Release notes for kOps 1.19 series

Significant changes

Terraform users on AWS should read the Required Actions section below to avoid potential etcd data loss.

  • Specific to 1.19.0: containerd is blocked with kubenet, kopeio, gce and external networking modes. This will be addressed in 1.19.1; if you are using containerd with one of those networking modes we recommend waiting for that release. As docker is the default container runtime, if you have not explicitly opted in to containerd you are likely not affected. (As described in the deprecations section below, containerd is planned to be the default in 1.20).

Changes to kubernetes config export

kOps will no longer automatically export user credentials on kops update cluster. In order to export user credentials on cluster update, you need to either add the --user <user> to reference an existing user block in kubeconfig, or --admin to export the cluster admin user. If neither flag is passed, the kubernetes user config will not be modified. This makes it easier to reuse user definitions across clusters should you, for example, use OIDC for authentication.

Similarly, kops export kubecfg will also require passing either the --admin or --user flag if the context does not already exist.

By default, the credentials of any exported admin user now have a lifetime of 18 hours. The lifetime of the exported credentials may be specified as a value of the --admin flag. To get the previous behavior, specify --admin=87600h to either kops update cluster or kops export kubecfg.

kops create cluster --yes exports the admin user along with rest of the cluster config, as was the previous behaviour (except for the 18-hour validity).

ARM64 support

kOps will install ARM64 artifacts and containerd images when the instance group supports ARM64 for both machine type and OS image. At the moment this is known to work with AWS m6g, c6g, r6g and t4g instances, with the latest Ubuntu 20.04 OS images for ARM64.

OpenStack Cinder plugin

kOps will install the Cinder plugin for kOps running kubernetes 1.16 or newer. If you already have this plugin installed you should remove it before upgrading.

If you already have a default StorageClass, you should set cloudConfig.Openstack.BlockStorage.CreateStorageClass: false to prevent kOps from installing one.

Other significant changes by kind

General

  • New clusters will now have one nodes group per zone. The number of nodes now defaults to the number of zones.

  • On AWS kOps now defaults to using launch templates instead of launch configurations.

  • There is now Alpha support for Hashicorp Vault as a store for secrets and keys. See the Vault state store docs.

  • The lifetimes of certificates used by various components have been substantially reduced. The certificates on a node will expire sometime between 455 and 485 days after the node’s creation. The expiration times vary randomly so that nodes are likely to have their certs expire at different times than other nodes.

  • kOps now supports using an AWS Network Load Balancer (NLB) for API access. See the documentation for more info.

  • Allow users to partially compress user-data, check the instance groups docs for more details.

  • Worker nodes on AWS will now be bootstrapped using kops-controller.

CLI

  • The kops update cluster command will now refuse to run on a cluster that has been updated by a newer version of kOps unless it is given the --allow-kops-downgrade flag.

  • New command for deleting a single instance: kops delete instance

CNI

  • Clusters using the Amazon VPC CNI provider now perform an ec2.DescribeInstanceTypes call at instance launch time. In large clusters or AWS accounts this may lead to API throttling which could delay node readiness. If this becomes a problem please open a GitHub issue.

  • Clusters using Calico with CrossSubnet enabled will switch to the new awsSrcDstCheck for disabling the AWS source/destination checks. The previous implementation using k8s-ec2-srcdst is now deprecated.

  • Clusters using Calico can now enable the eBPF dataplane mode for Ubuntu 20.04 (Focal) hosts. Add spec.networking.calico.bpfEnabled: true and spec.kubeProxy.enabled: false to the cluster spec to enable.

  • Clusters using Calico can now encrypt pod-to-pod traffic with WireGuard for Ubuntu hosts. Add spec.networking.calico.wireguardEnabled: true to the cluster spec to enable.

  • New clusters running Cilium now enable BPF NodePort by default if the Kubernetes version is 1.12 or newer.

  • Cilium will enable enable-remote-identity by default. This can affect network policies. If you want to keep old behaviour, add spec.networking.cilium.enableRemoteIdentity: false in your cluster spec.

Addons

  • Metrics Server is now available as a configurable addon. Add spec.metricsServer.enabled: true to the cluster spec to enable.
  • With this change, one no longer has to set the --kubelet-insecure-tls flag, making any metrics-server installation more secure. See #6879 for more details.

  • Cluster Autoscaler is now availalble as a configurable addon. Add spec.clusterAutoscaler.enabled: true to the cluster spec to enable.

  • AWS Node Termination Handler is now available as a configurable addon. Add spec.nodeTerminationHandler.enabled: true to the cluster spec to enable.

Breaking changes

  • Support for Kubernetes 1.9 and 1.10 has been removed.

  • Support for the Romana networking provider has been removed.

  • Support for legacy IAM permissions has been removed. This removal may be temporarily deferred to kOps 1.20 by setting the LegacyIAM feature flag.

Required Actions

  • See note about Openstack Cinder plugin above.

  • Terraform 0.12 users on AWS, in order to prevent downtime you will have to remove the state of any existing ELB or TargetGroup attachments from your Terraform state file. This is due to migrating the attachments to the in-line aws_autoscaling_group fields. See the terraform documentation for more information about the difference. This migration is required due to a bug described in #9913.

To prevent downtime, follow these steps with the new version of Kops:

  1. kops update cluster --target terraform ...
  2. terraform plan
  3. terraform state list | grep aws_autoscaling_attachment | xargs -L1 terraform state rm
  4. terraform plan
  5. # Ensure these resources are no longer being destroyed and recreated
  6. terraform apply
  • Terraform 0.12 users on AWS migrating clusters from Launch Configurations to Launch Templates may need to remove the state of the old Launch Configuration. This is due to potential errors with Terraform attempting to delete the Launch Configuration before updating the AutoScalingGroup to use the Launch Template. The Launch Configurations will need to be manually deleted afterwards. More information including detailed remediation steps is available in #10017.

    1. kops update cluster --target terraform ...
    2. terraform state list | grep aws_launch_configuration | xargs -L1 terraform state rm
    3. terraform plan
    4. # Ensure launch configurations are not being destroyed
    5. terraform apply
  • Terraform users on AWS may need to rename their EBS Volume resources to match 0.12’s stricter naming requirements. Volumes whose Terraform resource name begin with a digit are now prefixed with ebs-. This change will be made regardless of Terraform-0.12 feature flag value. More information is available in #9982. When upgrading to kOps 1.19, follow these steps to determine if a rename is necessary:

    1. kops update cluster --target terraform ...
    2. terraform plan
    3. # Look for any EBS volumes being recreated
    4. # Adjust these arguments as necessary
    5. terraform state mv aws_ebs_volume.1a-etcd-events-foo-k8s-local aws_ebs_volume.ebs-1a-etcd-events-foo-k8s-local
    6. terraform plan
    7. # Confirm no EBS volumes being changed
    8. terraform apply
  • If you are using Terraform with an additional .tf file and using “aws_autoscaling_attachment” to attach additional Load Balancers or ALB/NLB Target Groups you’ll need to migrate to attaching them through the InstanceGroup spec instead.

  • AWS clusters using an ACM Certificate on the API ELB (.spec.api.loadBalancer.sslCertificateID) will need to migrate from Classic LoadBalancer (CLB) to Network LoadBalancer (NLB) prior to upgrading to Kubernetes 1.19 by setting .spec.api.loadBalancer.class: Network. Any kubeconfig files using kOps’ admin client credentials will need to be regenerated with kops export kubecfg --admin. For more information see this page.

Deprecations

  • Support for Kubernetes versions 1.11 and 1.12 has been deprecated and will be removed in kOps 1.20.

  • Support for Docker container runtime has been deprecated and will be replaced with containerd for new clusters as of kOps 1.20. For existing clusters the default behaviour can be overridden by setting spec.containerRuntime: containerd in the cluster spec. This change closely tracks the upstream deprecation of Docker.

  • Support for Terraform version 0.11 has been deprecated and will be removed in kOps 1.20.

  • Support for feature flag Terraform-0.12 has been deprecated and will be removed in kOps 1.20. All generated Terraform HCL2/JSON files will support versions 0.12.26+ and 0.13.0+.

  • The manifest based metrics server addon has been deprecated in favour of a configurable addon.

  • The manifest based cluster autoscaler addon has been deprecated in favour of a configurable addon.

  • The experimental node authorizer is now ignored if you are using kubernetes 1.19. The feature will be removed in 1.20. Worker nodes will instead be authorized using kops-controller.

  • Support for AWS LaunchConfiguration has been deprecated and will be removed in kOps 1.21.

Full change list since 1.18.0 release

v1.18.0-alpha.3 to v1.19.0-alpha.1

1.19.0-alpha.1 to 1.19.0-alpha.2

1.19.0-alpha.2 to 1.19.0-alpha.3

1.19.0-alpha.3 to 1.19.0-alpha.4

1.19.0-alpha.4 to 1.19.0-alpha.5

1.19.0-alpha.5 to 1.19.0-beta.1

1.19.0-beta.1 to 1.19.0-beta.2

1.19.0-beta.2 to 1.19.0-beta.3

1.19.0-beta.3 to 1.19.0

1.19.0 to 1.19.1

1.19.1 to 1.19.2

1.19.2 to 1.19.3