Gitops repository controlling talos linux k8s cluster
- Python 83%
- Dockerfile 8.7%
- Shell 8.3%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
|
|
||
| .forgejo/workflows | ||
| docs | ||
| gitops | ||
| images/turnstone | ||
| scripts | ||
| talos | ||
| .gitignore | ||
| LICENSE | ||
| README.md | ||
| renovate.json | ||
talos-gitops (n100-talos-cluster)
A 3-node Talos Linux Kubernetes cluster on N100 mini PCs, managed by Argo CD. This repo is the source of truth for the Talos machine config and everything running on the cluster. It also builds a few images and runs jobs for hosts outside the cluster.
- Cluster name
clusterfuck, APIhttps://192.168.1.200:6443, nodes192.168.1.200-202(all control plane, all schedulable). - LoadBalancer IPs
192.168.1.40-45(Cilium L2):.40Ingress,.44Prometheus remote write,.45Loki push. - Internal apps are
https://<app>.k8s.internal.nonmundane.org; public oneshttps://<app>.nonmundane.orgthrough farnsworth's Traefik.
Layout
talos/ Talos machine config: gen-config.sh, patches/, talos-mcp image
images/turnstone/ Patched Turnstone image (runs on farnsworth)
gitops/
clusters/n100/ App-of-apps root
infrastructure/ Argo CD, Cilium, Longhorn, cert-manager, external-dns, External Secrets,
CloudNativePG, Prometheus/Grafana, Loki, Alloy, Headlamp, metrics-server
apps/ Workloads. Each folder becomes an Argo CD Application of the same name,
in a namespace of the same name (apps-appset.yaml)
docs/ The documents below
docs/templates/ Reference manifests, kept out of gitops/apps/ so they aren't deployed
scripts/ render-edge-routes.py (public routes for farnsworth's Traefik)
.forgejo/workflows/ Renovate (for this repo and homelab-compose), edge-routes, image builds
Day to day
- Change something: edit, commit, push. Argo CD reconciles. Never
kubectl editorkubectl scalea managed resource:selfHealreverts it within one sync. - Add an app: create
gitops/apps/<name>/with plain manifests. Nometadata.namespaceneeded there. Every pod must declare requests and limits (namespaceResourceQuota). - Secrets: AWS Secrets Manager under the
nonmundane/prefix, regionus-west-2, pulled by anExternalSecretfrom theClusterSecretStorenamedaws-secretsmanager(no hyphen in "secretsmanager"). - Not applied by a push:
gitops/infrastructure/argocd(apply by hand), Cilium's Helm values (helm upgrade), and Longhorn (manual sync). See bootstrap.md. - Image updates: Renovate runs daily at 06:00 UTC. PRs labelled
manual-reviewormanual-apply-requiredneed a step from you after merging; the PR's rule description inrenovate.jsonsays which.
Docs
| doc | covers |
|---|---|
| bootstrap.md | rebuilding the cluster from nothing; upgrading Argo CD; Cilium, cert-manager, external-dns, Longhorn |
| observability.md | Prometheus, Grafana, Loki, Alloy; how the Compose hosts and forbin report in |
| public-apps.md | exposing an app through farnsworth's Traefik; Basic Auth; IP allowlists |
| backups.md | what is backed up to Garage, and how to add a volume or database |
| migrating.md | moving a service from Docker Compose onto the cluster |
| apps.md | per-app notes: what is unusual about each workload |
| farnsworth-services.md | talos-mcp, kubernetes-mcp, grafana-mcp and Turnstone: images built here, run on farnsworth |
| troubleshooting.md | problems hit more than once. Read before debugging from scratch |
| homelab-maint.md | nightly OS updates and reboots of forbin, the Pis and farnsworth |
| webhook.md | the torrent-completion webhook: setup and gotchas |
| ddns-updater.md | home.nonmundane.org and the AWS setup behind it |
| sl-presence.md | the Second Life presence poller and its Grafana dashboard |
| kubernetes-mcp.md | building kubernetes-mcp's read-only kubeconfig |
All documentation is in docs/; there are no READMEs under gitops/.
Known gaps
- Restores are untested (backups.md).
- Backups are on the same LAN as the cluster; there is no off-site copy.