Skip to main content
The terraform-aws-eks module supports four distinct compute strategies for running workloads on Amazon EKS. Each option offers different trade-offs in control, operational overhead, and cost.

​
Compute options

EKS Auto Mode

Fully managed compute with automatic node provisioning, scaling, and lifecycle management. AWS manages the underlying EC2 fleet.

EKS Managed Node Groups

AWS-managed Auto Scaling Groups where EKS handles node provisioning and lifecycle, but you retain control over instance configuration via launch templates.

Self-Managed Node Groups

Fully self-managed Auto Scaling Groups. You control every aspect of the node lifecycle, launch template, and bootstrap process.

Fargate Profiles

Serverless compute for pods. No nodes to manage — AWS provisions isolated compute for each pod matching your selectors.

​
Comparison

FeatureEKS Auto ModeManaged Node GroupsSelf-ManagedFargate
Node managementAWSAWSYouAWS (serverless)
Launch templateNot applicableCustom (default on)Full controlNot applicable
Custom AMINoYesYesNo
Spot / On-Demand mixAutomaticVia capacity typeVia mixed instances policyNo
EFA supportNoYesYesNo
Bottlerocket OSNoYesYesNo
Windows nodesNoYesYesNo
Bin packing / cost optimizationAutomaticManualManualPer-pod billing
Best forSimplicity, managed operationsStandard workloads needing customizationAdvanced control or custom AMIsBatch, burst, or isolation workloads

​
Default node group configurations

All node group types inherit cluster-level defaults unless overridden at the node group level. The module waits 30s after the EKS cluster becomes active before creating any node groups (configurable via dataplane_wait_duration). For eks_managed_node_groups and self_managed_node_groups, when no values are specified:
  • min_size defaults to 1
  • max_size defaults to 3
  • desired_size defaults to 1
  • ami_type defaults to AL2023_x86_64_STANDARD
  • instance_type / instance_types defaults to t3.medium (managed) or m6i.large (self-managed)
The desired_size value is only respected on initial creation. Subsequent changes are ignored to avoid conflicts with the cluster autoscaler or Karpenter managing the desired count externally.

​
Combining compute types

The module supports running multiple compute types simultaneously in the same cluster. A common pattern is using Fargate for system components while managed node groups serve application workloads, or enabling EKS Auto Mode alongside a self-managed node group for GPU workloads that require specific drivers. 
module "eks" {
  source  = "terraform-aws-modules/eks/aws"
  version = "~> 21.0"

  name               = "my-cluster"
  kubernetes_version = "1.33"

  # Fargate for system workloads
  fargate_profiles = {
    kube-system = {
      selectors = [
        { namespace = "kube-system" }
      ]
    }
  }

  # Managed node groups for application workloads
  eks_managed_node_groups = {
    app = {
      ami_type       = "AL2023_x86_64_STANDARD"
      instance_types = ["m6i.large"]
      min_size       = 2
      max_size       = 10
      desired_size   = 2
    }
  }

  vpc_id     = "vpc-1234556abcdef"
  subnet_ids = ["subnet-abcde012", "subnet-bcde012a", "subnet-fghi345a"]

  tags = {
    Environment = "production"
    Terraform   = "true"
  }
}

Build docs developers (and LLMs) love

Get started for freeTalk to us