Luke Petzer

Case study · ixo · Platform

Multi-cloud Terraform platform

An open-source, config-driven Terraform platform across Vultr, AWS and GCP. A team adds one .tfvars file and gets its own environment, so anyone with code-level skills can bring up their own ixo-like stack.

Context

ixo runs its platform on Kubernetes across three clouds: AWS (EKS), GCP (GKE) and Vultr (VKE). I joined as the sole platform hire, half-time from February 2024 alongside my job at BMTX and full-time since March 2025.

Before the platform, the infrastructure was mostly click-ops: resources were created by hand in the cloud consoles, with little or no infrastructure as code.

Why

ixo wanted people with code-level skills to be able to pick up our Terraform and bring up their own ixo-like infrastructure on their own cloud. We run services like Matrix alongside services that work with ours, so the whole stack has to be reproducible, not just one cluster.

Approach

  • Built a config-driven Terraform platform: a team adds a .tfvars file with its values for the shared platform modules, instead of writing its own infrastructure code.
  • Environments live in an environments map, with application_configs switching services on or off. Secrets come in through TF_VAR_* environment variables, and each environment gets its own Terraform workspace.
  • Clusters run on Vultr (VKE) and AWS (EKS), with GCP optional for Vault auto-unseal and backups. DNS goes through Cloudflare (external-dns), with Tailscale for VPN access and GitHub OAuth for login.
  • Wrote the onboarding docs for teams plugging in, including a quickstart and a cost-estimate script.

An illustrative sketch of the shape (not a real config):

# illustrative only
environments = {
  staging = {
    application_configs = {
      matrix     = { enabled = true }
      monitoring = { enabled = true }
    }
  }
}

What it deploys

ArgoCD, Prometheus / Grafana / Loki, Vault + Dex, PostgreSQL clusters with backups, ingress-nginx, cert-manager, Matrix, Ghost and Uptime Kuma: the core services and the monitoring around them.

Result

  • Infrastructure that used to be clicked together by hand is now code, in one platform any team can plug into.
  • The platform is open source: ixofoundation/ixo-terra-infra (Apache-2.0).
  • Also at ixo: I ran cost comparisons of multi-cloud against an AWS-only setup (28–36% savings), and cut backup storage costs by 60–80% with GCS lifecycle policies.

All work