Deployment

The platform is containerized and deployed with Kamal, Rails’ own zero-downtime deploy tool. The goal is a production setup one engineer can run and reason about — no Kubernetes control plane to babysit.

Containers

Everything is defined in Docker:

• Dockerfile — the production image: a slim, multi-stage build with assets precompiled and Thruster in front for asset compression and HTTP caching.
• Dockerfile.dev + docker-compose.yml — the local stack: the app, PostgreSQL, and Redis, so a new developer is one docker compose up from a running store.

The running app is several processes from the same image: the Puma web server, a Solid Queue worker, and a Sidekiq worker — see Background Jobs.

Deploying with Kamal

Kamal builds the image, pushes it to a registry, and rolls it out across hosts with a health-checked, zero-downtime swap. Production configuration lives in config/deploy.yml.

Redis runs as a Kamal accessory (ecommerce_app-redis) on the shared kamal network. Operational note: rebooting it —

export KAMAL_REGISTRY_PASSWORD=...   # required first
kamal accessory reboot redis

— needs the registry password exported, because .kamal/secrets references it. And because Redis is a fail-open dependency, that reboot degrades the site rather than downing it.

Health checks

Two endpoints serve two audiences:

• /up — a pure boot check; “is the process alive?” Used by the orchestrator at start-up.
• /health — a deeper check reporting database, Redis, and queue status. It deliberately returns HTTP 200 even when Redis is degraded, so the load balancer never pulls a node that is still serving shoppers. See Redis Resilience for why.

Observability

• Prometheus — metrics exported from the app, including queue depth and request timings.
• Sentry — exception tracking, wired so production errors (including reported Redis failures) surface immediately.
• Mission Control — Jobs — a dashboard over both job backends.

Configuration

Configuration is environment-driven. env.example and .env.production.example document every variable — database URLs, Redis pool sizing, SES credentials, payment-gateway keys, and the RACK_ATTACK_* rate-limit knobs — so a new environment is a matter of filling in values, not reading source.

Key files