Migrating to a new node group

This topic describes how you can create a new node group, gracefully migrate your existing applications to the new group, and remove the old node group from your cluster. You can migrate to a new node group using eksctl or the AWS Management Console.

eksctl

To migrate your applications to a new node group with eksctl

For more information on using eksctl for migration, see Unmanaged nodegroup upgrades in the eksctl documentation.

This procedure requires eksctl version 0.135.0 or later. You can check your version with the following command:

eksctl version

For instructions on how to install or upgrade eksctl, see Installing or updating eksctl.

Note

This procedure only works for clusters and node groups that were created with eksctl.

1. Retrieve the name of your existing node groups, replacing my-cluster with your cluster name.

   eksctl get nodegroups --cluster=my-cluster

   The example output is as follows.

   CLUSTER      NODEGROUP          CREATED               MIN SIZE      MAX SIZE     DESIRED CAPACITY     INSTANCE TYPE     IMAGE ID
   default      standard-nodes   2019-05-01T22:26:58Z  1             4            3                    t3.medium         ami-05a71d034119ffc12

2. Launch a new node group with eksctl with the following command. In the command, replace every example value with your own values. The version number can't be later than the Kubernetes version for your control plane. Also, it can't be more than two minor versions earlier than the Kubernetes version for your control plane. We recommend that you use the same version as your control plane.

   We recommend blocking pod access to IMDS if the following conditions are true:

   • You plan to assign IAM roles to all of your Kubernetes service accounts so that pods only have the minimum permissions that they need.

   • No pods in the cluster require access to the Amazon EC2 instance metadata service (IMDS) for other reasons, such as retrieving the current AWS Region.

   For more information, see Restrict access to the instance profile assigned to the worker node.

   To block pod access to IMDS, add the --disable-pod-imds option to the following command.

   Note

   For more available flags and their descriptions, see https://eksctl.io/.

   eksctl create nodegroup \
     --cluster my-cluster \
     --version 1.25 \
     --name standard-nodes-new \
     --node-type t3.medium \
     --nodes 3 \
     --nodes-min 1 \
     --nodes-max 4 \
     --managed=false
   

3. When the previous command completes, verify that all of your nodes have reached the Ready state with the following command:

   kubectl get nodes

4. Delete the original node group with the following command. In the command, replace every example value with your cluster and node group names:

   eksctl delete nodegroup --cluster my-cluster --name standard-nodes-old

AWS Management Console and AWS CLI

To migrate your applications to a new node group with the AWS Management Console and AWS CLI

1. Launch a new node group by following the steps that are outlined in Launching self-managed Amazon Linux nodes.

2. When your stack has finished creating, select it in the console and choose Outputs.

3. Record the NodeInstanceRole for the node group that was created. You need this to add the new Amazon EKS nodes to your cluster.

   Note

   If you attached any additional IAM policies to your old node group IAM role, attach those same policies to your new node group IAM role to maintain that functionality on the new group. This applies to you if you added permissions for the Kubernetes Cluster Autoscaler, for example.

4. Update the security groups for both node groups so that they can communicate with each other. For more information, see Amazon EKS security group requirements and considerations.

   1. Record the security group IDs for both node groups. This is shown as the NodeSecurityGroup value in the AWS CloudFormation stack outputs.

      You can use the following AWS CLI commands to get the security group IDs from the stack names. In these commands, oldNodes is the AWS CloudFormation stack name for your older node stack, and newNodes is the name of the stack that you are migrating to. Replace every example value with your own values.

      oldNodes="old_node_CFN_stack_name"
      newNodes="new_node_CFN_stack_name"
      
      oldSecGroup=$(aws cloudformation describe-stack-resources --stack-name $oldNodes \
      --query 'StackResources[?ResourceType==`AWS::EC2::SecurityGroup`].PhysicalResourceId' \
      --output text)
      newSecGroup=$(aws cloudformation describe-stack-resources --stack-name $newNodes \
      --query 'StackResources[?ResourceType==`AWS::EC2::SecurityGroup`].PhysicalResourceId' \
      --output text)

   2. Add ingress rules to each node security group so that they accept traffic from each other.

      The following AWS CLI commands add inbound rules to each security group that allow all traffic on all protocols from the other security group. This configuration allows pods in each node group to communicate with each other while you're migrating your workload to the new group.

      aws ec2 authorize-security-group-ingress --group-id $oldSecGroup \
      --source-group $newSecGroup --protocol -1
      aws ec2 authorize-security-group-ingress --group-id $newSecGroup \
      --source-group $oldSecGroup --protocol -1
      

5. Edit the aws-auth configmap to map the new node instance role in RBAC.

   kubectl edit configmap -n kube-system aws-auth

   Add a new mapRoles entry for the new node group. If your cluster is in the AWS GovCloud (US-East) or AWS GovCloud (US-West) AWS Regions, then replace arn:aws: with arn:aws-us-gov:.

   apiVersion: v1
   data:
     mapRoles: |
       - rolearn: ARN of instance role (not instance profile)
         username: system:node:{{EC2PrivateDNSName}}
         groups:
           - system:bootstrappers
           - system:nodes>
       - rolearn: arn:aws:iam::111122223333:role/nodes-1-16-NodeInstanceRole-U11V27W93CX5
         username: system:node:{{EC2PrivateDNSName}}
         groups:
           - system:bootstrappers
           - system:nodes

   Replace the ARN of instance role (not instance profile) snippet with the NodeInstanceRole value that you recorded in a previous step. Then, save and close the file to apply the updated configmap.

6. Watch the status of your nodes and wait for your new nodes to join your cluster and reach the Ready status.

   kubectl get nodes --watch

7. (Optional) If you're using the Kubernetes Cluster Autoscaler, scale the deployment down to zero (0) replicas to avoid conflicting scaling actions.

   kubectl scale deployments/cluster-autoscaler --replicas=0 -n kube-system

8. Use the following command to taint each of the nodes that you want to remove with NoSchedule. This is so that new pods aren't scheduled or rescheduled on the nodes that you're replacing. For more information, see Taints and Tolerations in the Kubernetes documentation.

   kubectl taint nodes node_name key=value:NoSchedule
   

   If you're upgrading your nodes to a new Kubernetes version, you can identify and taint all of the nodes of a particular Kubernetes version (in this case, 1.23) with the following code snippet. The version number can't be later than the Kubernetes version of your control plane. It also can't be more than two minor versions earlier than the Kubernetes version of your control plane. We recommend that you use the same version as your control plane.

   K8S_VERSION=1.23
   nodes=$(kubectl get nodes -o jsonpath="{.items[?(@.status.nodeInfo.kubeletVersion==\"v$K8S_VERSION\")].metadata.name}")
   for node in ${nodes[@]}
   do
       echo "Tainting $node"
       kubectl taint nodes $node key=value:NoSchedule
   done

9. Determine your cluster's DNS provider.

   kubectl get deployments -l k8s-app=kube-dns -n kube-system
   

   The following is the output. This cluster is using CoreDNS for DNS resolution, but your cluster can return kube-dns instead):

   NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
   coredns   1         1         1            1           31m

10. If your current deployment is running fewer than two replicas, scale out the deployment to two replicas. Replace coredns with kubedns if your previous command output returned that instead.

    kubectl scale deployments/coredns --replicas=2 -n kube-system

11. Drain each of the nodes that you want to remove from your cluster with the following command:

    kubectl drain node_name --ignore-daemonsets --delete-local-data
    

    If you're upgrading your nodes to a new Kubernetes version, identify and drain all of the nodes of a particular Kubernetes version (in this case, 1.23) with the following code snippet.

    K8S_VERSION=1.23
    nodes=$(kubectl get nodes -o jsonpath="{.items[?(@.status.nodeInfo.kubeletVersion==\"v$K8S_VERSION\")].metadata.name}")
    for node in ${nodes[@]}
    do
        echo "Draining $node"
        kubectl drain $node --ignore-daemonsets --delete-local-data
    done

12. After your old nodes finished draining, revoke the security group inbound rules you authorized earlier. Then, delete the AWS CloudFormation stack to terminate the instances.

    Note

    If you attached any additional IAM policies to your old node group IAM role, such as adding permissions for the Kubernetes Cluster Autoscaler), detach those additional policies from the role before you can delete your AWS CloudFormation stack.

    1. Revoke the inbound rules that you created for your node security groups earlier. In these commands, oldNodes is the AWS CloudFormation stack name for your older node stack, and newNodes is the name of the stack that you are migrating to.

       oldNodes="old_node_CFN_stack_name"
       newNodes="new_node_CFN_stack_name"
       
       oldSecGroup=$(aws cloudformation describe-stack-resources --stack-name $oldNodes \
       --query 'StackResources[?ResourceType==`AWS::EC2::SecurityGroup`].PhysicalResourceId' \
       --output text)
       newSecGroup=$(aws cloudformation describe-stack-resources --stack-name $newNodes \
       --query 'StackResources[?ResourceType==`AWS::EC2::SecurityGroup`].PhysicalResourceId' \
       --output text)
       aws ec2 revoke-security-group-ingress --group-id $oldSecGroup \
       --source-group $newSecGroup --protocol -1
       aws ec2 revoke-security-group-ingress --group-id $newSecGroup \
       --source-group $oldSecGroup --protocol -1

    2. Open the AWS CloudFormation console at https://console.aws.amazon.com/cloudformation.

    3. Select your old node stack.

    4. Choose Delete.

    5. In the Delete stack confirmation dialog box, choose Delete stack.

13. Edit the aws-auth configmap to remove the old node instance role from RBAC.

    kubectl edit configmap -n kube-system aws-auth
    

    Delete the mapRoles entry for the old node group. If your cluster is in the AWS GovCloud (US-East) or AWS GovCloud (US-West) AWS Regions, then replace arn:aws: with arn:aws-us-gov:.

    apiVersion: v1
    data:
      mapRoles: |
        - rolearn: arn:aws:iam::111122223333:role/nodes-1-16-NodeInstanceRole-W70725MZQFF8
          username: system:node:{{EC2PrivateDNSName}}
          groups:
            - system:bootstrappers
            - system:nodes
        - rolearn: arn:aws:iam::111122223333:role/nodes-1-15-NodeInstanceRole-U11V27W93CX5
          username: system:node:{{EC2PrivateDNSName}}
          groups:
            - system:bootstrappers
            - system:nodes>

    Save and close the file to apply the updated configmap.

14. (Optional) If you are using the Kubernetes Cluster Autoscaler, scale the deployment back to one replica.

    Note

    You must also tag your new Auto Scaling group appropriately (for example, k8s.io/cluster-autoscaler/enabled,k8s.io/cluster-autoscaler/my-cluster) and update the command for your Cluster Autoscaler deployment to point to the newly tagged Auto Scaling group. For more information, see Cluster Autoscaler on AWS.

    kubectl scale deployments/cluster-autoscaler --replicas=1 -n kube-system
    

15. (Optional) Verify that you're using the latest version of the Amazon VPC CNI plugin for Kubernetes. You might need to update your CNI version to use the latest supported instance types. For more information, see Working with the Amazon VPC CNI plugin for Kubernetes Amazon EKS add-on.

16. If your cluster is using kube-dns for DNS resolution (see previous step), scale in the kube-dns deployment to one replica.

    kubectl scale deployments/kube-dns --replicas=1 -n kube-system