Files

T

Antigravity Agent 93ec7e9056 [G1-Sync] Manual knowledge update

2026-05-09 21:08:02 +09:00

5.4 KiB

Raw Blame History

id, title, category, status, source_trust_level, verification_status, created_at, updated_at, tags, tech_stack, applied_in, aliases

title

FinOps

Cloud bill 가 OOC (out of control). Tag → cost allocation, RI / Savings Plan, spot, rightsizing, idle 제거 5종. 보통 30-50% 절감 가능.

📖 핵심 개념

Tag: 자원에 메타 (env, team, project).
Cost allocation: tag 기반 청구.
Reserved / Savings: 1년+ 약정 = 30-70% 할인.
Spot: 대기 가능한 작업 = 70-90% 할인.

💻 코드 패턴

Tagging strategy

# Terraform — default tags
provider "aws" {
  default_tags {
    tags = {
      Environment = var.env
      Team        = var.team
      Project     = var.project
      CostCenter  = var.cost_center
      ManagedBy   = "terraform"
    }
  }
}

→ AWS Cost Explorer 가 tag 별 청구 분석.

IAM 강제 — 미태그 금지

{
  "Effect": "Deny",
  "Action": ["ec2:RunInstances", "rds:CreateDBInstance"],
  "Resource": "*",
  "Condition": {
    "Null": { "aws:RequestTag/Team": "true" }
  }
}

비용 모니터링

// AWS Cost Explorer API
const ce = new AWS.CostExplorer({ region: 'us-east-1' });
const r = await ce.getCostAndUsage({
  TimePeriod: { Start: '2026-05-01', End: '2026-05-09' },
  Granularity: 'DAILY',
  Metrics: ['UnblendedCost'],
  GroupBy: [{ Type: 'TAG', Key: 'Team' }],
}).promise();

# Slack alert if cost spike
- alert: DailyCostSpike
  expr: aws_cost_today > 2 * aws_cost_avg_7d

Reserved Instances / Savings Plans

On-demand:        $100/month
Reserved 1y:      $60 (40% off)
Reserved 3y:      $40 (60% off)
Savings Plan:     $50 (50% off, 더 유연)

→ 안정적 baseline = RI / SP. spike = on-demand.

Spot instances

resource "aws_instance" "worker" {
  instance_market_options {
    market_type = "spot"
    spot_options { max_price = "0.05" }
  }
}

→ 1-2분 알림 후 종료. Stateless / batch / autoscaling group.

# K8s — Karpenter
provisioner:
  spec:
    requirements:
      - { key: karpenter.sh/capacity-type, operator: In, values: [spot, on-demand] }

Rightsizing

# CloudWatch + Compute Optimizer
aws compute-optimizer get-ec2-instance-recommendations
# → "이 t3.large 는 t3.small 로 충분"

→ CPU / memory < 20% 사용 = 작게.

Idle 자원 (가장 흔한 낭비)

# 미사용 EBS volumes
aws ec2 describe-volumes --filters Name=status,Values=available

# Idle ELB (no traffic)
# Stopped EC2 (EBS 비용 그대로)
# Old snapshots
aws ec2 describe-snapshots --owner-ids self --query 'Snapshots[?StartTime<`2025-01-01`]'

# Unused EIP
aws ec2 describe-addresses --filters Name=association-id,Values=

# 매일 정리
aws ec2 delete-volume --volume-id $UNATTACHED

Scheduled scaling (dev)

# Dev 환경 = 9-18 만 켜기 (50% 절감)
resource "aws_autoscaling_schedule" "off" {
  scheduled_action_name = "off-evening"
  recurrence            = "0 18 * * 1-5"
  desired_capacity      = 0
  ...
}

resource "aws_autoscaling_schedule" "on" {
  scheduled_action_name = "on-morning"
  recurrence            = "0 9 * * 1-5"
  desired_capacity      = 2
}

Data transfer (숨은 비용)

Same-AZ: free
Cross-AZ: $0.01/GB
Cross-region: $0.02-0.09/GB
Internet egress: $0.05-0.09/GB

NAT gateway: $0.045/GB + $0.045/hour
→ VPC endpoint 로 S3 / DynamoDB 직접 (free)

S3 storage class

Standard:        $23/TB/mo
Intelligent-Tier: 자동 전환
Standard-IA:     $12.5/TB/mo (가끔 access)
Glacier:         $4/TB/mo (long backup)
Glacier Deep:    $1/TB/mo (rare)

resource "aws_s3_bucket_lifecycle_configuration" "main" {
  rule {
    status = "Enabled"
    transition { days = 30; storage_class = "STANDARD_IA" }
    transition { days = 90; storage_class = "GLACIER" }
  }
}

Budget alarm

resource "aws_budgets_budget" "monthly" {
  name         = "monthly"
  budget_type  = "COST"
  limit_amount = "10000"
  limit_unit   = "USD"
  time_unit    = "MONTHLY"
  
  notification {
    threshold = 80
    threshold_type = "PERCENTAGE"
    notification_type = "ACTUAL"
    subscriber_email_addresses = ["finance@acme.com"]
  }
}

LLM cost (위 AI_LLM_Cost_Optimization)

LLM = 새로운 cloud bill. 별도 추적.

🤔 의사결정 기준

절감 영역	우선순위
Idle 자원	즉시 (매주 cleanup)
Rightsizing	월별
Tag + visibility	즉시
RI / Savings	6개월 안정 후
Spot	Stateless 작업
Data transfer	VPC endpoint
S3 lifecycle	항상

❌ 안티패턴

Tag 없음: cost 누구 책임 모름.
모든 자원 on-demand: RI / SP 없으면 30-70% 더.
Dev 24h 켜둠: 70% 낭비.
Spot prod stateful: 강제 종료 시 데이터 잃음.
Snapshot / EBS / EIP 청소 X: 매월 누적.
Cross-AZ 무절제: $$$/GB.
Cost monitoring 없음: 청구서 보고 놀람.

🤖 LLM 활용 힌트

Tag → visibility → 액션.
Idle 청소가 가장 ROI.
Spot + Karpenter 자동.

5.4 KiB Raw Blame History