Skip to main content
Amazon EKS 0.65.5Last updated in version 0.65.0

K8S Cluster Autoscaler Module

View SourceRelease Notes

This Terraform Module installs a Cluster Autoscaler to automatically scale up and down the nodes in a cluster in response to resource utilization.

This module is responsible for manipulating each Auto Scaling Group (ASG) that was created by the EKS cluster workers module. By default, the ASG is configured to allow zero-downtime deployments but is not configured to scale automatically. You must launch an EKS control plane with worker nodes for this module to function.

Important Considerations

  • The autoscaler doesn't account for CPU or Memory usage in deciding to scale up, it scales up when Pods fail to schedule due to insufficient resources. This means it's important to carefully the manage the compute resources you assign to your deployments. See the Kubernetes documentation on compute resources for more information.
  • Scaling down happens when utilization dips below a specified threshold and there are pods that are able to be moved to another node. There are a variety of conditions to be aware of that can prevent pods from being automatically removable which can result in wasted capacity.

How do I deploy the Pods to Fargate?

To deploy the Pods to Fargate, you can use the create_fargate_profile variable to true and specify the subnet IDs for Fargate using vpc_worker_subnet_ids. Note that if you are using Fargate, you must rely on the IAM Roles for Service Accounts (IRSA) feature to grant the necessary AWS IAM permissions to the Pod. This is configured using the use_iam_role_for_service_accounts, eks_openid_connect_provider_arn, and eks_openid_connect_provider_url input variables.

Sample Usage

# ------------------------------------------------------------------------------------------------------
# ------------------------------------------------------------------------------------------------------

module "eks_k_8_s_cluster_autoscaler" {

source = ""

# ----------------------------------------------------------------------------------------------------
# ----------------------------------------------------------------------------------------------------

# The AWS region that the EKS cluster resides in.
aws_region = <string>

# The name of the EKS cluster (e.g. eks-prod). This is used to assist with
# auto-discovery of the cluster workers ASG.
eks_cluster_name = <string>

# Configuration for using the IAM role with Service Accounts feature to
# provide permissions to the helm charts. This expects a map with two
# properties: `openid_connect_provider_arn` and `openid_connect_provider_url`.
# The `openid_connect_provider_arn` is the ARN of the OpenID Connect Provider
# for EKS to retrieve IAM credentials, while `openid_connect_provider_url` is
# the URL. Set to null if you do not wish to use IAM role with Service
# Accounts.
iam_role_for_service_accounts_config = <object(
openid_connect_provider_arn = string
openid_connect_provider_url = string

# ----------------------------------------------------------------------------------------------------
# ----------------------------------------------------------------------------------------------------

# ARN of permissions boundary to apply to the autoscaler IAM role - the IAM
# role created for the Autoscaler
autoscaler_iam_role_permissions_boundary = null

# The AWS partition used for default AWS Resources.
aws_partition = "aws"

# Restrict the cluster autoscaler to a list of absolute ASG ARNs upon initial
# apply to ensure no new ASGs can be managed by the autoscaler without
# explicitly running another apply. Setting this to false will ensure that the
# cluster autoscaler is automatically given access to manage any new ASGs with
# the tag applied.
cluster_autoscaler_absolute_arns = true

# The version of the cluster-autoscaler helm chart to deploy. Note that this
# is different from the app/container version, which is sepecified with
# var.cluster_autoscaler_version.
cluster_autoscaler_chart_version = "9.21.0"

# Which docker repository to use to install the cluster autoscaler. Check the
# following link for valid repositories to use
cluster_autoscaler_repository = ""

# Which version of the cluster autoscaler to install.
cluster_autoscaler_version = "v1.28.0"

# Map of extra arguments to pass to the container.
container_extra_args = {}

# When set to true, create a dedicated Fargate execution role for the cluster
# autoscaler. When false, you must provide an existing fargate execution role
# in the variable var.pod_execution_iam_role_arn. Only used if
# var.create_fargate_profile is true.
create_fargate_execution_role = false

# When set to true, create a dedicated Fargate execution profile for the
# cluster autoscaler.
create_fargate_profile = false

# Create a dependency between the resources in this module to the interpolated
# values in this list (and thus the source resources). In other words, the
# resources in this module will now depend on the resources backing the values
# in this list such that those resources need to be created before the
# resources in this module, and the resources in this module need to be
# destroyed before the resources in the list.
dependencies = []

# If scaling_strategy is set to 'priority', you can use this variable to
# define cluster-autoscaler-priority-expander priorities. See:
expander_priorities = {}

# Which Kubernetes Namespace to deploy the chart into.
namespace = "kube-system"

# Annotations to apply to the Pod that is deployed, as key value pairs.
pod_annotations = {}

# ARN of IAM Role to use as the Pod execution role for Fargate. Set to null
# (default) to create a new one. Only used when var.create_fargate_profile is
# true.
pod_execution_iam_role_arn = null

# Labels to apply to the Pod that is deployed, as key value pairs.
pod_labels = {}

# Configure affinity rules for the Pod to control which nodes to schedule on.
# Each item in the list should be a map with the keys `key`, `values`, and
# `operator`, corresponding to the 3 properties of matchExpressions. Note that
# all expressions must be satisfied to schedule on the node.
pod_node_affinity = []

# Configure priorityClassName of pods to allow scheduler to order pending pods
# by their priority.
pod_priority_class_name = null

# Number of replicas of the cluster autoscaler Pod to deploy.
pod_replica_count = 1

# Pod resource requests and limits to use. Refer to
# for more information.
pod_resources = null

# Configure tolerations rules to allow the Pod to schedule on nodes that have
# been tainted. Each item in the list specifies a toleration rule.
pod_tolerations = []

# If scaling_strategy is set to 'priority', you can use this to specify
# annotations to add to the cluster-autoscaler-priority-expander ConfigMap.
priority_config_map_annotations = {}

# The name of the helm release to use. Using different release names are
# useful for deploying different copies of the cluster autoscaler.
release_name = "cluster-autoscaler"

# Specifies an 'expander' for the cluster autoscaler. This helps determine
# which ASG to scale when additional resource capacity is needed.
scaling_strategy = "least-waste"

# A list of the subnets into which the EKS Cluster's administrative pods will
# be launched. These should usually be all private subnets and include one in
# each AWS Availability Zone. Required when var.create_fargate_profile is
# true.
vpc_worker_subnet_ids = []