Automating Node Management with Exostellar Karpenter

This Karpenter is an enhanced version of the open-source Karpenter, tailored to integrate seamlessly with our scheduler and core technology platform.

Prerequisites

Helm: Version 3.14+

Installation Steps

Uninstalling Exostellar Karpenter (v1.x)

If version v1.x of Exostellar Karpenter is installed in your cluster, follow the steps below to uninstall it cleanly:

Delete existing exokarpenter CRDs:

BASH

kubectl delete crds exonodeclaims.exokarpenter.sh \
                   exonodeclasses.karpenter.k8s.exo \
                   exonodepools.exokarpenter.sh

Verify exokarpenter CRDs' deletion:

BASH

kubectl get crds

Force-remove finalizers (if CRDs are stuck in Terminating state):

BASH

kubectl patch crd exonodeclasses.karpenter.k8s.exo -p '{"metadata":{"finalizers":[]}}' --type=merge

Repeat this command for any other CRDs that are stuck due to finalizers.

Uninstall the Helm chart

BASH

helm uninstall xkarpenter  --namespace exokarpenter

Installing Exostellar Karpenter (v2.x)

Replace the HEADNODE variable with the private IP address of your EMS head node.

BASH

cat <<'EOF' | bash
#!/bin/bash
set -euo pipefail

# Define variables
HEADNODE="http://10.0.186.213:5000"  # Replace head node private address
CLUSTER_ENDPOINT=$(kubectl config view --minify -o jsonpath='{.clusters[].cluster.server}')

# Retrieve AWS account ID if not already set in the environment
if [ -z "${AWS_ACCOUNT_ID-}" ]; then
  AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
fi

if [ -z "${CLUSTER_NAME:-}" ]; then
  CLUSTER_NAME=$(kubectl config view --minify -o jsonpath='{.clusters[].name}' | rev | cut -d'/' -f1 | rev | cut -d'.' -f1)
fi

echo "Cluster name is: $CLUSTER_NAME"

# Install helm chart using variables
helm install xkarpenter oci://public.ecr.aws/u8h5n6o4/exostellar-karpenter/karpenter \
  --version v2.0.1 \
  --namespace exokarpenter \
  --set settings.clusterName="$CLUSTER_NAME" \
  --set settings.clusterEndpoint="$CLUSTER_ENDPOINT" \
  --set settings.featureGates.drift=true \
  --set controller.resources.requests.cpu=1 \
  --set controller.resources.requests.memory=1Gi \
  --set controller.resources.limits.cpu=1 \
  --set controller.resources.limits.memory=1Gi \
  --set "headnode=$HEADNODE" \
  --set defaultControllerRole="arn:aws:iam::${AWS_ACCOUNT_ID}:instance-profile/"$CLUSTER_NAME"-xio-controller" \
  --set defaultWorkerRole="arn:aws:iam::${AWS_ACCOUNT_ID}:instance-profile/"$CLUSTER_NAME"-xio-worker" \
  --create-namespace
EOF

Step 1: Setting Environment Variables

Set the necessary environment variables to match your system configuration:

BASH

export HEADNODE="http://192.0.7.xx:5000" #Please use the Private IPv4 address
export CLUSTER_NAME="EksClusterName"

export CONTROLLER_ROLE="arn:aws:iam::${AWS_ACCOUNT_ID}:instance-profile/"$CLUSTER_NAME"-xio-controller" # use controller's IAM instance profile that you or x-install created
export WORKER_ROLE="arn:aws:iam::${AWS_ACCOUNT_ID}:instance-profile/"$CLUSTER_NAME"-xio-worker" #use worker IAM instance profile that you or x-install created

Step 2: Installing the Helm Chart

Execute the Helm chart installation with the command below:

BASH

helm upgrade --install xkarpenter oci://public.ecr.aws/u8h5n6o4/exostellar-karpenter/karpenter \
   --version v2.0.1 \
   --namespace exokarpenter \
   --create-namespace \
   --set "settings.clusterName=${CLUSTER_NAME}" \
   --set controller.resources.requests.cpu=1 \
   --set controller.resources.requests.memory=1Gi \
   --set controller.resources.limits.cpu=1 \
   --set controller.resources.limits.memory=1Gi \
   --set headnode=$HEADNODE \
   --set defaultControllerRole=$CONTROLLER_ROLE \
   --set defaultWorkerRole=$WORKER_ROLE \
   --wait