Plane	Services	How
Control	backend, consumer, tracking, realtime, web	Container hosting in one region (Railway in production). Stable region-pinning so KMS/S3 calls stay local.
Execution	worker	One process per VPS, anywhere with a public IPv4. Managed from the admin dashboard over SSH.

Directory layout

deploy/
├── docker/
│   ├── backend.Dockerfile
│   ├── consumer.Dockerfile
│   ├── worker.Dockerfile
│   └── realtime.Dockerfile
└── config/
    └── env.example

The tracking Dockerfile lives at tracking/Dockerfile. The local-dev compose is docker-compose.yml at the repo root.

Building images

docker build -f deploy/docker/backend.Dockerfile  -t warmbly/backend  .
docker build -f deploy/docker/consumer.Dockerfile -t warmbly/consumer .
docker build -f deploy/docker/worker.Dockerfile   -t warmbly/worker   .
docker build -f deploy/docker/realtime.Dockerfile -t warmbly/realtime .
docker build -f tracking/Dockerfile               -t warmbly/tracking tracking/

GitHub Actions publishes these to GHCR automatically. See the deployment guide.

Local development

make infra  # postgres, redis, kafka, etc. (leave running, shared across worktrees)
make app    # backend, consumer, worker, tracking, realtime, web (hot reload)
make sim    # adds premium + dedicated workers (prod-image flow)
make seed   # rich fixture
make tools  # kafka-ui at :18090
make reset  # nuke volumes

Full reference: local development.

Deploying the control plane

The Dockerfiles in deploy/docker/ are the deployment unit. Production runs on Railway. Other valid targets: Fly.io, ECS Fargate, single-VPS systemd. Migrations run automatically on backend boot.

Configuration is env-driven — see deploy/config/env.example for the full env reference, or the deployment guide for a step-by-step.

Realtime transport

Backend, consumer, and the Elixir realtime service all pick their event transport from one flag, PUBSUB_ENABLED, so they cannot disagree:

PUBSUB_ENABLED=false (default): events bridge over Redis (REDIS_URL). No GCP needed. This is the local-dev and simple self-host path.
PUBSUB_ENABLED=true: events flow through Google Pub/Sub. Also set GCP_PROJECT_ID and GOOGLE_APPLICATION_CREDENTIALS_JSON on every service. The backend and consumer auto-provision the realtime topics and their <topic>-sub pull subscriptions on boot (idempotent), so there is no manual gcloud step. The service account needs roles/pubsub.editor.

Set the flag the same on all three services. A publisher on Pub/Sub with a subscriber on Redis silently drops every realtime event.

Worker deployment

Workers run on per-VPS machines so cold-mail traffic spreads across many IPs. Worker identity is a deterministic UUIDv5 derived from the VPS's public IPv4 — same IP, same worker.

Add a worker from the admin dashboard:

Provision a VPS, note its public IP + root user
Admin → Workers → Add Worker
Copy the generated enrollment command
Run it on the VPS as root

The installer is served by the backend at /worker-install.sh. It exchanges the one-time token for worker config, writes /etc/warmbly/worker.env, configures systemd, enables a daily randomized self-update timer, and starts the worker container. The worker then heartbeats back to the backend and marks itself installed.

The older SSH-managed path is still supported: paste the generated SSH public key into the VPS's ~/.ssh/authorized_keys, then click Test and Install. From then on, lifecycle operations (restart, update, system updates, reboot, rotate keys, logs, uninstall) can happen from the dashboard.

Manual install on the VPS is also supported:

curl -fsSL https://api.example.com/worker-install.sh | sudo bash -s -- \
  --enroll wmenroll_... \
  --api-base https://api.example.com

# or fully manual:
sudo bash scripts/install-worker.sh \
  --kafka kafka.example.com:9092 \
  --schema-registry https://schema.example.com \
  --redis redis://cache.example.com:6379 \
  --aws-region us-east-1 --aws-key ... --aws-secret ...

Why per-VPS instead of Kubernetes DaemonSet

Cold-mail reputation lives at the IP level. K8s nodes typically NAT pods through a small set of egress IPs, so a per-node DaemonSet does not deliver IP diversity. Workers don't depend on Postgres, so cluster-level service discovery isn't needed. Spreading across VPS providers and regions is the only thing that actually moves the deliverability needle.

Worker env reference

Workers in production should be assigned to a worker profile in the dashboard. The profile bundles all of these:

Env var	Source	Notes
`APP_ENV`	profile	`prod` selects `alias/master-key`; otherwise `alias/master-key-dev`
`AWS_REGION`	profile (via AWS credentials row)
`AWS_ACCESS_KEY_ID`	profile (via AWS credentials row)
`AWS_SECRET_ACCESS_KEY`	profile (via AWS credentials row)	encrypted at rest
`KAFKA_BOOTSTRAP_SERVERS`	profile
`KAFKA_SASL_USERNAME`	profile
`KAFKA_SASL_PASSWORD`	profile	encrypted at rest
`SCHEMA_REGISTRY_URL`	profile
`SCHEMA_REGISTRY_KEY`	profile
`SCHEMA_REGISTRY_SECRET`	profile	encrypted at rest
`REDIS`	profile	full URL with embedded password; encrypted at rest
`WORKER_TIER`	(worker row)	`shared` or `dedicated`

The worker does not open a Postgres connection. Do not add one.

Auto-update

Each worker profile picks a release channel (pinned / stable / dev) and an auto_update toggle. When a GitHub release fires the webhook, the backend resolves the channel and (if auto_update=true) rolls every assigned worker. See the deployment guide.

Health checks

curl http://localhost:8080/health    # backend
curl http://localhost:3000/health   # tracking
curl http://localhost:4000/health   # realtime

README.md

Warmbly Deployment