第 4 回 V-G1 で LLM が route フィルタのない 5 分窓の p95(0.017 秒)を根拠に 「遅延なし」と結論した。slo.md にアラートごとの基準クエリ(規則と同じ式、 legacy のインスタンス別遅延と要求率を含む)を書き、確かめ直すときは同じ式を 使わせる。各規則の annotations.condition に式をそのまま入れ、AI 側が 「アラート条件が解消したか」を計算できるようにする(Jev の recovering 再定義)。
selfheal-lab-gitops
GitOps and configuration repository for the self-healing verification lab
(VM + container mixed environment). Infrastructure lives in
seon-labs/selfheal-lab-terraform; this repository holds everything that runs
on top of it.
This repository is public on purpose: every EC2 node and GCE VM clones it anonymously at first boot (cloud-init / startup script), and MIG instances are recreated by autohealing without an operator. Never commit secrets here; values live in GCP Secret Manager and are fetched at runtime.
| Path | Owner | Filled in |
|---|---|---|
bootstrap/kubeadm/ |
operator, run once | WP-07: containerd/kubeadm init+join, flannel, Argo CD install, root-app |
argocd/root-app.yaml |
Argo CD app-of-apps root | WP-07 |
apps/ |
child Argo CD Applications | WP-14 |
workloads/{demo-app,ingress-nginx,chaos-mesh,alloy}/ |
Argo CD | WP-14, WP-15 |
vm/<role>/apply.sh |
GCE startup script and make vm-sync |
WP-08..12 |
vm/<role> matches the role metadata each VM gets from Terraform:
observability, harness, ai-env, postgres-primary, postgres-replica,
legacy-api. The startup script runs vm/<role>/apply.sh when it exists and is
executable; apply.sh must be idempotent because make vm-sync re-runs it.