Files

T

Antigravity Agent 504fd5fb42 [G1-Sync] Manual knowledge update

2026-05-10 22:08:15 +09:00

6.9 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

Platform Engineering

매 한 줄

"매 dev → prod 의 길을 product 처럼 디자인하는 분야". 매 2022 Team Topologies + Gartner Hype Cycle 의 trigger → 매 2026 모든 mid-large 조직의 default. 매 IDP (Internal Developer Platform) 가 product, application team 이 customer.

매 핵심

매 DevOps 와 차이

DevOps = 매 culture / practice ("you build it, you run it").
Platform Engineering = 매 그 cognitive load 를 줄이는 product approach.
매 DevOps 의 "everyone owns infra" → 매 burnout. PE 는 매 "platform team owns paved road".

매 4 component (Humanitec, CNCF Platform WG)

Developer Control Plane: 매 portal (Backstage), CLI, API.
Integration & Delivery: 매 CI/CD, GitOps (ArgoCD, Flux).
Resource Plane: 매 K8s, cloud, DB, message queue.
Security Plane: 매 secrets, policy (OPA), supply chain (SLSA, Sigstore).

매 Golden Path

매 80% case 의 paved road. 매 옆길 가능 but cost 명시.
매 service template (cookiecutter) + 매 infra module (Terraform) + 매 deploy pipeline + 매 observability default.

매 핵심 원칙

Treat platform as product: 매 PM, roadmap, NPS, SLO.
Self-service: 매 ticket → 매 button.
Opinionated: 매 freedom of choice 의 cost > 매 standardization 가치.
Thinnest viable platform (Pulumi 표현): 매 build 보다 reuse.
Telemetry-driven: 매 DORA + DevEx (SPACE) measurement.

매 metric

DORA: deploy freq, lead time, MTTR, change fail rate.
DevEx (SPACE): Satisfaction, Performance, Activity, Communication, Efficiency.
Platform: time-to-first-deploy, % services on golden path, ticket reduction.

💻 패턴

Backstage software template

# template.yaml
apiVersion: scaffolder.backstage.io/v1beta3
kind: Template
metadata:
  name: nodejs-service
  title: 매 Node.js Service (Golden Path)
spec:
  parameters:
    - title: Basics
      properties:
        name: { type: string, pattern: "^[a-z][a-z0-9-]*$" }
        owner: { type: string, ui:field: OwnerPicker }
  steps:
    - id: fetch
      action: fetch:template
      input:
        url: ./skeleton
        values: { name: "${{ parameters.name }}" }
    - id: publish
      action: publish:github
      input:
        repoUrl: github.com?owner=acme&repo=${{ parameters.name }}
        defaultBranch: main
    - id: register
      action: catalog:register
      input:
        repoContentsUrl: ${{ steps.publish.output.repoContentsUrl }}

Score / workload spec (platform-agnostic)

# score.yaml — 매 dev 가 작성, 매 platform 이 K8s/ECS/Cloud Run으로 번역
apiVersion: score.dev/v1b1
metadata: { name: hello-svc }
containers:
  api:
    image: ghcr.io/acme/hello:1.2.3
    variables:
      DB_URL: ${resources.db.url}
resources:
  db:
    type: postgres
service:
  ports:
    web: { port: 80, targetPort: 3000 }

Crossplane composition (provisioning)

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata: { name: postgres-aws }
spec:
  compositeTypeRef:
    apiVersion: platform.acme/v1alpha1
    kind: XPostgres
  resources:
    - base:
        apiVersion: rds.aws.upbound.io/v1beta2
        kind: Instance
        spec:
          forProvider:
            engine: postgres
            engineVersion: "16.3"
            instanceClass: db.t4g.micro
            allocatedStorage: 20
            backupRetentionPeriod: 7

OPA / Gatekeeper policy

package k8sallowedrepos

violation[{"msg": msg}] {
  c := input.review.object.spec.containers[_]
  not startswith(c.image, "ghcr.io/acme/")
  msg := sprintf("매 image '%v' 는 허용된 registry 아님", [c.image])
}

ArgoCD ApplicationSet (multi-env)

apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata: { name: hello-svc }
spec:
  generators:
    - list:
        elements:
          - { env: dev, cluster: dev-cluster, replicas: "1" }
          - { env: prod, cluster: prod-cluster, replicas: "5" }
  template:
    metadata: { name: "hello-svc-{{env}}" }
    spec:
      project: default
      source:
        repoURL: https://github.com/acme/hello
        path: deploy
        helm:
          parameters:
            - { name: replicas, value: "{{replicas}}" }
      destination: { server: "{{cluster}}" }
      syncPolicy: { automated: { prune: true, selfHeal: true } }

DORA metric collection

// 매 GitHub deploy event → DORA metric
github.on("deployment_status", async (e) => {
  if (e.deployment_status.state !== "success") return;
  const deployedAt = new Date(e.deployment_status.created_at);
  const commitTime = await getCommitTime(e.deployment.sha);
  const leadTimeSec = (+deployedAt - +commitTime) / 1000;
  await metrics.write({
    service: e.deployment.environment.split("-")[0],
    deploy_freq_inc: 1,
    lead_time_sec: leadTimeSec,
  });
});

매 결정 기준

상황	Approach
<50 engineers	Lightweight: shared Helm charts, GitOps, basic Backstage
50-500	Full IDP: Backstage + golden paths + Crossplane
500+	Multi-cluster, multi-region, FinOps integrated
Heterogeneous stack	Score / OAM 로 abstract
Highly regulated (gov/fin)	OPA policy + SLSA L3 + signed images

기본값: 매 Backstage + GitOps (ArgoCD) + opinionated golden path + DORA.

🔗 Graph

부모: DevOps · SRE · Team Topologies
변형: Internal Developer Platform · DevEx Platform
응용: Backstage · Crossplane · ArgoCD · Humanitec
Adjacent: GitOps · Service Catalog · Golden Path · DORA Metrics

🤖 LLM 활용

언제: 매 50+ engineer 조직, 매 multi-team friction, 매 cognitive load 증가, 매 onboarding 느림. 언제 X: 매 single team / startup pre-PMF — 매 platform overhead 의 prematurely.

❌ 안티패턴

Platform team = ticket queue: 매 product mindset 의 X. → 매 self-service 못 함.
Build everything: 매 OSS 안 쓰고 자체 — 매 cost 폭발.
No customers: 매 application team 의 needs 무시.
YAML hell: 매 abstraction 안 만들고 raw K8s manifest 강요.
Tightly coupled stack: 매 Score/OAM 없이 → 매 cloud lock.
Measure 부재: 매 NPS/DORA 없으면 매 ROI 증명 X.

🧪 검증 / 중복

Verified (CNCF Platform WG whitepaper, Team Topologies, ThoughtWorks Tech Radar 2026, Humanitec State of Platform Engineering reports).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — IDP components + golden path patterns

6.9 KiB Raw Blame History