Remove from Amazon EKS - Poolside

Overview

These steps remove the inference Helm release, its Kubernetes resources, and the AWS artifacts used by a model inference deployment on Amazon EKS. The removal process includes the following phases:

Uninstall the inference release: Uninstall the inference Helm release, which also tears down the Application Load Balancer.
Delete cluster resources: Delete the poolside-models namespace, which removes the remaining workloads, secrets, ConfigMaps, and Ingress objects.
Clean up AWS artifacts: Remove the IRSA role, ECR repository, S3 checkpoints, ACM certificate, and DNS records.

Prerequisites

These instructions assume that you deployed model inference using Install on Amazon EKS. Before you start, make sure you have:

Cluster administrator access to the EKS cluster
helm 3.12 or later
kubectl, configured for your EKS cluster
aws CLI, with permission to delete the AWS resources you provisioned

Step 1: Uninstall the inference release

List the Helm releases in the poolside-models namespace:

helm list -n poolside-models

Uninstall the release. This deletes the per-model Ingress objects, which prompts the AWS Load Balancer Controller to delete the shared Application Load Balancer:

helm uninstall inference -n poolside-models

Wait for the inference pods to stop. Continue when kubectl get pods -n poolside-models shows no Poolside pods remaining:

kubectl get pods -n poolside-models

Confirm the load balancer is gone before you continue:

kubectl get ingress -n poolside-models

Step 2: Delete cluster resources

Deleting the namespace removes the remaining workloads, secrets, ConfigMaps, and any Ingress objects that the inference release created:

kubectl delete namespace poolside-models

Step 3: Clean up AWS artifacts

Clean up the following AWS resources separately. Remove only the resources that you no longer need.

IRSA role
Delete the inference-pod IAM role and its permissions policy. If you created them with eksctl create iamserviceaccount, delete them with eksctl:

eksctl delete iamserviceaccount \
  --cluster <cluster-name> \
  --namespace poolside-models \
  --name inference

If you created the role and policy directly, detach the policy and delete both with the aws iam commands you use for IAM cleanup.

Amazon ECR
Delete the atlas repository if no other deployment uses it:

aws ecr delete-repository --repository-name atlas --force --region <aws-region>

Amazon S3
Delete the model checkpoints from the bucket you referenced in models.<key>.model in the inference_values.yaml file you used to install:

aws s3 rm s3://<bucket-name>/checkpoints --recursive --region <aws-region>

If the bucket only stored Poolside checkpoints, you can delete the bucket itself.

TLS certificate and DNS records
Remove the DNS records that pointed to the model hostnames. If the ACM certificate covered only those hostnames, delete it:

aws acm delete-certificate --certificate-arn <acm-certificate-arn> --region <aws-region>

Local files
Delete the extracted Helm bundle directory and your inference_values.yaml file from your workstation.