Uses

Self-hosted on its own VM. It is the GitOps source, container registry, CI control plane, and durable operating record. Runner capacity spans Linux on x86-64 and ARM64, macOS, and Windows. The macOS and Windows runners stay powered down until a job needs them.

Five-node cluster with Ceph underneath. The metal the whole lab runs on.

Storage for the cluster, on the order of 230 TiB raw. It taught me the most, mostly the hard way.

The workload layer. Immutable nodes, API-driven, no SSH.

GitOps. A push to the repo is the deploy. Rollback is a git revert.

Provisions the Proxmox VMs and LXCs below the cluster.

Configures the platform underneath the cluster.

Local builds and testing, on the workstation and the dev box both.

HA firewall pair with CARP failover. Routes the whole lab, and the one platform I have code merged into. Twice, six years apart.

The cluster CNI, doing L2/BGP so services get real routable IPs.

Ingress and TLS for everything the cluster serves.

Internal resolvers. dnsweaver keeps their records honest.

DNS and edge in front of the public sites.

HA Postgres on dedicated VMs, kept off the cluster on purpose so it stays rebuildable from git.

The other stateful backends the apps lean on.

Metrics and dashboards. If it runs, it emits.

Runtime secrets, kept out of git.

The glue and the scripts that run the lab, and my preference anywhere it will run. PowerShell when I have no choice, with the reference docs open.

Public-repository CI for linting, tests, CodeQL, documentation, and container health.

Keeps dependencies and container images current through reviewable pull requests.

Everywhere. WSL on the workstation, Talos on the cluster.

I set the architecture, standards, and acceptance criteria, then direct agents to implement. Scheduled work starts report-only; I promote it only after its output has been verified.

Small changes, reviewed, reconciled from git.

If it's important, it's written down. Decisions get recorded.