EKS Fully-Private Cluster¶
eksctl supports creation of fully-private clusters that have no outbound internet access and have only private subnets. VPC endpoints are used to enable private access to AWS services.
This guide describes how to create a private cluster without outbound internet access.
Creating a fully-private cluster¶
The only required field to create a fully-private cluster is privateCluster.enabled
:
privateCluster:
enabled: true
Note
Post cluster creation, eksctl commands that need access to the Kubernetes API server will have to be run from within the cluster's VPC, a peered VPC or using some other means like AWS Direct Connect. eksctl commands that need access to the EKS APIs will not work if they're being run from within the cluster's VPC. To fix this, create an interface endpoint for Amazon EKS to privately access the Amazon Elastic Kubernetes Service (Amazon EKS) management APIs from your Amazon Virtual Private Cloud (VPC). In a future release, eksctl will add support to create this endpoint so it does not need to be manually created. Commands that need access to the OpenID Connect provider URL will need to be run from outside of your cluster's VPC once you've enabled AWS PrivateLink for Amazon EKS. Creating managed nodegroups will continue to work, and creating self-managed nodegroups will work as it needs access to the API server via the EKS interface endpoint if the command is run from within the cluster's VPC, a peered VPC or using some other means like AWS Direct Connect.
Info
VPC endpoints are charged by the hour and based on usage. More details about pricing can be found at AWS PrivateLink pricing
Warning
Fully-private clusters are not supported in eu-south-1
.
Configuring private access to additional AWS services¶
To enable worker nodes to access AWS services privately, eksctl creates VPC endpoints for the following services:
- Interface endpoints for ECR (both
ecr.api
andecr.dkr
) to pull container images (AWS CNI plugin etc) - A gateway endpoint for S3 to pull the actual image layers
- An interface endpoint for EC2 required by the
aws-cloud-provider
integration - An interface endpoint for STS to support Fargate and IAM Roles for Services Accounts (IRSA)
- An interface endpoint for CloudWatch logging (
logs
) if CloudWatch logging is enabled
These VPC endpoints are essential for a functional private cluster, and as such, eksctl does not support configuring or disabling them. However, a cluster might need private access to other AWS services (e.g., Autoscaling required by the Cluster Autoscaler). These services can be specified in privateCluster.additionalEndpointServices
, which instructs eksctl to create a VPC endpoint for each of them.
For example, to allow private access to Autoscaling and CloudWatch logging:
privateCluster:
enabled: true
additionalEndpointServices:
# For Cluster Autoscaler
- "autoscaling"
# CloudWatch logging
- "logs"
The endpoints supported in additionalEndpointServices
are autoscaling
, cloudformation
and logs
.
Skipping endpoint creations¶
If a VPC has already been created with the necessary AWS endpoints set up and linked to the subnets described in the EKS documentation, eksctl
can skip creating them by providing the option skipEndpointCreation
like this:
privateCluster:
enabled: true
skipEndpointCreation: true
Note: this setting cannot be used together with additionalEndpointServices
. It will skip all endpoint creation. Also, this setting is only recommended if the endpoint <-> subnet topology is correctly set up. I.e.: subnet ids are correct, vpce
routing is set up with prefix addresses, all the necessary EKS endpoints are created and linked to the provided VPC. eksctl
will not alter any of these resources.
Nodegroups¶
Only private nodegroups (both managed and self-managed) are supported in a fully-private cluster because the cluster's VPC is created without any public subnets. The privateNetworking
field (nodeGroup[*].privateNetworking
and managedNodeGroup[*].privateNetworking
) must be explicitly set. It is an error to leave privateNetworking
unset in a fully-private cluster.
nodeGroups:
- name: ng1
instanceType: m5.large
desiredCapacity: 2
# privateNetworking must be explicitly set for a fully-private cluster
# Rather than defaulting this field to `true`,
# we require users to explicitly set it to make the behaviour
# explicit and avoid confusion.
privateNetworking: true
managedNodeGroups:
- name: m1
instanceType: m5.large
desiredCapacity: 2
privateNetworking: true
Cluster Endpoint Access¶
A fully-private cluster does not support modifying clusterEndpointAccess
during cluster creation. It is an error to set either clusterEndpoints.publicAccess
or clusterEndpoints.privateAccess
, as a fully-private cluster can have private access only, and allowing modification of these fields can break the cluster.
User-supplied VPC and subnets¶
eksctl supports creation of fully-private clusters using a pre-existing VPC and subnets. Only private subnets can be specified and it's an error to specify subnets under vpc.subnets.public
.
eksctl creates VPC endpoints in the supplied VPC and modifies route tables for the supplied subnets. Each subnet should have an explicit route table associated with it because eksctl does not modify the main route table.
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
name: private-cluster
region: us-west-2
privateCluster:
enabled: true
additionalEndpointServices:
- "autoscaling"
vpc:
subnets:
private:
us-west-2b:
id: subnet-0818beec303f8419b
us-west-2c:
id: subnet-0d42ef09490805e2a
us-west-2d:
id: subnet-0da7418077077c5f9
nodeGroups:
- name: ng1
instanceType: m5.large
desiredCapacity: 2
# privateNetworking must be explicitly set for a fully-private cluster
# Rather than defaulting this field to true for a fully-private cluster, we require users to explicitly set it
# to make the behaviour explicit and avoid confusion.
privateNetworking: true
managedNodeGroups:
- name: m1
instanceType: m5.large
desiredCapacity: 2
privateNetworking: true
Managing a fully-private cluster¶
For all commands to work post cluster creation, eksctl will need private access to the EKS API server endpoint, and outbound internet access (for EKS:DescribeCluster
). Commands that do not need access to the API server will be supported if eksctl has outbound internet access.
Force-delete a fully-private cluster¶
Errors are likely to occur when deleting a fully-private cluster through eksctl since eksctl does not automatically have access to all of the cluster's resources. --force
exists to solve this: it will force delete the cluster and continue when errors occur.
Limitations¶
A limitation of the current implementation is that eksctl initially creates the cluster with both public and private endpoint access enabled, and disables public endpoint access after all operations have completed. This is required because eksctl needs access to the Kubernetes API server to allow self-managed nodes to join the cluster and to support GitOps and Fargate. After these operations have completed, eksctl switches the cluster endpoint access to private-only. This additional update does mean that creation of a fully-private cluster will take longer than for a standard cluster. In the future, eksctl may switch to a VPC-enabled Lambda function to perform these API operations.
Outbound access via HTTP proxy servers¶
eksctl is able to talk to the AWS APIs via a configured HTTP(S) proxy server, however you will need to ensure you set your proxy exclusion list correctly.
Generally, you will need to ensure that requests for the VPC endpoint for your cluster are not routed via your proxies by setting an appropriate no_proxy
environment variable including the value .eks.amazonaws.com
.
If your proxy server performs "SSL interception" and you are using IAM Roles for Service Accounts (IRSA), you will need to ensure that you explicitly bypass SSL Man-in-the-Middle for the domain oidc.<region>.amazonaws.com
. Failure to do so will result in eksctl obtaining the incorrect root certificate thumbprint for the OIDC provider, and the AWS VPC CNI plugin will fail to start due to being unable to obtain IAM credentials, rendering your cluster inoperative.