The platform is split into API, worker, connectors, callback gateway, storage, Kafka, and observability. That separation is useful in production, but the docs should tell operators how to bring it up and how to shut it down safely.
docker compose up -d docker compose ps docker compose logs -f api worker callback-gateway
Use the local stack to validate request flow, admin APIs, template resolution, and callback handling before touching a public environment.
Store provider credentials as secret references or mounted secret files, not inside the request payload or docs.
Keep the admin token in your deployment secret store and inject it as an environment variable or secret mount.
Callback routes can use no verification, shared-secret verification, or HMAC verification depending on the provider.
FCM service accounts should be mounted as files or secret references so the connector can resolve them at runtime.
Run API, worker, connectors, and callback gateway in the application namespace.
Run Prometheus, Grafana, and dashboard components separately so app rollouts do not disturb observability.
Run Kafka, Postgres, and node exporters where you prefer to keep scrape targets isolated from application pods.
Expose only the public endpoints you actually need. Admin and callback routes should not be reachable without intent.
Create a fresh EKS cluster in the chosen region, then deploy the same namespace split you used locally.
Publish images to a registry, then point the manifests or Helm values at that registry.
Run notification smoke tests after deployment so you can verify send, callback, and metrics paths together.
Keep the previous image and manifest set available so you can roll back without rebuilding the environment.
When you need to stop the system, scale the workloads down in reverse order and remove the external resources only after the data has been backed up.
docker compose down kubectl scale deploy/api --replicas=0 kubectl scale deploy/worker --replicas=0 kubectl scale deploy/callback-gateway --replicas=0
