{0xc00044b600 0xc0004cf0c0} Clean up resources :: Distributed training with Amazon SageMaker / Amazon EKS Workshop

Clean up resources

Amazon EKS resources

Kill all distributed training jobs

kubectl delete MPIJobs --all

Delete StorageClass, PersistentVolumeClaim and FSx for Lustre CSI Driver

Note: This will automatically delete the FSx for luster file system. Your files are safe in Amazon S3.

kubectl delete -f specs/storage-class-fsx-s3.yaml
kubectl delete -f specs/claim-fsx-s3.yaml
kubectl delete -f https://raw.githubusercontent.com/kubernetes-sigs/aws-fsx-csi-driver/master/deploy/kubernetes/manifest.yaml

Delete security group

aws ec2 delete-security-group --group-id ${SECURITY_GROUP_ID}

Delete policies attached to the instance role

These policies were automatically added to the node IAM roles, but we’ll need to manually remove them.

  • Copy the role associated with the worker instances

    echo $INSTANCE_ROLE_NAME
    
  • Navigate to IAM console

  • Click on Roles on the left pane

  • Search for the output of echo $INSTANCE_ROLE_NAME

  • Delete the two inline policies.

    • iam_alb_ingress_policy
    • iam_csi_fsx_policy

Finally, delete the cluster

eksctl delete cluster aws-tf-cluster-cpu

SageMaker resources

SageMaker resources are easier to clear. Login into the SageMaker console and click dashboard Make sure that you don’t have any resources that are Green as shown below. Click on the resources that is shown as green and either stop or delete them.

sm_dashboard

Other resources

It’s always good idea to ensure that: