12 minute read  

Using the AWS provider extension with Gardener as end-user

The core.gardener.cloud/v1beta1.Shoot resource declares a few fields that are meant to contain provider-specific configuration.

In this document we are describing how this configuration looks like for AWS and provide an example Shoot manifest with minimal configuration that you can use to create an AWS cluster (modulo the landscape-specific information like cloud profile names, secret binding names, etc.).

Provider Secret Data

Every shoot cluster references a SecretBinding which itself references a Secret, and this Secret contains the provider credentials of your AWS account. This Secret must look as follows:

apiVersion: v1
kind: Secret
metadata:
  name: core-aws
  namespace: garden-dev
type: Opaque
data:
  accessKeyID: base64(access-key-id)
  secretAccessKey: base64(secret-access-key)

The AWS documentation explains the necessary steps to enable programmatic access, i.e. create access key ID and access key, for the user of your choice.

⚠️ For security reasons, we recommend creating a dedicated user with programmatic access only. Please avoid re-using a IAM user which has access to the AWS console (human user).

⚠️ Depending on your AWS API usage it can be problematic to reuse the same AWS Account for different Shoot clusters in the same region due to rate limits. Please consider spreading your Shoots over multiple AWS Accounts if you are hitting those limits.

Permissions

Please make sure that the provided credentials have the correct privileges. You can use the following AWS IAM policy document and attach it to the IAM user backed by the credentials you provided (please check the official AWS documentation as well):

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": "autoscaling:*",
      "Resource": "*"
    },
    {
      "Effect": "Allow",
      "Action": "ec2:*",
      "Resource": "*"
    },
    {
      "Effect": "Allow",
      "Action": "elasticloadbalancing:*",
      "Resource": "*"
    },
    {
      "Action": [
        "iam:GetInstanceProfile",
        "iam:GetPolicy",
        "iam:GetPolicyVersion",
        "iam:GetRole",
        "iam:GetRolePolicy",
        "iam:ListPolicyVersions",
        "iam:ListRolePolicies",
        "iam:ListAttachedRolePolicies",
        "iam:ListInstanceProfilesForRole",
        "iam:CreateInstanceProfile",
        "iam:CreatePolicy",
        "iam:CreatePolicyVersion",
        "iam:CreateRole",
        "iam:CreateServiceLinkedRole",
        "iam:AddRoleToInstanceProfile",
        "iam:AttachRolePolicy",
        "iam:DetachRolePolicy",
        "iam:RemoveRoleFromInstanceProfile",
        "iam:DeletePolicy",
        "iam:DeletePolicyVersion",
        "iam:DeleteRole",
        "iam:DeleteRolePolicy",
        "iam:DeleteInstanceProfile",
        "iam:PutRolePolicy",
        "iam:PassRole",
        "iam:UpdateAssumeRolePolicy"
      ],
      "Effect": "Allow",
      "Resource": "*"
    }
  ]
}

InfrastructureConfig

The infrastructure configuration mainly describes how the network layout looks like in order to create the shoot worker nodes in a later step, thus, prepares everything relevant to create VMs, load balancers, volumes, etc.

An example InfrastructureConfig for the AWS extension looks as follows:

apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
kind: InfrastructureConfig
enableECRAccess: true
networks:
  vpc: # specify either 'id' or 'cidr'
  # id: vpc-123456
    cidr: 10.250.0.0/16
  # gatewayEndpoints:
  # - s3
  zones:
  - name: eu-west-1a
    internal: 10.250.112.0/22
    public: 10.250.96.0/22
    workers: 10.250.0.0/19
  # elasticIPAllocationID: eipalloc-123456
ignoreTags:
  keys: # individual ignored tag keys
  - SomeCustomKey
  - AnotherCustomKey
  keyPrefixes: # ignored tag key prefixes
  - user.specific/prefix/

The enableECRAccess flag specifies whether the AWS IAM role policy attached to all worker nodes of the cluster shall contain permissions to access the Elastic Container Registry of the respective AWS account. If the flag is not provided it is defaulted to true. Please note that if the iamInstanceProfile is set for a worker pool in the WorkerConfig (see below) then enableECRAccess does not have any effect. It only applies for those worker pools whose iamInstanceProfile is not set.

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "ec2:DescribeInstances"
      ],
      "Resource": [
        "*"
      ]
    },
    // Only if `.enableECRAccess` is `true`.
    {
      "Effect": "Allow",
      "Action": [
        "ecr:GetAuthorizationToken",
        "ecr:BatchCheckLayerAvailability",
        "ecr:GetDownloadUrlForLayer",
        "ecr:GetRepositoryPolicy",
        "ecr:DescribeRepositories",
        "ecr:ListImages",
        "ecr:BatchGetImage"
      ],
      "Resource": [
        "*"
      ]
    }
  ]
}

The networks.vpc section describes whether you want to create the shoot cluster in an already existing VPC or whether to create a new one:

• If networks.vpc.id is given then you have to specify the VPC ID of the existing VPC that was created by other means (manually, other tooling, …). Please make sure that the VPC has attached an internet gateway - the AWS controller won’t create one automatically for existing VPCs. To make sure the nodes are able to join and operate in your cluster properly, please make sure that your VPC has enabled DNS Support, explicitly the attributes enableDnsHostnames and enableDnsSupport must be set to true.
• If networks.vpc.cidr is given then you have to specify the VPC CIDR of a new VPC that will be created during shoot creation. You can freely choose a private CIDR range.
• Either networks.vpc.id or networks.vpc.cidr must be present, but not both at the same time.
• networks.vpc.gatewayEndpoints is optional. If specified then each item is used as service name in a corresponding Gateway VPC Endpoint.

The networks.zones section contains configuration for resources you want to create or use in availability zones. For every zone, the AWS extension creates three subnets:

• The internal subnet is used for internal AWS load balancers.
• The public subnet is used for public AWS load balancers.
• The workers subnet is used for all shoot worker nodes, i.e., VMs which later run your applications.

For every subnet, you have to specify a CIDR range contained in the VPC CIDR specified above, or the VPC CIDR of your already existing VPC. You can freely choose these CIDRs and it is your responsibility to properly design the network layout to suit your needs.

Also, the AWS extension creates a dedicated NAT gateway for each zone. By default, it also creates a corresponding Elastic IP that it attaches to this NAT gateway and which is used for egress traffic. The elasticIPAllocationID field allows you to specify the ID of an existing Elastic IP allocation in case you want to bring your own. If provided, no new Elastic IP will be created and, instead, the Elastic IP specified by you will be used.

⚠️ If you change this field for an already existing infrastructure then it will disrupt egress traffic while AWS applies this change. The reason is that the NAT gateway must be recreated with the new Elastic IP association. Also, please note that the existing Elastic IP will be permanently deleted if it was earlier created by the AWS extension.

You can configure Gateway VPC Endpoints by adding items in the optional list networks.vpc.gatewayEndpoints. Each item in the list is used as a service name and a corresponding endpoint is created for it. All created endpoints point to the service within the cluster’s region. For example, consider this (partial) shoot config:

spec:
  region: eu-central-1
  provider:
    type: aws
    infrastructureConfig:
      apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
      kind: InfrastructureConfig
      networks:
        vpc:
          gatewayEndpoints:
          - s3

The service name of the S3 Gateway VPC Endpoint in this example is com.amazonaws.eu-central-1.s3.

If you want to use multiple availability zones then add a second, third, … entry to the networks.zones[] list and properly specify the AZ name in networks.zones[].name.

Apart from the VPC and the subnets the AWS extension will also create DHCP options and an internet gateway (only if a new VPC is created), routing tables, security groups, elastic IPs, NAT gateways, EC2 key pairs, IAM roles, and IAM instance profiles.

The ignoreTags section allows to configure which resource tags on AWS resources managed by Gardener should be ignored during infrastructure reconciliation. By default, all tags that are added outside of Gardener’s reconciliation will be removed during the next reconciliation. This field allows users and automation to add custom tags on AWS resources created and managed by Gardener without loosing them on the next reconciliation. Tags can ignored either by specifying exact key values (ignoreTags.keys) or key prefixes (ignoreTags.keyPrefixes). In both cases it is forbidden to ignore the Name tag or any tag starting with kubernetes.io or gardener.cloud.
Please note though, that the tags are only ignored on resources created on behalf of the Infrastructure CR (i.e. VPC, subnets, security groups, keypair, etc.), while tags on machines, volumes, etc. are not in the scope of this controller.

ControlPlaneConfig

The control plane configuration mainly contains values for the AWS-specific control plane components. Today, the only component deployed by the AWS extension is the cloud-controller-manager.

An example ControlPlaneConfig for the AWS extension looks as follows:

apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
kind: ControlPlaneConfig
cloudControllerManager:
  featureGates:
    CustomResourceValidation: true
  useCustomRouteController: true
storage:
  managedDefaultClass: false

The cloudControllerManager.featureGates contains a map of explicitly enabled or disabled feature gates. For production usage it’s not recommend to use this field at all as you can enable alpha features or disable beta/stable features, potentially impacting the cluster stability. If you don’t want to configure anything for the cloudControllerManager simply omit the key in the YAML specification.

The cloudControllerManager.useCustomRouteController controls if the custom routes controller should be enabled. If enabled, it will add routes to the pod CIDRs for all nodes in the route tables for all zones.

The storage.managedDefaultClass controls if the default storage / volume snapshot classes are marked as default by Gardener. Set it to false to mark another storage / volume snapshot class as default without Gardener overwriting this change. If unset, this field defaults to true.

WorkerConfig

The AWS extension supports encryption for volumes plus support for additional data volumes per machine. For each data volume, you have to specify a name. By default (if not stated otherwise), all the disks (root & data volumes) are encrypted. Please make sure that your instance-type supports encryption. If your instance-type doesn’t support encryption, you will have to disable encryption (which is enabled by default) by setting volume.encrpyted to false (refer below shown YAML snippet).

The following YAML is a snippet of a Shoot resource:

spec:
  provider:
    workers:
    - name: cpu-worker
      ...
      volume:
        type: gp2
        size: 20Gi
        encrypted: false
      dataVolumes:
      - name: kubelet-dir
        type: gp2
        size: 25Gi
        encrypted: true

Note: The AWS extension does not support EBS volume (root & data volumes) encryption with customer managed CMK. Support for customer managed CMK is out of scope for now. Only AWS managed CMK is supported.

Additionally, it is possible to provide further AWS-specific values for configuring the worker pools. It can be provided in .spec.provider.workers[].providerConfig and is evaluated by the AWS worker controller when it reconciles the shoot machines.

An example WorkerConfig for the AWS extension looks as follows:

apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
kind: WorkerConfig
volume:
  iops: 10000
  throughput: 200 
dataVolumes:
- name: kubelet-dir
  iops: 12345
  throughput: 150
  snapshotID: snap-1234
iamInstanceProfile: # (specify either ARN or name)
  name: my-profile
# arn: my-instance-profile-arn
nodeTemplate: # (to be specified only if the node capacity would be different from cloudprofile info during runtime)
  capacity:
    cpu: 2
    gpu: 0
    memory: 50Gi

The .volume.iops is the number of I/O operations per second (IOPS) that the volume supports. For io1 and gp3 volume type, this represents the number of IOPS that are provisioned for the volume. For gp2 volume type, this represents the baseline performance of the volume and the rate at which the volume accumulates I/O credits for bursting. For more information about General Purpose SSD baseline performance, I/O credits, IOPS range and bursting, see Amazon EBS Volume Types (http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EBSVolumeTypes.html) in the Amazon Elastic Compute Cloud User Guide.
Constraint: IOPS should be a positive value. Validation of IOPS (i.e. whether it is allowed and is in the specified range for a particular volume type) is done on aws side.

The volume.throughput is the throughput that the volume supports, in MiB/s. As of 16th Aug 2022, this parameter is valid only for gp3 volume types and will return an error from the provider side if specified for other volume types. Its current range of throughput is from 125MiB/s to 1000 MiB/s. To know more about throughput and its range, see the official AWS documentation here.

The .dataVolumes can optionally contain configurations for the data volumes stated in the Shoot specification in the .spec.provider.workers[].dataVolumes list. The .name must match to the name of the data volume in the shoot. It is also possible to provide a snapshot ID. It allows to restore the data volume from an existing snapshot.

The iamInstanceProfile section allows to specify the IAM instance profile name xor ARN that should be used for this worker pool. If not specified, a dedicated IAM instance profile created by the infrastructure controller is used (see above).

Example Shoot manifest (one availability zone)

Please find below an example Shoot manifest for one availability zone:

apiVersion: core.gardener.cloud/v1beta1
kind: Shoot
metadata:
  name: johndoe-aws
  namespace: garden-dev
spec:
  cloudProfileName: aws
  region: eu-central-1
  secretBindingName: core-aws
  provider:
    type: aws
    infrastructureConfig:
      apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
      kind: InfrastructureConfig
      networks:
        vpc:
          cidr: 10.250.0.0/16
        zones:
        - name: eu-central-1a
          internal: 10.250.112.0/22
          public: 10.250.96.0/22
          workers: 10.250.0.0/19
    controlPlaneConfig:
      apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
      kind: ControlPlaneConfig
    workers:
    - name: worker-xoluy
      machine:
        type: m5.large
      minimum: 2
      maximum: 2
      volume:
        size: 50Gi
        type: gp2
    # The following provider config is valid if the volume type is `io1`.
    # providerConfig:
    #   apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
    #   kind: WorkerConfig
    #   volume:
    #     iops: 10000
      zones:
      - eu-central-1a
  networking:
    nodes: 10.250.0.0/16
    type: calico
  kubernetes:
    version: 1.24.3
  maintenance:
    autoUpdate:
      kubernetesVersion: true
      machineImageVersion: true
  addons:
    kubernetesDashboard:
      enabled: true
    nginxIngress:
      enabled: true

Example Shoot manifest (three availability zones)

Please find below an example Shoot manifest for three availability zones:

apiVersion: core.gardener.cloud/v1beta1
kind: Shoot
metadata:
  name: johndoe-aws
  namespace: garden-dev
spec:
  cloudProfileName: aws
  region: eu-central-1
  secretBindingName: core-aws
  provider:
    type: aws
    infrastructureConfig:
      apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
      kind: InfrastructureConfig
      networks:
        vpc:
          cidr: 10.250.0.0/16
        zones:
        - name: eu-central-1a
          workers: 10.250.0.0/26
          public: 10.250.96.0/26
          internal: 10.250.112.0/26
        - name: eu-central-1b
          workers: 10.250.0.64/26
          public: 10.250.96.64/26
          internal: 10.250.112.64/26
        - name: eu-central-1c
          workers: 10.250.0.128/26
          public: 10.250.96.128/26
          internal: 10.250.112.128/26
    controlPlaneConfig:
      apiVersion: aws.provider.extensions.gardener.cloud/v1alpha1
      kind: ControlPlaneConfig
    workers:
    - name: worker-xoluy
      machine:
        type: m5.large
      minimum: 3
      maximum: 9
      volume:
        size: 50Gi
        type: gp2
      zones:
      - eu-central-1a
      - eu-central-1b
      - eu-central-1c
  networking:
    nodes: 10.250.0.0/16
    type: calico
  kubernetes:
    version: 1.24.3
  maintenance:
    autoUpdate:
      kubernetesVersion: true
      machineImageVersion: true
  addons:
    kubernetesDashboard:
      enabled: true
    nginxIngress:
      enabled: true

CSI volume provisioners

Every AWS shoot cluster that has at least Kubernetes v1.18 will be deployed with the AWS EBS CSI driver. It is compatible with the legacy in-tree volume provisioner that was deprecated by the Kubernetes community and will be removed in future versions of Kubernetes. End-users might want to update their custom StorageClasses to the new ebs.csi.aws.com provisioner. Shoot clusters with Kubernetes v1.17 or less will use the in-tree kubernetes.io/aws-ebs volume provisioner in the kube-controller-manager and the kubelet.

Node-specific Volume Limits

The Kubernetes scheduler allows configurable limit for the number of volumes that can be attached to a node. See https://k8s.io/docs/concepts/storage/storage-limits/#custom-limits.

CSI drivers usually have a different procedure for configuring this custom limit. By default, the EBS CSI driver parses the machine type name and then decides the volume limit. However, this is only a rough approximation and not good enough in most cases. Specifying the volume attach limit via command line flag (--volume-attach-limit) is currently the alternative until a more sophisticated solution presents itself (dynamically discovering the maximum number of attachable volume per EC2 machine type, see also https://github.com/kubernetes-sigs/aws-ebs-csi-driver/issues/347). The AWS extension allows the --volume-attach-limit flag of the EBS CSI driver to be configurable via aws.provider.extensions.gardener.cloud/volume-attach-limit annotation on the Shoot resource. If the annotation is added to an existing Shoot, then reconciliation needs to be triggered manually (see Immediate reconciliation), as in general adding annotation to resource is not a change that leads to .metadata.generation increase in general.

Kubernetes Versions per Worker Pool

This extension supports gardener/gardener’s WorkerPoolKubernetesVersion feature gate, i.e., having worker pools with overridden Kubernetes versions since gardener-extension-provider-aws@v1.34. Note that this feature is only usable for Shoots whose .spec.kubernetes.version is greater or equal than the CSI migration version (1.18).

Shoot CA Certificate and ServiceAccount Signing Key Rotation

This extension supports gardener/gardener’s ShootCARotation and ShootSARotation feature gates since gardener-extension-provider-aws@v1.36.