Skip to content

Release notes for kOps 1.22 series

⚠ kOps 1.22 has not been released yet! ⚠

This is a document to gather the release notes prior to the release.

Significant changes

Instance metadata service version 2

On AWS, kOps will enable Instance Metadata Service Version 2 and require tokens on new clusters with Kubernetes 1.22. In addition, the following max hop limits will be set by default:

  • worker and API server Nodes, and bastions, will have a limit of 1 hop.
  • control plane nodes will have a limit of 3 hops to accommodate for controller Pods without host networking that need to assume roles.

This will increase security by default, but may break some types of workloads. In order to revert to old behavior, add the following to the InstanceGroup:

spec:
  instanceMetadata:
    httpTokens: optional

External ServiceAccountPermissions

Many of kOps addons can now make direct use of external permissions. This can be enabled by adding the following to the Cluster spec:

spec:
  iam:
    useServiceAcountExternalPermissions: true

Currently this is only available using the AWS cloud provider.

Managed nvidia instances

kOps can now provision instances with nvidia GPUs and configure it for container workloads without the need of hooks and operators. See GPU support

Other significant changes

  • New clusters on AWS will no longer provision an SSH public key by default. To provision an SSH public key on a new cluster, use the --ssh-public-key flag to kops create cluster.

  • The kOps Terraform support now renders managed files through the Terraform configuration instead of writing them to S3 directly. This defers changes to these files until the time of terraform apply. This feature may be temporarily disabled by turning off the TerraformManagedFiles feature flag using export KOPS_FEATURE_FLAGS="-TerraformManagedFiles".

  • kOps now implements graceful rotation of its Certificate Authorities and the service account signing key. See the documentation on How to rotate all secrets / credentials

  • New clusters running Kubernetes 1.22 will have AWS EBS CSI driver enabled by default.

  • kOps now supports Debian 11 (Bullseye).

  • kOps can now use external-dns as a drop-in replacement for dns-controller.

Breaking changes

kube-apiserver no longer mounts /srv/kubernetes

For security reasons, /srv/kubernetes is no longer mounted in the kube-apiserver Pod. This also means the files in the default file assets path will be unavailable. If you have file assets or other files needed by kube-apiserver, you must put these into /srv/kubernetes/kube-apiserver/.

For file assets, it means adding an explicit path as shown below:

  fileAssets:
  - name: audit-policy-config
    path: /srv/kubernetes/kube-apiserver/audit-policy-config.yaml # make sure you add the path
    roles:
    - Master
    content: |
      apiVersion: audit.k8s.io/v1
      kind: Policy
      rules:
      - level: Metadat

Other breaking changes

  • Support for Kubernetes versions 1.15 and 1.16 has been removed.

  • The legacy addons from https://github.com/kubernetes/kops/tree/master/addons have been deprecated and will not be available in Kubernetes 1.23+. Use managed addons instead.

  • The legacy location for downloads s3://https://kubeupv2.s3.amazonaws.com/kops/ has been deprecated and will not be used for new releases. The new canonical downloads location is https://artifacts.k8s.io/binaries/kops/.

  • The assets phase of kops update cluster has been removed. It is replaced by the new kops get assets --copy command.

  • Support for importing and converting kubeup clusters has been removed.

Required actions

  • Terraform support now requires Terraform >=0.15.0. Users on older versions must follow Terraform's recommended upgrade path of applying one minor version at a time prior to running kops update cluster --target terraform.

  • The kOps Terraform support now renders managed files through the Terraform configuration instead of writing them to S3 directly. If, after upgrading kOps and applying a new Terraform plan, you subsequently downgrade to an earlier version of kOps, the generated plan will delete these files, breaking the cluster. Prior to applying the plan, you will need to orphan all the aws_s3_bucket_object objects the plan wants to destroy. Use terraform state rm on each of them. Then re-run terraform plan until there are no such objects in the plan.

If you applied the plan without first orphaning all of these objects, fix the cluster by re-running kops update cluster --target terraform.

  • Terraform users of clusters with names beginning with digits will need to move resources prior to upgrading to kOps 1.22. Some of the following commands will need to be run depending on the particular cluster configuration. Confirm the Terraform plan doesn't destroy any of these resources before running terraform apply.
    # View the existing terraform resource names for the exact value to use
    HYPHENATED_CLUSTER_NAME=123-cluster-example-com
    terraform state mv "aws_iam_openid_connect_provider.${HYPHENATED_CLUSTER_NAME}" "aws_iam_openid_connect_provider.prefix_${HYPHENATED_CLUSTER_NAME}"
    terraform state mv "aws_internet_gateway.${HYPHENATED_CLUSTER_NAME}" "aws_internet_gateway.prefix_${HYPHENATED_CLUSTER_NAME}"
    terraform state mv "aws_route_table.${HYPHENATED_CLUSTER_NAME}" "aws_route_table.prefix_${HYPHENATED_CLUSTER_NAME}"
    terraform state mv "aws_vpc.${HYPHENATED_CLUSTER_NAME}" "aws_vpc.prefix_${HYPHENATED_CLUSTER_NAME}"
    terraform state mv "aws_vpc_dhcp_options.${HYPHENATED_CLUSTER_NAME}" "aws_vpc_dhcp_options.prefix_${HYPHENATED_CLUSTER_NAME}"
    terraform state mv "aws_vpc_dhcp_options_association.${HYPHENATED_CLUSTER_NAME}" "aws_vpc_dhcp_options_association.prefix_${HYPHENATED_CLUSTER_NAME}"
    

Deprecations

  • Support for Kubernetes version 1.17 is deprecated and will be removed in kOps 1.23.

  • Support for Kubernetes version 1.18 is deprecated and will be removed in kOps 1.24.

  • Support for the Lyft CNI is deprecated and will be removed in kOps 1.23.

  • Support for CentOS 7 is deprecated and will be removed in future versions of kOps.

  • Support for CentOS 8 is deprecated and will be removed in future versions of kOps.

  • Support for Debian 9 (Stretch) is deprecated and will be removed in future versions of kOps.

  • Support for RHEL 7 is deprecated and will be removed in future versions of kOps.

  • Support for Ubuntu 18.04 (Bionic) is deprecated and will be removed in future versions of kOps.

  • All legacy addons are deprecated in favor of managed addons, including the metrics server addon and the autoscaler addon.

  • The node-role.kubernetes.io/master and kubernetes.io/role labels are deprecated and might be removed from control plane nodes in kOps 1.23.

  • The TerraformJSON feature flag is deprecated and will be removed in kOps 1.23. Only native HCL2 Terraform output will be supported.

  • Due to lack of maintainers, the Aliyun/Alibaba Cloud support has been deprecated. The current implementation will be left as-is until the implementation needs updates or otherwise becomes incompatible. At that point, it will be removed. We very much welcome anyone willing to contribute to this cloud provider.

  • Due to lack of maintainers, the CloudFormation support has been deprecated. The current implementation will be left as-is until the implementation needs updates or otherwise becomes incompatible. At that point, it will be removed. We very much welcome anyone willing to contribute to this target.

Other changes of note

  • Support for shell completion has been substantially improved. kOps has added support for shell completion in fish and PowerShell.

  • It is no longer necessary to set AWS_SDK_LOAD_CONFIG=1 in the environment when using AWS assumed roles with the kops CLI.

  • There is a new command kops get assets for listing image and file assets used by a cluster. It also includes a --copy flag to copy the assets to local repositories. See the documentation on Using local asset repositories for more information.

  • kOps now provisions TLS server certificates signed by the Kubernetes general CA to kube-controller-manager and kube-scheduler. The previous behavior of using self-signed certs may be restored by setting kubeControllerManager.tlsCertFile and/or kubeScheduler.tlsCertFile to "" in the cluster spec.

  • Cilium now supports the wireguard protocol for transparent encryption.

  • Amazon Linux 2 users are encouraged to use the AMIs based on the 5.10 Linux kernel. See the documentation for more information.

Full change list since 1.21.0 release

1.22.0-alpha.1 to 1.22.0-alpha.2

1.22.0-alpha.2 to 1.22.0-beta.1