Spot instances¶

Managed Nodegroups¶

eksctl supports Spot worker nodes using EKS Managed Nodegroups, a feature that allows EKS customers with fault-tolerant applications to easily provision and manage EC2 Spot Instances for their EKS clusters. EKS Managed Nodegroup will configure and launch an EC2 Autoscaling group of Spot Instances following Spot best practices and draining Spot worker nodes automatically before the instances are interrupted by AWS. There is no incremental charge to use this feature and customers pay only for using the AWS resources, such as EC2 Spot Instances and EBS volumes.

To create a cluster with a managed nodegroup using Spot instances, pass the --spot flag and an optional list of instance types:

$ eksctl create cluster --spot --instance-types=c3.large,c4.large,c5.large

To create a managed nodegroup using Spot instances on an existing cluster:

$ eksctl create nodegroup --cluster=<clusterName> --spot --instance-types=c3.large,c4.large,c5.large

To create Spot instances using managed nodegroups via a config file:

# spot-cluster.yaml

apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: spot-cluster
  region: us-west-2

managedNodeGroups:
- name: spot
  instanceTypes: ["c3.large","c4.large","c5.large","c5d.large","c5n.large","c5a.large"]
  spot: true


# `instanceTypes` defaults to [`m5.large`]
- name: spot-2
  spot: true

# On-Demand instances
- name: on-demand
  instanceTypes: ["c3.large", "c4.large", "c5.large"]

$ eksctl create cluster -f spot-cluster.yaml

Note

Unmanaged nodegroups do not support the spot and instanceTypes fields, instead the instancesDistribution field is used to configure Spot instances. See below

Further information¶

Unmanaged Nodegroups¶

eksctl has support for spot instances through the MixedInstancesPolicy for Auto Scaling Groups.

Here is an example of a nodegroup that uses 50% spot instances and 50% on demand instances:

nodeGroups:
  - name: ng-1
    minSize: 2
    maxSize: 5
    instancesDistribution:
      maxPrice: 0.017
      instanceTypes: ["t3.small", "t3.medium"] # At least one instance type should be specified
      onDemandBaseCapacity: 0
      onDemandPercentageAboveBaseCapacity: 50
      spotInstancePools: 2

Note that the nodeGroups.X.instanceType field shouldn't be set when using the instancesDistribution field.

This example uses GPU instances:

nodeGroups:
  - name: ng-gpu
    instanceType: mixed
    desiredCapacity: 1
    instancesDistribution:
      instanceTypes:
        - p2.xlarge
        - p2.8xlarge
        - p2.16xlarge
      maxPrice: 0.50

This example uses the capacity-optimized spot allocation strategy:

nodeGroups:
  - name: ng-capacity-optimized
    minSize: 2
    maxSize: 5
    instancesDistribution:
      maxPrice: 0.017
      instanceTypes: ["t3.small", "t3.medium"] # At least one instance type should be specified
      onDemandBaseCapacity: 0
      onDemandPercentageAboveBaseCapacity: 50
      spotAllocationStrategy: "capacity-optimized"

This example uses the capacity-optimized-prioritized spot allocation strategy:

nodeGroups:
  - name: ng-capacity-optimized-prioritized
    minSize: 2
    maxSize: 5
    instancesDistribution:
      maxPrice: 0.017
      instanceTypes: ["t3a.small", "t3.small"] # At least two instance types should be specified
      onDemandBaseCapacity: 0
      onDemandPercentageAboveBaseCapacity: 0
      spotAllocationStrategy: "capacity-optimized-prioritized"

Use the capacity-optimized-prioritized allocation strategy and then set the order of instance types in the list of launch template overrides from highest to lowest priority (first to last in the list). Amazon EC2 Auto Scaling honors the instance type priorities on a best-effort basis but optimizes for capacity first. This is a good option for workloads where the possibility of disruption must be minimized, but also the preference for certain instance types matters.

Note that the spotInstancePools field shouldn't be set when using the spotAllocationStrategy field. If the spotAllocationStrategy is not specified, EC2 will default to use the lowest-price strategy.

Here is a minimal example:

nodeGroups:
  - name: ng-1
    instancesDistribution:
      instanceTypes: ["t3.small", "t3.medium"] # At least one instance type should be specified

To distinguish nodes between spot or on-demand instances you can use the kubernetes label node-lifecycle which will have the value spot or on-demand depending on its type.

Parameters in instancesDistribution¶

Please see the config parameters for details.