b. Create EKS cluster

In this section, you will create a new Amazon EKS cluster. Prior to the start, we can ensure our environment variables are set by executing:

echo export EKS_CLUSTER_NAME=eks-hpc >> env_vars
source env_vars

export AWS_REGION=${AWS_REGION:-us-east-1}

# Select two random availability zones for the cluster
export AZ_IDS=(use1-az4 use1-az5 use1-az6)
echo "AZ_IDS=(${AZ_IDS[@]})"
export AZ_COUNT=${#AZ_IDS[@]}
export AZ_IND=($(python3 -S -c "import random; az_ind=random.sample(range(${AZ_COUNT}),2); print(*az_ind)"))
echo "AZ_IND=(${AZ_IND[@]})"
export AZ1_NAME=$(aws ec2 describe-availability-zones --region ${AWS_REGION} --query "AvailabilityZones[?ZoneId == '${AZ_IDS[${AZ_IND[0]}]}'].ZoneName" --output text)
echo "AZ1_NAME=${AZ1_NAME}"
export AZ2_NAME=$(aws ec2 describe-availability-zones --region ${AWS_REGION} --query "AvailabilityZones[?ZoneId == '${AZ_IDS[${AZ_IND[1]}]}'].ZoneName" --output text)
echo "AZ2_NAME=${AZ2_NAME}"

export IMAGE_URI=$(aws ecr --region ${AWS_REGION} describe-repositories --repository-name sc22-container --query "repositories[0].repositoryUri" --output text)

Please note that IMAGE_URI is required. If the value of IMAGE_URI above is blank, then please go back to Lab 3 and make sure that the container image is successfully pushed to ECR.

1. Create the EKS manifest file

The EKS cluster manifest specifies the version of Kubernetes to deploy, the AWS region as well as the associated availability zones to use, the instance type and network settings. For this lab, you will mainly focus on those parameters but many more parameters can be specified in the EKS manifest. In this example, you will create an Amazon EKS cluster with one managed node group.

Let’s create the manifest file by pasting the following commands into the Cloud9 terminal:

cat > ~/environment/eks-hpc.yaml << EOF
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

  version: "1.21"
  region: ${AWS_REGION}

  - ${AZ1_NAME}
  - ${AZ2_NAME}

  withOIDC: true

  - name: hpc
    instanceType: c5.24xlarge
    instancePrefix: hpc
    privateNetworking: true
    availabilityZones: ["${AZ1_NAME}"]
    efaEnabled: false
    minSize: 0
    desiredCapacity: 1
    maxSize: 10
    volumeSize: 30
        autoScaler: true
        ebs: true
        fsx: true

Notice the efaEnabled flag in the manifest file. When set to true, eksctl will create a node group with the correct setup for using the Elastic Fabric Adapter (EFA) network interface to run a tightly-coupled workload with MPI. eksctl will leverage AWS CloudFormation to create a Placement group that puts instances close together and a EFA-enabled security group. If EFA is enabled, instanceType must be set to one of the EC2 instance types with EFA support, also the managedNodeGroup availabilityZones must be constrained to a single AZ. In this lab, we will not use EFA, so we are setting the value of this setting to false.

2. Create the EKS cluster

It will take ~10 minutes to create the Kubernetes control plane and ~10 minutes to create the node group, in total ~20 minutes.

eksctl create cluster -f ~/environment/eks-hpc.yaml

Upon successful completion, you will see a log line similar to this:

2022-09-29 03:34:37 [✔]  EKS cluster "eks-hpc" in "us-east-1" region is ready