lab
the home lab, documented and measured — the tracker records intent, the snapshot records reality.
live snapshot
probed 2026-11-17 (0h ago)cluster
kind-lab
carried forward — k8s v1.37.0 · control-plane ready
pods
32 / 34 ready
616 restarts (lifetime)
forge
down
https://forge.lab.local/
vms
none up
ssh fwd loopback / loopback
deployments
control-plane: cpu 357m · mem 2368Mi
the network, in one glance
the full map — addresses, ports, and the who-can-reach-whom matrix — lives in the workspace home-lab/network.md.
expansion tracker
7 / 9 completenginx + flask + postgres in one compose file — the baseline stack every later lab re-implements.
the compose stack translated to kubernetes: deployments, services, configmaps, the secret that is a bag of keys.
metrics-server, kube-state-metrics, prometheus, grafana, postgres-exporter bolted onto lab-02; alerts verified end-to-end with a chaos pod.
the three tiers declared in hcl — all 8 exercises run: s3 remote state with versioning + locking, the aws provider swap, and the stack instanced twice from a module.
network map (addresses, ports, who-can-reach-whom) plus a baseline playbook that is idempotent on both vm — changed=0 on the second run.
forgejo + runner (dind sidecar) + in-cluster registry over tls from cert-manager; the compose forge retired with its volumes kept as a rollback.
the chart is deployed by argo cd from the forge's own repo — sync-gitops.sh is the only path from editing the chart to a release.
first self-built stripped elf analyzed in the kali vm; five reproducible notes and a reused helper remain for completion.
everything currently faces loopback; going public needs a tunnel (cloudflared) in front of the ingress — deliberately last.
the study
cloud topics in progress — progress is the fraction of the topic worked through, proficiency is the honest self-grade (learning → working → comfortable), evidenced by what actually ran.
lab-01 compose stack, lab-04 containers via terraform, colima migration after docker desktop, the dind sidecar in the forge runner.
labs 02/03 on kind: deployments, services, configmaps/secrets, nodeport vs clusterip, boot-time retries, crashloop forensics.
prometheus + grafana + ksm + postgres-exporter on the cluster; alert fired → chaos pod deleted → self-resolved, verified end to end.
lab-04: hcl modules, s3 remote state with versioning and locking, provider swap to aws, drift created and destroyed on purpose.
baseline playbook idempotent on ubuntu + kali (changed=0 second run); learned serial-console debugging when ssh lied.
argo cd deploys the forge chart from the forge's own repo; sync-gitops.sh is the only release path.
scoped iam user limited to the terraform state bucket, root locked behind mfa with no keys, remote state in s3 with versioning and locking, a real bucket created/drifting/destroyed.
the home-lab network map: slirp nat quirks, docker bridges 172.17/18/19, the reachability matrix proven by curl, not by belief.
the write-ups of these labs are under blog — the whole arc is from a compose file to a certificate chain.