Customizing Terraform Stack Templates

• AWS
• Azure
• GCP

​

How the templates read their configuration

data "stack_config" "this" {
  install_id = var.install_id
}

locals {
  runner_id                  = data.stack_config.this.runner_id
  runner_api_url             = data.stack_config.this.runner_api_url
  phone_home_url             = data.stack_config.this.phone_home_url
  nuon_support_iam_role_arns = data.stack_config.this.aws.nuon_support_iam_role_arns
  # ...
}

​

AWS

​

Runner instance

• An EC2 instance (typically launched by an Auto Scaling Group) in a subnet with outbound HTTPS to the Nuon API and to GitHub raw — the boot process downloads init-mng-v2.sh and the runner binary over the public internet.
• IMDSv2 must be enabled with required tokens. The runner authenticates to ctl-api using its Instance Identity Document, which is read from IMDS.
• Instance tags must include:
  • nuon_runner_id
  • nuon_runner_api_url
  • nuon_install_id — used in the per-install CloudWatch log group path.
  init-mng-v2.sh reads these via ec2:DescribeTags. Missing or misnamed tags cause the runner to fail to start.
• user_data must export RUNNER_AUTH_METHOD=iid before invoking the init script. The init script defaults to sts auth (legacy) if this is not set.

​

Runner IAM role

• sts:AssumeRole on every operation, break-glass, and custom role the stack creates. The runner switches into these roles to execute components.
• secretsmanager:GetSecretValue and secretsmanager:DescribeSecret on the secret ARN pattern your stack uses (the default stack uses <prefix>-*). Required for components that consume secrets.
• logs:CreateLogGroup, CreateLogStream, PutLogEvents, DescribeLogStreams on /nuon/<install_id>/* and runner-* log groups.
• ec2:DescribeTags on * — the init script’s tag lookups.

​

Operation, break-glass, and custom roles

• One IAM role per [operation_role], [[break_glass_role]], and [[custom_role]] block declared in stack.toml.
• Trust policy must allow sts:AssumeRole from:
  • The runner role ARN — the runner assumes them when executing components or actions.
  • The Nuon control-plane principals (data.stack_config.this.aws.nuon_support_iam_role_arns, falling back to account root) — used for ad-hoc actions and break-glass operations initiated from ctl-api.
• Permissions on each role come from your stack.toml (provision_inline_policy, provision_managed_policy_arns, etc.). The stack’s job is to translate those into IAM resources.
• Role names must match each.key from stack.toml verbatim. ctl-api looks roles up by exact name, and the default stack deliberately doesn’t double-prefix break-glass / custom roles. If you rewrap names with a prefix you’ll hit the 64-character IAM role-name limit and break role lookups.

​

Phone-home payload

​

GCP

​

Runner instance

• A google_compute_instance (or MIG) in a subnet with outbound HTTPS to the Nuon API and to GitHub raw.
• The runner authenticates with the token returned by data.stack_config.this.gcp.runner_api_token. Protect Terraform state as sensitive because the token is stored in instance metadata.
• The same init-mng-v2.sh flow applies: instance metadata must carry nuon_runner_id, nuon_runner_api_url, and nuon_install_id so the init script can read them.

​

Runner service account

• iam.serviceAccounts.getAccessToken and iam.serviceAccounts.signBlob on every operation, break-glass, and custom service account the stack creates — the runner impersonates these to execute components.
• Read access to whatever Secret Manager secrets the stack provisions (the default stack scopes by name prefix).
• logging.logEntries.create for runner logs.

​

Operation, break-glass, and custom service accounts

• One service account per [operation_role], [[break_glass_role]], and [[custom_role]] block in stack.toml.
• Each must grant the runner service account roles/iam.serviceAccountTokenCreator so the runner can impersonate it.
• Permissions on each service account come from your stack.toml (inline + managed policies are translated into IAM bindings on the project).
• Report each role under its exact configured role name. Resource names may use deterministic hashes to satisfy GCP’s 30-character service-account ID limit.

​

Phone-home payload

​

Azure

​

Runner virtual machine

• A Linux virtual machine in a subnet with outbound HTTPS to the Nuon API and artifact hosts.
• Metadata passed to cloud-init must include the runner ID and runner API URL from stack_config.
• The runner identity needs the role assignments required to assume configured operation, break-glass, and custom roles.

​

Phone-home payload

​

What you can customize freely

• VPC / VNet / VPC layout, CIDRs, subnet count and sizes — as long as the runner subnet has working egress to the Nuon API.
• NAT vs public-subnet egress.
• Instance type / machine type and image — anything that supports cloud-init and can run the runner binary works (AL2023, Ubuntu LTS, Amazon Linux 2, Debian, COS).
• Tags / labels, KMS keys, access logging, flow logs, private service endpoints, peering, transit gateways, DNS zones.
• Wrapping the published registry module from a parent Terraform configuration instead of forking:
  provider "stack" {}
  
  module "nuon" {
    source  = "nuonco/stack/azure"
    version = "~> 1.0"
  
    install_id = var.install_id
  }
  
  The module reads the rest of its configuration from the control plane via the stack_config data source, so you pass only the install ID and supported overrides. Use each module’s README for its current source, version, required cloud provider configuration, and supported overrides.

​

Contributing