Part 5: Scaling

What Will You Do

In this part, you will scale the EKS Cluster to zero worker nodes. This will allow you to save on AWS cloud costs such as over a weekend or at night when the worker nodes are not required.

---

Update Cluster Spec

Update the EKS cluster specification file (eks-gpu.yaml) with the following changes for both the Regular and GPU node groups.

GPU Node Group

<deleted text>
managedNodeGroups:
- amiFamily: AmazonLinux2
  desiredCapacity: 0
  iam:
    withAddonPolicies:
      albIngress: true
      autoScaler: true
      efs: true
      externalDNS: true
      imageBuilder: true
  instanceType: g4dn.xlarge
<deleted text>

---

Regular Node Group

<deleted text>
  desiredCapacity: 0
  iam:
    withAddonPolicies:
      albIngress: true
      autoScaler: true
      efs: true
      externalDNS: true
      imageBuilder: true
  instanceType: t3.large
  <deleted text>

---

Scale to Zero

Now, you will use the RCTL CLI to apply the updated cluster specification from the previous step.

rctl apply -f eks-gpu.yaml

You should see a response like the following:

{
  "taskset_id": "7kr1r2l",
  "operations": [
    {
      "operation": "NodegroupScaling",
      "resource_name": "t3-nodegroup",
      "status": "PROVISION_TASK_STATUS_PENDING"
    },
    {
      "operation": "NodegroupScaling",
      "resource_name": "gpu-nodegroup",
      "status": "PROVISION_TASK_STATUS_PENDING"
    }
  ],
  "comments": "The status of the operations can be fetched using taskset_id",
  "status": "PROVISION_TASKSET_STATUS_PENDING"
}

In a few minutes, you should see all the worker nodes in the cluster spin down.

---

Scale Up

When you need the worker nodes back, update the same entries in the cluster specification and apply the file using RTCL. The EKS cluster will automatically scale back to the requested number of worker nodes.

---