Spot Instances and Interruption Management

EC2 Spot Instances are spare compute capacity in the AWS cloud offered at a significant discount — typically 60-90% off On-Demand prices. AWS sells this unused capacity through a bidding mechanism. You specify the maximum price you are willing to pay per instance-hour (your bid), and your instance runs as long as the current Spot price is below your bid and capacity is available. When the Spot price exceeds your bid or capacity is reclaimed, AWS terminates or stops your instance with a 2-minute warning.

Spot Instances are ideal for fault-tolerant, flexible, and stateless workloads: big data analytics, containerized applications, CI/CD pipelines, web servers (with auto-scaling), image and media rendering, high-performance computing (HPC), and test/development environments. They are NOT suitable for stateful, time-sensitive, or single-instance workloads like databases, long-running batch jobs that cannot checkpoint, or applications that require constant availability.

When you launch a Spot Instance, you do not get dedicated hardware. Instead, AWS manages a pool of spare EC2 capacity across Availability Zones and instance types. The Spot price is determined by supply and demand for that capacity pool. AWS updates the Spot price every hour (or more frequently) based on current utilization. The price can fluctuate, but in practice, it is relatively stable except during spikes.

Your request can be in one of three forms: a one-time request (for a specific number of instances), a persistent request (maintains the desired count, re-launching instances if they are terminated), or a Spot Fleet request (a collection of instance types and pools). For the SOA-C02 exam, focus on persistent requests and Spot Fleets.

When your instance is marked for interruption, AWS sends a notification via two mechanisms: - Instance Metadata: The events/maintenance/scheduled endpoint in the instance metadata (http://169.254.169.254/latest/meta-data/events/maintenance/scheduled) shows an event with code instance-rebalance or instance-stop or instance-terminate. The action field includes stop, terminate, or hibernate. - CloudWatch Events: A EC2 Spot Instance Interruption Warning event is emitted approximately 2 minutes before the action.

You can also enable Spot Instance rebalance recommendation (newer feature) which gives a 2-minute notice before interruption, allowing you to proactively rebalance your workload.

When AWS needs to reclaim capacity, the interruption behavior can be: - Terminate: The instance is stopped (if configured) or terminated. By default, the instance is terminated. Root EBS volumes are deleted (unless DeleteOnTermination is set to false). - Stop: The instance is stopped (not terminated) — only available for instances backed by Amazon EBS. When stopped, you can start it later if capacity and price are favorable. - Hibernate: The instance is hibernated — RAM contents are saved to the EBS root volume, and on start, the instance resumes with the same state. This requires an EBS-backed instance with hibernation enabled at launch.

You can set the interruption behavior when launching a Spot Instance. For persistent requests, you typically use stop or hibernate to preserve state and reduce relaunch times.

A Spot Instance request goes through these states: - open: The request is active and waiting for fulfillment. - active: The request is fulfilled and the instance is running. - closed: The request is closed (you canceled it, or the instance was terminated and the request is not persistent). - cancelled: You canceled the request. - disabled: The request is disabled (e.g., due to an invalid parameter).

For persistent requests, if the instance is interrupted, the request goes back to open and AWS tries to relaunch.

You can use Spot Instances with Auto Scaling Groups (ASGs). The ASG can launch Spot Instances by specifying a mixed instances policy or by using a launch template with Spot options. When Spot Instances are interrupted, the ASG can automatically replace them if the desired capacity is not met. This is a common pattern for cost-optimized, fault-tolerant architectures.

To launch a Spot Instance via AWS CLI:

aws ec2 request-spot-instances \
    --spot-price "0.05" \
    --instance-count 1 \
    --type "persistent" \
    --launch-specification file://spec.json

Example spec.json:

{
  "ImageId": "ami-0abcdef1234567890",
  "InstanceType": "t3.medium",
  "KeyName": "my-key",
  "SecurityGroups": ["sg-12345678"],
  "BlockDeviceMappings": [
    {
      "DeviceName": "/dev/xvda",
      "Ebs": {
        "VolumeSize": 20,
        "DeleteOnTermination": true
      }
    }
  ]
}

To check Spot price history:

aws ec2 describe-spot-price-history \
    --instance-types t3.medium \
    --product-description "Linux/UNIX" \
    --start-time 2025-01-01T00:00:00Z