jgrusewski
8fa7eb4bcf
fix(infra): fix alloy-values for k8s-monitoring v1.x + gitlab prometheus OOM
...
- Alloy: split host/writeEndpoint, use authMode none + extraHeaders
- Alloy: disable logs/events/opencost, set dummy Loki host (chart requires it)
- Alloy: remove kube-state-metrics nodeSelector (DEV1-M 4GB too small)
- GitLab: increase prometheus memory 512Mi→1Gi (OOMKilled on WAL replay)
- GitLab: remove writeRelabelConfigs (Prom v3 rejects camelCase config)
- GitLab: add install instructions for passing secrets via --set
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com >
2026-02-27 01:00:40 +01:00
jgrusewski
51257dbe13
feat(infra): add Cockpit metrics + NAT gateway
...
Workstream 1 — NAT Gateway:
- New public-gateway TF module (VPC-GW-S, masquerade, push_default_route)
- Disable public IPs on all 6 Kapsule node pools
- Terragrunt live config with kapsule dependency
Workstream 2 — Scaleway Cockpit Metrics:
- New cockpit TF module (data source, push token, Grafana user)
- Grafana Alloy k8s-monitoring Helm values (node-exporter, kube-state-metrics)
- DCGM Exporter DaemonSet for GPU metrics (affinity on nvidia.com/gpu.present)
- GitLab Prometheus remote_write to Cockpit (filtered to gitlab_/gitaly_/sidekiq_)
- Apply-order runbook (3 phases: Cockpit → GitLab → NAT)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com >
2026-02-27 00:33:55 +01:00
