Amazon EKS kubectl Context: Difference between revisions
Line 9: | Line 9: | ||
=Use the Current AWS Identity= | =Use the Current AWS Identity= | ||
If no --role-arn option is specified for the <code>aws eks</code> command, kubectl context is configured to accesses the EKS cluster with the default AWS CLI IAM user identity at the time of aws eks execution, so make sure that AWS_PROFILE is set to the AWS profile that carries the identity that was used to create the cluster. | |||
<syntaxhighlight lang='text'> | <syntaxhighlight lang='text'> |
Revision as of 23:56, 28 September 2020
External
Internal
Overview
The local kubectl context for an EKS cluster is generated automatically with the aws eks update-kubeconfig
command, which updates .kube/config. The command can be used to generate a context for the current AWS identity, or for an AWS IAM role. The command updates .kube/config with a pre-populated server endpoint and certificate authority data values for the cluster specified by name. If the right region is configured in the profile, there is no need to be specified.
Use the Current AWS Identity
If no --role-arn option is specified for the aws eks
command, kubectl context is configured to accesses the EKS cluster with the default AWS CLI IAM user identity at the time of aws eks execution, so make sure that AWS_PROFILE is set to the AWS profile that carries the identity that was used to create the cluster.
aws --region <region> eks update-kubeconfig --name <cluster-name> --alias <context-name>
This creates the following context elements:
1. A "cluster" with the name "arn:aws:eks:<region>:999999999999:cluster/<cluster-name>". The cluster's API Server URL and certificate authority are retrieved automatically from the EKS cluster configuration:
- name: arn:aws:eks:us-west-2:999999999999:cluster/<cluster-name>
cluster:
certificate-authority-data: LS0...Qo=
server: https://C5000000000000000000000000000000.gr7.us-west-2.eks.amazonaws.com
2. A "user" with the name "arn:aws:eks:<region>:999999999999:cluster/<cluster-name>" (same as the cluster name), whose identity is a token generated via an "aws get-token" command:
- name: arn:aws:eks:us-west-2:999999999999:cluster/<cluster-name>
user:
exec:
apiVersion: client.authentication.k8s.io/v1alpha1
args:
- --region
- us-west-2
- eks
- get-token
- --cluster-name
- <cluster-name>
command: aws
env:
- name: AWS_PROFILE
value: default
This is equivalent with executing the command:
export AWS_PROFILE=<profile-used-when-command-was-run>
aws --region <region> eks get-token --cluster-name <cluster-name>
The result is an authentication token.
3. The context, named "<context-name>", that binds these two together. If no alias is specified, the context name is the same as the cluster name, and the user name.
- context:
name: <context-name>
cluster: arn:aws:eks:us-west-2:999999999999:cluster/<cluster-name>
user: arn:aws:eks:us-west-2:999999999999:cluster/<cluster-name>
Use an IAM Role
aws --region <region> eks update-kubeconfig --name <cluster-name> --alias <context-alias> --role-arn <role-arn>
The only difference to using current identity is the way the authentication token is generated:
- name: arn:aws:eks:us-west-2:999999999999:cluster/<cluster-name>
user:
exec:
apiVersion: client.authentication.k8s.io/v1alpha1
args:
- --region
- us-west-2
- eks
- get-token
- --cluster-name
- <cluster-name>
- --role
- arn:aws:iam::999999999999:role/<role-name>
command: aws
env:
- name: AWS_PROFILE
value: default
This is equivalent with executing:
export AWS_PROFILE=<profile-used-when-command-was-run>
aws --region <region> eks get-token --cluster-name <cluster-name> --role arn:aws:iam::999999999999:role/<role-name>
⚠️ If "aws eks update-kubeconfig" command is used twice with the same .kube/config, the last execution will overwrite the user entry, even if a different context alias was used, producing two identical contexts, yet named differently. That is why is best to only use "aws eks update-kubeconfig" once, while generating the initial configuration, then update .kube/config manually with additional contexts.
TO PROCESS
If the right region is configured in the profile, there is no need to be specified. If no alias is used, the default name of the context is the cluster ARN. The result is to add a new context to .kube/config:
Added new context arn:aws:eks:us-east-1:999999999999:cluster/example-eks-cluster to /Users/testuser/.kube/config
If no --role-arn option is specified for the aws eks command, kubectl context is configured to accesses the EKS cluster with the default AWS CLI IAM user identity at the time of aws eks execution. This identity can be obtained with aws sts get-caller-identity. The IAM identity associated with the context can be changed with the --role-arn option. If the --role-arn option is specified, the Kubernetes context will be configured as such that it will not be necessary to explicitly assume the role; kubectl operations in the correct context will simply work. Note that the IAM role used for --role-arn is NOT the cluster service role, but a completely different role altogether.
For more details on how the IAM user or role identity is linked to a specific set of RBAC permissions, see:
Building upon this capability, it is possible to create two different Kuberenetes context that imply to different sets of RBAC permission on the Kubernetes clusters:
aws eks update-kubeconfig --name example-eks-cluster --alias access-with-cluster-admin-permissions --role-arn arn:aws:iam::999999999999:role/eks-clusterrole-cluster-admin
aws eks update-kubeconfig --name example-eks-cluster --alias access-with-limited-permissions --role-arn arn:aws:iam::999999999999:role/eks-clusterrole-limited-permissions
Switching between Kubernetes contexts is done with kubectl config use-context:
kubectl config use-context access-with-cluster-admin-permissions
kubectl config current-context
access-with-cluster-admin-permissions
kubectl config use-context access-with-limited-permissions
kubectl config current-context
access-with-limited-permissions
I ran into trouble using a role to configure access to a cluster, it overwrote something locally so I could not login. To investigate.