预留实例、Savings Plans 与 Spot

EC2 购买模式概览

# EC2 pricing models (m5.large in us-east-1, approximate):
# On-Demand:        $0.096/hr ($70/month)
# 1yr RI (No Upf): $0.062/hr ($45/month) - 35% savings
# 1yr RI (All Upf): $0.051/hr ($37/month) - 47% savings
# 3yr RI (All Upf): $0.032/hr ($23/month) - 67% savings
# Spot:            $0.030-0.050/hr       - 50-70% savings

# Rule: RI/SP for steady state, Spot for burst/interruptible

Reserved Instances：类型与付款选项

# Purchase Reserved Instance
aws ec2 purchase-reserved-instances-offering \
  --reserved-instances-offering-id offering-id-here \
  --instance-count 5

# List RI offerings
aws ec2 describe-reserved-instances-offerings \
  --instance-type m5.large \
  --product-description 'Linux/UNIX' \
  --offering-type 'No Upfront' \
  --duration 31536000 \
  --query 'ReservedInstancesOfferings[].{Price:RecurringCharges[0].Amount,Class:OfferingClass}'

RI 范围：Regional 与 Zonal

# Regional RI: flexible across sizes in a family
# 1x m5 regional RI covers:
# - m5.large (1.0 units)
# - 2x m5.medium (0.5 units each)
# - 0.5x m5.xlarge (2.0 units)

# Zonal RI: specific AZ, reserves capacity
# 1x m5.large in us-east-1a:
# - Only covers m5.large in us-east-1a
# - Guarantees capacity is available

# Most use cases: prefer REGIONAL for flexibility

Savings Plans：比 RIs 更灵活

# Purchase Compute Savings Plan
aws savingsplans create-savings-plan \
  --savings-plan-offering-id offering-id \
  --commitment 10.00 \
  --purchase-time '2026-07-01T00:00:00Z'

# $10/hr commitment covers:
# - EC2 instances (any family, region, OS)
# - Lambda invocations and duration
# - Fargate vCPU and memory

# Savings Plans apply AFTER Reserved Instances
# Check coverage in Cost Explorer > Savings Plans

如何在 RI 和 Savings Plans 之间进行选择

# Savings Plans coverage analysis
aws ce get-savings-plans-coverage \
  --time-period Start=2026-05-01,End=2026-06-01 \
  --granularity MONTHLY

# Response shows:
# OnDemandCost: $5,000 (not covered)
# SpendCoveredBySavingsPlans: $10,000
# CoveragePercentage: 66%

# Recommendation:
# If coverage < 60%, consider purchasing more
# If coverage > 85%, you may be over-committed

Spot 实例：大幅折扣与中断风险

# Check Spot instance interruption notice
# From inside EC2 instance via Instance Metadata Service
curl http://169.254.169.254/latest/meta-data/spot/termination-time
# Returns: 2026-06-21T15:00:00Z if interruption is coming
# Returns 404 if no interruption

# Handle in your application:
# 1. Poll this endpoint every 5 seconds
# 2. On 2-min notice: checkpoint work to S3
# 3. Terminate gracefully
# 4. AWS terminates the instance 2 minutes after notice

Spot 实例策略与多样化配置

# ASG with diversified Spot pools
aws autoscaling create-auto-scaling-group \
  --mixed-instances-policy '{
    "LaunchTemplate": {
      "LaunchTemplateSpecification": {
        "LaunchTemplateId": "lt-12345",
        "Version": "$Latest"
      },
      "Overrides": [
        {"InstanceType": "m5.large"},
        {"InstanceType": "m5a.large"},
        {"InstanceType": "m4.large"},
        {"InstanceType": "c5.large"},
        {"InstanceType": "c5a.large"}
      ]
    },
    "InstancesDistribution": {
      "SpotAllocationStrategy": "capacity-optimized",
      "OnDemandPercentageAboveBaseCapacity": 0
    }
  }'

用于批处理和数据处理的 Spot

# EMR cluster with On-Demand master + Spot workers
aws emr create-cluster \
  --instance-fleets \
    InstanceFleetType=MASTER,TargetOnDemandCapacity=1,InstanceTypeConfigs=[{InstanceType=m5.xlarge}] \
    InstanceFleetType=CORE,TargetSpotCapacity=10,InstanceTypeConfigs=[{InstanceType=m5.xlarge},{InstanceType=m5a.xlarge}],LaunchSpecifications={SpotSpecification={TimeoutAction=SWITCH_TO_ON_DEMAND,TimeoutDurationMinutes=10}}

# SWITCH_TO_ON_DEMAND: if Spot unavailable after 10 min,
# provision On-Demand instead (ensures job completes)

组合使用购买模式

# Hybrid ASG: On-Demand base + Spot burst
# Savings Plan covers On-Demand baseline cost
aws autoscaling create-auto-scaling-group \
  --mixed-instances-policy '{
    "LaunchTemplate": { ... },
    "InstancesDistribution": {
      "OnDemandBaseCapacity": 4,
      "OnDemandPercentageAboveBaseCapacity": 0,
      "SpotAllocationStrategy": "capacity-optimized"
    }
  }' \
  --min-size 4 --max-size 40 --desired-capacity 8

# 4 On-Demand instances (covered by Savings Plan)
# Additional instances = 100% Spot (cheapest burst)

监控承诺使用率

# Check Savings Plans utilisation
aws ce get-savings-plans-utilization \
  --time-period Start=2026-06-01,End=2026-06-21 \
  --granularity DAILY \
  --query 'SavingsPlansUtilizationsByTime[].{Date:TimePeriod.Start,Utilization:Utilization.UtilizationPercentage}'

# Also check: Cost Explorer > Savings Plans > Utilization report
# Target: >80% utilisation
# If <80%: you over-committed or workload shrank

专用主机与专用实例

# Allocate Dedicated Host for Oracle licensing
aws ec2 allocate-hosts \
  --quantity 1 \
  --instance-type m5.xlarge \
  --availability-zone us-east-1a \
  --auto-placement on \
  --host-recovery on

# Dedicated Host pricing is per host per hour
# Dedicated Instances: $2/region/hr + instance price

# Use Dedicated Host when:
# - Oracle/SQL Server socket-based licensing
# - Windows Server with own license
# - Compliance requires physical server isolation

快速检查

课程回顾