Giant Swarm Control Plane admission controller that implements the following rules:
Mutating Webhook:
In an AWSCluster resource, the Region is defaulted if it is not set.
In an AWSCluster resource, the Description is defaulted if it is not set.
In an AWSCluster resource, the DNS Domain is defaulted if it is not set.
In an AWSCluster resource, the Pod CIDR is defaulted if it is not set.
In a G8sControlPlane resource, when the .spec.replicas is changed from 1 to 3, the Availability Zones of the according AWSControlPlane will be defaulted if needed.
In a G8sControlPlane resource, the replicas attribute will be defaulted if it is not defined.
For HA-Versions, in case the matching AWSControlPlane already exists, the number of AZs determines the value of replicas.
In case no such AWSControlPlane exists, the default number of AZs is assigned.
For pre-HA versions, replicas is always set to 1 for a single master cluster.
In a G8sControlPlane resource, the infrastructure reference will be set to point to the matching AWSControlPlane resource if it already exists.
In an AWSControlPlane resource, the Availability Zones will be defaulted if they are nil.
For HA-Versions, in case the matching G8sControlPlane already exists, the number of AZs is determined by the number of replicas defined there.
In case no such G8sControlPlane exists, the default number of AZs is assigned.
For Pre-HA-Versions, in case the matching AWSCluster already exists, the AZ is taken from there.
In an AWSControlPlane resource, the Instance Type will be defaulted if it is not defined.
For HA-Versions, the default Instance Type is chosen.
For Pre-HA-Versions, in case the matching AWSCluster already exists, the Instance Type is taken from there.
On creation of an AWSControlPlane resource, the infrastructure reference of the according G8sControlPlane will be set if needed.
When a new AWSMachineDeployment is created, details are logged.
Validating Webhook:
In a G8sControlPlane resource, it validates the Master Node Replicas are a valid count (Right now either 1 or 3).
In a G8sControlPlane resource, it validates the Master Node Replicas are matching the number of Availability Zones in the AWSControlPlane resource.
In a AWSControlPlane resource, it validates the Master Instance Type is a valid Instance Type for the installation.
In a AWSControlPlane resource, it validates that the order of Master Node Availability Zones does not change on update.
In a AWSControlPlane resource, it validates that the number of distinct Master Node Availability Zones is maximal.
In a AWSControlPlane resource, it validates the Master Node Availability Zones are valid AZs for the installation.
In a AWSControlPlane resource, it validates the Master Node Availability Zones are a valid count (Right now either 1 or 3).
In a AWSControlPlane resource, it validates the Master Node Availability Zones are matching the number of Replicas in the G8sControlPlane resource.
In a AWSMachineDeployment resource, it validates the Machine Deployment ID is matching against MachineDeployment resource.
In a NetworkPool resource, it validates the .Spec.CIDRBlock from other NetworkPools and also checks if there's overlapping from Docker CIDR, Kubernetes cluster IP range or tenant cluster CIDR.
The certificates for the webhook are created with CertManager and injected through the CA Injector.
Ownership
Firecracker Team
Local Development
Testing the aws-admission-controller in a kind cluster on your local machine:
kind create cluster
# Build a linux image
CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build .
docker build . -t aws-admission-controller:dev
kind load docker-image aws-admission-controller:dev
# Make sure the Custom Resource Definitions are in place
opsctl ensure crds -k "$(kind get kubeconfig)" -p aws
# Insert the certificate
kubectl apply --context kind-kind -f local_dev/certmanager.yml
## Wait until certmanager is up
kubectl apply --context kind-kind -f local_dev/clusterissuer.yml
helm template aws-admission-controller -f helm/aws-admission-controller/ci/default-values.yaml helm/aws-admission-controller > local_dev/deploy.yaml
## Replace image name with aws-admission-controller:dev
kubectl apply --context kind-kind -f local_dev/deploy.yaml
kind delete cluster