — GitHub Actions → Cloudflare Workers on every main merge
• Per-environment Secrets management — Doppler / Cloudflare Secrets
• Database backups — automated nightly Supabase snapshots + S3 dump