Skip to main content
Version: 1.5

Deploying on Google Cloud

This is the remote runtimes topology as one terraform apply on Google Cloud: a small control-plane VM running Fadebox, a powerful worker VM running only Docker, and the two connected over mutual TLS — with a GCP-specific improvement over the generic guide: the daemon port never faces the internet. Fadebox reaches it on the worker's VPC-internal address, and the firewall rule for 2376 matches the control plane's network tag, not an IP you have to keep in sync.

you ──▶ fadebox-control (e2-small): Fadebox app + PostgreSQL, port 8080
              │
              │  tcp://10.10.0.x:2376 — VPC-internal only, mutual TLS
              ▼
        fadebox-worker (e2-standard-8): Docker daemon
              ├── Traefik ingress ◀── *.envs.example.com → static external IP
              └── deployed instances

Terraform creates the network, firewall rules, both VMs, static IPs, and the whole TLS story — CA, server and client certificates — and prints the three PEMs you paste into Fadebox at the end. Machine types, disk size, region and a Spot worker are all variables.

You need Terraform, the gcloud CLI authenticated with application default credentials (gcloud auth application-default login), a project with the Compute Engine API enabled, and — for instance URLs — a domain.

Running both runtimes at once — the worker plus the control plane's own local — needs a Team license, because Free allows one runtime in service. Free still runs this topology: disable local first and the worker becomes that one runtime, which is the arrangement you want anyway. See Licensing.

The files​

Put these four files in one directory, plus the compose file from the installation page saved as docker-compose.yml next to them — Terraform ships it to the control plane via instance metadata:

fadebox-gcp/
├── main.tf
├── variables.tf
├── control-plane-startup.sh
├── worker-startup.sh
└── docker-compose.yml        # verbatim from the installation page

variables.tf​

variable "project" {
  description = "GCP project id"
  type        = string
}

variable "region" {
  type    = string
  default = "us-central1"
}

variable "zone" {
  description = "Must lie in var.region"
  type        = string
  default     = "us-central1-a"
}

variable "control_machine_type" {
  description = "Fadebox itself is light — 2 GB of RAM is comfortable"
  type        = string
  default     = "e2-small"
}

variable "worker_machine_type" {
  description = "Size this for the environments you intend to run"
  type        = string
  default     = "e2-standard-8"
}

variable "worker_disk_gb" {
  description = "Images, containers and volumes all live here"
  type        = number
  default     = 100
}

variable "worker_spot" {
  description = "Run the worker as a Spot VM — a fraction of the price, but instances die on preemption"
  type        = bool
  default     = false
}

variable "admin_cidr" {
  description = "Who may reach the Fadebox UI on 8080. Narrow this to your own address"
  type        = string
  default     = "0.0.0.0/0"
}

variable "ingress_domain" {
  description = "Wildcard domain for instance URLs, e.g. envs.example.com. Empty skips the DNS record"
  type        = string
  default     = ""
}

variable "dns_managed_zone" {
  description = "Cloud DNS managed zone holding ingress_domain. Empty if DNS lives elsewhere"
  type        = string
  default     = ""
}

main.tf​

terraform {
  required_providers {
    google = { source = "hashicorp/google" }
    tls    = { source = "hashicorp/tls" }
  }
}

provider "google" {
  project = var.project
  region  = var.region
  zone    = var.zone
}

# ── Network and firewall ────────────────────────────────────────────────────

resource "google_compute_network" "fadebox" {
  name                    = "fadebox"
  auto_create_subnetworks = false
}

resource "google_compute_subnetwork" "fadebox" {
  name          = "fadebox"
  network       = google_compute_network.fadebox.id
  region        = var.region
  ip_cidr_range = "10.10.0.0/24"
}

# SSH via IAP tunnels only — no public port 22.
resource "google_compute_firewall" "ssh_iap" {
  name          = "fadebox-ssh-iap"
  network       = google_compute_network.fadebox.name
  source_ranges = ["35.235.240.0/20"]
  allow {
    protocol = "tcp"
    ports    = ["22"]
  }
}

# The Fadebox UI.
resource "google_compute_firewall" "fadebox_ui" {
  name          = "fadebox-ui"
  network       = google_compute_network.fadebox.name
  source_ranges = [var.admin_cidr]
  target_tags   = ["fadebox-control"]
  allow {
    protocol = "tcp"
    ports    = ["8080"]
  }
}

# Instance URLs on the worker.
resource "google_compute_firewall" "ingress" {
  name          = "fadebox-ingress"
  network       = google_compute_network.fadebox.name
  source_ranges = ["0.0.0.0/0"]
  target_tags   = ["fadebox-worker"]
  allow {
    protocol = "tcp"
    ports    = ["80", "443"]
  }
}

# The Docker daemon — only from the control plane, only VPC-internally.
resource "google_compute_firewall" "docker_tls" {
  name        = "fadebox-docker-tls"
  network     = google_compute_network.fadebox.name
  source_tags = ["fadebox-control"]
  target_tags = ["fadebox-worker"]
  allow {
    protocol = "tcp"
    ports    = ["2376"]
  }
}

# Lets Traefik on the worker call Fadebox — needed for instances that require a sign-in.
# It calls the public URL, the control VM's external address, so the packets come from the
# worker's external address: GCP matches source_tags against internal addresses only.
resource "google_compute_firewall" "worker_to_fadebox" {
  name          = "fadebox-worker-to-control"
  network       = google_compute_network.fadebox.name
  source_ranges = ["${google_compute_address.worker.address}/32"]
  target_tags   = ["fadebox-control"]
  allow {
    protocol = "tcp"
    ports    = ["8080"]
  }
}

# ── Addresses ───────────────────────────────────────────────────────────────

resource "google_compute_address" "control" {
  name = "fadebox-control"
}

resource "google_compute_address" "worker" {
  name = "fadebox-worker"
}

# Reserved up front so the server certificate can name it before the VM exists.
resource "google_compute_address" "worker_internal" {
  name         = "fadebox-worker-internal"
  address_type = "INTERNAL"
  subnetwork   = google_compute_subnetwork.fadebox.id
}

# ── TLS material for the Docker daemon ──────────────────────────────────────

resource "tls_private_key" "ca" {
  algorithm = "RSA"
  rsa_bits  = 4096
}

resource "tls_self_signed_cert" "ca" {
  private_key_pem       = tls_private_key.ca.private_key_pem
  is_ca_certificate     = true
  validity_period_hours = 87600 # 10 years
  allowed_uses          = ["cert_signing", "crl_signing"]
  subject {
    common_name = "fadebox-docker-ca"
  }
}

resource "tls_private_key" "server" {
  algorithm = "RSA"
  rsa_bits  = 4096
}

resource "tls_cert_request" "server" {
  private_key_pem = tls_private_key.server.private_key_pem
  ip_addresses    = [google_compute_address.worker_internal.address]
  subject {
    common_name = "fadebox-worker"
  }
}

resource "tls_locally_signed_cert" "server" {
  cert_request_pem      = tls_cert_request.server.cert_request_pem
  ca_private_key_pem    = tls_private_key.ca.private_key_pem
  ca_cert_pem           = tls_self_signed_cert.ca.cert_pem
  validity_period_hours = 26280 # 3 years
  allowed_uses          = ["server_auth", "digital_signature", "key_encipherment"]
}

resource "tls_private_key" "client" {
  algorithm = "RSA"
  rsa_bits  = 4096
}

resource "tls_cert_request" "client" {
  private_key_pem = tls_private_key.client.private_key_pem
  subject {
    common_name = "fadebox"
  }
}

resource "tls_locally_signed_cert" "client" {
  cert_request_pem      = tls_cert_request.client.cert_request_pem
  ca_private_key_pem    = tls_private_key.ca.private_key_pem
  ca_cert_pem           = tls_self_signed_cert.ca.cert_pem
  validity_period_hours = 26280
  allowed_uses          = ["client_auth", "digital_signature"]
}

# ── The two VMs ─────────────────────────────────────────────────────────────

resource "google_compute_instance" "control" {
  name         = "fadebox-control"
  machine_type = var.control_machine_type
  zone         = var.zone
  tags         = ["fadebox-control"]

  boot_disk {
    initialize_params {
      image = "debian-cloud/debian-12"
      size  = 20
    }
  }

  network_interface {
    subnetwork = google_compute_subnetwork.fadebox.id
    access_config {
      nat_ip = google_compute_address.control.address
    }
  }

  metadata = {
    startup-script  = file("${path.module}/control-plane-startup.sh")
    fadebox-compose = file("${path.module}/docker-compose.yml")
  }
}

resource "google_compute_instance" "worker" {
  name         = "fadebox-worker"
  machine_type = var.worker_machine_type
  zone         = var.zone
  tags         = ["fadebox-worker"]

  boot_disk {
    initialize_params {
      image = "debian-cloud/debian-12"
      size  = var.worker_disk_gb
    }
  }

  network_interface {
    subnetwork = google_compute_subnetwork.fadebox.id
    network_ip = google_compute_address.worker_internal.address
    access_config {
      nat_ip = google_compute_address.worker.address
    }
  }

  scheduling {
    provisioning_model          = var.worker_spot ? "SPOT" : "STANDARD"
    preemptible                 = var.worker_spot
    automatic_restart           = var.worker_spot ? false : true
    instance_termination_action = var.worker_spot ? "STOP" : null
  }

  metadata = {
    startup-script  = file("${path.module}/worker-startup.sh")
    tls-ca          = tls_self_signed_cert.ca.cert_pem
    tls-server-cert = tls_locally_signed_cert.server.cert_pem
    tls-server-key  = tls_private_key.server.private_key_pem
  }
}

# ── Wildcard DNS, if Cloud DNS hosts the domain ─────────────────────────────

resource "google_dns_record_set" "wildcard" {
  count        = var.dns_managed_zone == "" ? 0 : 1
  managed_zone = var.dns_managed_zone
  name         = "*.${var.ingress_domain}."
  type         = "A"
  ttl          = 300
  rrdatas      = [google_compute_address.worker.address]
}

# ── What you need afterwards ────────────────────────────────────────────────

output "fadebox_url" {
  value = "http://${google_compute_address.control.address}:8080"
}

output "worker_external_ip" {
  description = "Point *.your-ingress-domain here if DNS lives outside Cloud DNS"
  value       = google_compute_address.worker.address
}

output "runtime_docker_host" {
  description = "The Docker host URI for the new runtime"
  value       = "tcp://${google_compute_address.worker_internal.address}:2376"
}

output "runtime_ca_cert" {
  value = tls_self_signed_cert.ca.cert_pem
}

output "runtime_client_cert" {
  value = tls_locally_signed_cert.client.cert_pem
}

output "runtime_client_key" {
  value     = tls_private_key.client.private_key_pem
  sensitive = true
}

control-plane-startup.sh​

Runs on every boot; everything in it is guarded, so re-boots are no-ops. It installs Docker, pulls the compose file out of instance metadata, generates the secrets .env once, and starts Fadebox:

#!/bin/bash
set -euo pipefail

if ! command -v docker >/dev/null; then
  curl -fsSL https://get.docker.com | sh
fi

# Cap every container's log store at about 50 MB (Concepts → Runtimes → Container logs).
if [ ! -f /etc/docker/daemon.json ]; then
  cat > /etc/docker/daemon.json <<'EOF'
{ "log-driver": "json-file", "log-opts": { "max-size": "10m", "max-file": "5" } }
EOF
  systemctl restart docker
fi

mkdir -p /opt/fadebox
cd /opt/fadebox

curl -fsSL -H "Metadata-Flavor: Google" \
  "http://metadata.google.internal/computeMetadata/v1/instance/attributes/fadebox-compose" \
  -o docker-compose.yml

if [ ! -f .env ]; then
  {
    echo "DOCKER_GID=$(stat -c '%g' /var/run/docker.sock)"
    echo "FADEBOX_SESSION_KEY=$(openssl rand -base64 32)"
    echo "FADEBOX_ENCRYPTION_KEY=$(openssl rand -base64 32)"
    echo "DB_PASSWORD=$(openssl rand -base64 24)"
  } > .env
  chmod 600 .env
fi

docker compose up -d

The secrets are generated on the VM, not in Terraform, so they never enter the state file. FADEBOX_ENCRYPTION_KEY lives only in /opt/fadebox/.env — back it up along with the database (see Secrets at rest).

worker-startup.sh​

Installs Docker, fetches the TLS material from instance metadata, and puts the daemon on 2376 with tlsverify — including the systemd override that drops the stock -H fd:// flag, which otherwise conflicts with hosts in daemon.json:

#!/bin/bash
set -euo pipefail

if ! command -v docker >/dev/null; then
  curl -fsSL https://get.docker.com | sh
fi

md() {
  curl -fsSL -H "Metadata-Flavor: Google" \
    "http://metadata.google.internal/computeMetadata/v1/instance/attributes/$1"
}

mkdir -p /etc/docker/tls
md tls-ca          > /etc/docker/tls/ca.pem
md tls-server-cert > /etc/docker/tls/server-cert.pem
md tls-server-key  > /etc/docker/tls/server-key.pem
chmod 600 /etc/docker/tls/server-key.pem

cat > /etc/docker/daemon.json <<'EOF'
{
  "hosts": ["unix:///var/run/docker.sock", "tcp://0.0.0.0:2376"],
  "tlsverify": true,
  "tlscacert": "/etc/docker/tls/ca.pem",
  "tlscert": "/etc/docker/tls/server-cert.pem",
  "tlskey": "/etc/docker/tls/server-key.pem",
  "log-driver": "json-file",
  "log-opts": { "max-size": "10m", "max-file": "5" }
}
EOF

mkdir -p /etc/systemd/system/docker.service.d
cat > /etc/systemd/system/docker.service.d/override.conf <<'EOF'
[Service]
ExecStart=
ExecStart=/usr/bin/dockerd
EOF

systemctl daemon-reload
systemctl restart docker

Apply​

terraform init
terraform apply -var project=my-project

Give the startup scripts two or three minutes after the apply finishes — they still have Docker to install and images to pull. Then open the URL from the output and fetch the initial admin password from the app log:

terraform output -raw fadebox_url

gcloud compute ssh fadebox-control --zone us-central1-a --tunnel-through-iap \
  --command 'sudo docker compose --project-directory /opt/fadebox logs app | grep "generated password"'

Sign in and change it.

Register the worker runtime​

On a single-runtime tier, disable local first — an enabled runtime holds the slot, and the limit is checked as the new one is created. On Team and above the order does not matter.

Under Settings → Runtimes, add a runtime and fill it from the Terraform outputs:

terraform output -raw runtime_docker_host    # → Docker host
terraform output -raw runtime_ca_cert        # → CA certificate
terraform output -raw runtime_client_cert    # → Client certificate
terraform output -raw runtime_client_key     # → Client key

Use Test connection — it does a real round trip to the daemon. Then disable the seeded local runtime so nothing deploys onto the small VM — see the remote runtimes guide for why.

Ingress and HTTPS​

If you set dns_managed_zone, the wildcard record already points at the worker; otherwise create *.envs.example.com → worker_external_ip wherever the domain lives. Then, on the worker runtime under Settings → Runtimes, set the ingress domain and install the managed ingress stack — once FADEBOX_PUBLIC_URL is set, as the end of this section explains; the full story is in Ingress and instance URLs.

For HTTPS with the wildcard certificate issued through Cloud DNS, create a service account that may edit the zone and give the runtime its key as ACME settings:

gcloud iam service-accounts create fadebox-acme
gcloud projects add-iam-policy-binding my-project \
  --member="serviceAccount:fadebox-acme@my-project.iam.gserviceaccount.com" \
  --role=roles/dns.admin
gcloud iam service-accounts keys create acme-key.json \
  --iam-account=fadebox-acme@my-project.iam.gserviceaccount.com
jq -c . acme-key.json    # the single-line form you paste below

ACME DNS provider is gcloud, and the credentials are two VAR=value lines — the key JSON must be on one line (that is what the jq -c above produces), because each line is parsed separately:

GCE_PROJECT=my-project
GCE_SERVICE_ACCOUNT={"type":"service_account","project_id":"my-project",...}

Set FADEBOX_PUBLIC_URL on the app container to the URL you open Fadebox at before installing the worker's ingress stack. Traefik on the worker polls Fadebox for routes, and the address Fadebox would work out from inside its container is not one the worker can reach, so without the setting the install is refused (reason: control_plane_address_unusable). If any instance requires a sign-in, the same Traefik sends visitors to that URL to sign in and checks their sessions by calling Fadebox there. It is the control VM's external address, so those calls arrive from the worker's external address, which is what the fadebox-worker-to-control firewall rule admits.

What to know about this setup​

  • The Terraform state contains private keys — the Docker CA, server and client keys. Keep it where you would keep a password: a remote backend with restricted access, not a public repo. The worker's server key also sits in its instance metadata, readable by anyone with compute viewer on the project — project IAM is the trust boundary here.
  • The UI is plain HTTP on 8080. Narrow admin_cidr to your own address, and put a TLS-terminating proxy in front before letting a team use it. The worker keeps reaching Fadebox through fadebox-worker-to-control, not fadebox-ui. If the control VM runs a managed ingress stack of its own, that stack calls the public URL too — add the control VM's external address (google_compute_address.control) to fadebox-ui's source_ranges.
  • A Spot worker (-var worker_spot=true) is a fine fit for disposable dev environments and a poor fit for anything that must stay up: preemption stops the VM and every instance on it. Deploys after the restart work unchanged — the reserved internal IP and the TLS material both survive.
  • Certificates expire — the ones above after 3 years. terraform taint the cert resources (or shorten validity_period_hours and let a plan rotate them), re-apply, reboot the worker to re-run its startup script, and paste the new client material into the runtime.
  • terraform destroy deletes everything, disks and database included. The only state worth backing up first is on the control plane: the PostgreSQL volume and /opt/fadebox/.env.