89 lines
3.0 KiB
YAML
89 lines
3.0 KiB
YAML
# ============================================
|
|
# MLOps Architecture - Infrastructure Config
|
|
# 클러스터/컴포넌트 구성 정보 (시크릿 제외)
|
|
# ============================================
|
|
|
|
cluster:
|
|
name: gpulive
|
|
api_server: https://114.110.128.68:6443
|
|
k8s_version: v1.34.3
|
|
masters: 3
|
|
workers: 14 # GPU 노드 포함
|
|
storage_classes:
|
|
default: component-nas # NFS provisioner
|
|
internal: internal-nas
|
|
user: user-nas
|
|
node_groups:
|
|
nd: dv2-kr3-ins-nd-worker-01 # 인프라 컴포넌트용
|
|
prometheus: dv2-kr3-ins-nd-worker-02 # 모니터링용
|
|
|
|
# ── 스케줄링 정책 ──────────────────────────────
|
|
scheduling:
|
|
# 워크로드 제외 모든 컴포넌트는 nodeSelector 지정
|
|
infra_node_selector:
|
|
nodegroup: nd # 기본값
|
|
monitoring_node_selector:
|
|
nodegroup: prometheus
|
|
|
|
# ── 컴포넌트 현황 ────────────────────────────
|
|
components:
|
|
gitea:
|
|
location: external # 다른 클러스터에 설치됨
|
|
url: https://gitea.inje-private.com
|
|
version: 1.22.1
|
|
installed: true # 외부에서 운영 중
|
|
|
|
gitea_runner:
|
|
location: vm # 전용 VM에 직접 설치
|
|
host: soo-runner # ssh ubuntu@runner
|
|
version: v0.3.0 # act_runner version
|
|
deploy_type: systemd # systemd service on VM
|
|
runner_name: soo-runner
|
|
scope: organization # 조직(soo-rnd) 레벨 runner
|
|
organization: soo-rnd
|
|
binary: /usr/local/bin/act_runner
|
|
workdir: /opt/act_runner
|
|
service_file: /etc/systemd/system/act_runner.service
|
|
labels:
|
|
- ubuntu-latest
|
|
- ubuntu-22.04
|
|
- self-hosted
|
|
installed: true
|
|
|
|
argocd:
|
|
namespace: argocd
|
|
version: v2.12.3
|
|
ingress: argocd.gpulive.cloud
|
|
installed: true
|
|
|
|
harbor:
|
|
url: harbor.inje-private.com
|
|
project: mlops # harbor.inje-private.com/mlops/
|
|
installed: true
|
|
location: external
|
|
|
|
dvc:
|
|
version: 3.66.1
|
|
remote_type: nas
|
|
remote_name: nas
|
|
nas_server: 192.168.0.43
|
|
nas_path: /GJ_SHARE_FS2/afed9d8d-f9c8-4053-aa89-23cdc70938e3
|
|
nas_mount: /ainas # 실제 마운트 포인트
|
|
dvc_storage: /ainas/dvc-storage
|
|
installed: true
|
|
|
|
# ── 네임스페이스 매핑 ────────────────────────
|
|
namespaces:
|
|
infra: gitea # Gitea, Runner
|
|
cd: argocd # ArgoCD
|
|
workload: soo # ML 워크로드 배포 대상
|
|
monitoring: monitoring # Prometheus, Grafana
|
|
|
|
# ── GitOps 구조 ──────────────────────────────
|
|
gitops:
|
|
# single: 앱 코드 + manifest 한 repo
|
|
# split: 앱 repo / manifest repo 분리
|
|
repo_strategy: "" # TODO: 결정 후 채우기
|
|
app_repo: ""
|
|
manifest_repo: ""
|