How to Use Spot Instances on AWS

Spot vs. On-Demand vs. Reserved Instances

Understanding the difference between AWS pricing models is crucial for optimization. On-Demand Instances offer predictable pricing with no commitments—you pay by the hour with no interruptions, making them ideal for steady-state workloads. Reserved Instances require 1-3 year commitments but offer 30-70% savings for predictable, always-on applications like databases and enterprise systems.

Spot Instances provide the deepest discounts (50-90%) but can be interrupted, making them perfect for flexible, stateless workloads. The smart strategy? Use Reserved Instances for your baseline capacity, On-Demand for predictable peaks, and Spot Instances for burst capacity and batch processing. This hybrid approach—often called a "spot-first" strategy—maximizes savings while maintaining application reliability. Many organizations now run 50-80% of their non-critical workloads on Spot Instances, dramatically reducing their AWS bills.

Exploring AWS EC2 Pricing Options

For detailed pricing information across all instance types and regions, visit the official AWS EC2 Pricing page to compare costs and calculate potential savings.

Managing Interruptions with Two-Minute Warnings

AWS provides a two-minute warning before terminating a Spot Instance, giving your application time to gracefully shut down. This interruption notice is delivered via the EC2 instance metadata service and CloudWatch Events, allowing you to implement automated handling.

Best practices include: continuously polling the instance metadata endpoint for termination notices, implementing graceful shutdown procedures that save state and checkpoint work, using AWS Auto Scaling groups with multiple instance types to automatically launch replacement capacity, and leveraging Spot Instance interruption notices with AWS Lambda for automated failover. Modern orchestration platforms like Kubernetes and ECS handle these interruptions automatically by draining nodes and rescheduling containers. For batch jobs, implement checkpointing every few minutes so jobs can resume from the last saved state rather than starting over.

# Poll metadata endpoint for termination notice
TOKEN=$(curl -X PUT "http://169.254.169.254/latest/api/token" \
  -H "X-aws-ec2-metadata-token-ttl-seconds: 21600")

TERMINATION=$(curl -H "X-aws-ec2-metadata-token: $TOKEN" \
  http://169.254.169.254/latest/meta-data/spot/instance-action)

if [ ! -z "$TERMINATION" ]; then
  echo "Spot termination notice received. Saving state..."
  # Implement graceful shutdown logic
  save_checkpoint()
  drain_tasks()
  exit 0
fi

Diversifying Instance Types and Availability Zones

The secret to Spot Instance reliability is diversification—never depend on a single instance type or availability zone. AWS Spot capacity varies by instance family, size, and region, so spreading your requests across multiple options dramatically reduces interruption rates.

Use Spot Fleet or EC2 Auto Scaling to automatically select from 10-15 different instance types with similar performance characteristics. For example, if you need 4 vCPUs and 16GB RAM, configure your fleet to accept m5.xlarge, m5a.xlarge, m5n.xlarge, m4.xlarge, and similar instances across all availability zones. Enable the "capacity-optimized" allocation strategy, which automatically launches instances in the pools with the most available capacity, minimizing interruptions.

This diversification approach, combined with automatic replacement, can reduce Spot interruption rates to less than 5%, making them reliable enough for most production workloads. The result: consistent capacity at a fraction of On-Demand costs.

Automating with Capacity-Optimized Allocation

The capacity-optimized allocation strategy is a game-changer for Spot Instance reliability. Unlike lowest-price allocation, which simply picks the cheapest instances (often the most volatile), capacity-optimized uses AWS's real-time capacity data to launch instances in pools with the deepest available capacity. This dramatically reduces interruption rates—in many cases below 5%—while still delivering 70-90% savings.

To automate Spot adoption at scale, use infrastructure-as-code tools like Terraform or AWS CloudFormation with EC2 Auto Scaling groups. Define launch templates with multiple instance types, enable capacity-optimized-prioritized allocation to favor your preferred instance types while maintaining flexibility, and configure CloudWatch alarms to monitor Spot interruption rates.

resource "aws_spot_fleet_request" "app_fleet" {
  allocation_strategy      = "capacityOptimized"
  target_capacity         = 10
  valid_until             = "2026-12-31T00:00:00Z"
  terminate_instances     = true
  
  launch_specification {
    ami                    = "ami-0c55b159cbfafe1f0"
    instance_type          = "m5.large"
    availability_zone      = "us-east-1a"
    vpc_security_group_ids = [aws_security_group.app.id]
  }
  
  launch_specification {
    ami                    = "ami-0c55b159cbfafe1f0"
    instance_type          = "m5a.large"
    availability_zone      = "us-east-1b"
    vpc_security_group_ids = [aws_security_group.app.id]
  }
  
  # Add more instance types for diversification
}

For containerized workloads, Kubernetes cluster autoscaler with AWS Node Termination Handler automatically manages Spot lifecycle events. The result is a fully automated, self-healing infrastructure that continuously optimizes for both cost and reliability, adapting to changing capacity conditions without manual intervention.