Deploy Patu as a Kubernetes CNI

Patu deployment consist of CNI binary and KPNG eBPF backend.

Deploy Kubernetes without kube-proxy

If you didn’t deploy the cluster yet, you can skip the kube-proxy addon phase during the cluster deployment

kubeadm init  --upload-certs --pod-network-cidr=10.200.0.0/16 --v=6 --skip-phases=addon/kube-proxy

If you already have a running cluster, remove the existing deployed CNI (if any), and remove the existing kube-proxy

kubectl delete daemonsets -n kube-system kube-proxy

Clone Patu Repository

git clone https://github.com/redhat-et/patu.git

Kubernetes Kubeconfig

By default, kubeadm places kubeconfig file at /etc/kubernetes/admin.conf with root permissions. The installer defaults to that location as well. Specify a custom kubeconfig location with the following example passing KUBECONFIG as an environmental variable:

KUBECONFIG=~/.kube/config ./deploy/kubernetes/patu-installer apply all

Install PATU CNI Binary

./deploy/kubernetes/patu-installer apply cni

Ensure all pods are in running state. Coredns pods should have IP from patu CIDR as mentioned in ./deploy/patu.yaml, and should be running but not in Ready state because Cluster-Ip is yet not enabled.

Install KPNG eBPF backend

./deploy/kubernetes/patu-installer apply kpng

All pods should be in running and ready state. Coredns pods should have IP from patu CIDR, KPNG pod should have 3 containers Running and Ready state.

Remove KPNG

./deploy/kubernetes/patu-installer delete kpng

Remove PATU CNI

./deploy/kubernetes/patu-installer delete cni

Install full Patu stack (CNI Binary and KPNG eBPF backend)

./deploy/kubernetes/patu-installer apply all

Uninstall full Patu stack (CNI Binary and KPNG eBPF backend)

./deploy/kubernetes/patu-installer delete all

Manual Instructions:

Install Patu CNI

1. Deploy Patu CNI from the patu directory

   
   `kubectl apply -f ./deploy/patu.yaml`  
   

2. Ensure status - All pods should be in running state. Coredns pods should have IP from patu CIDR as mentioned in patu.yaml, and should be running but not ready

Install KPNG

1. Extract node name:

   
    local_node=$(kubectl get nodes -l 'node-role.kubernetes.io/control-plane' -o custom-columns=NAME:.metadata.name --no-headers)
    

2. Remove taints concerning control-plane, master:

   kubectl taint nodes $local_node node-role.kubernetes.io/master:NoSchedule- node-role.kubernetes.io/control-plane:NoSchedule- 

3. Label node:

    kubectl label node $local_node kube-proxy=kpng 

4. Create configmap:

    kubectl create configmap kpng --namespace kube-system --from-file /etc/kubernetes/admin.conf 

5. Deploy kpng from the patu directory:

   kubectl apply -f ./deploy/kpngebpf.yaml

6. Ensure status - All pods should be in running and ready state. Coredns pods should have IP from patu CIDR, KPNG pod should have 3 containers running and ready

Remove KPNG

1. Remove KPNG Daemon set, service account, cluster role+ binding:

   kubectl delete -f ./deploy/kpngebpf.yaml

2. Remove configmap:

   kubectl delete cm kpng -n kube-system

3. Extract node name:

   
    local_node=$(kubectl get nodes -l 'node-role.kubernetes.io/control-plane' -o custom-columns=NAME:.metadata.name --no-headers)
    

4. Remove label from node:

   kubectl label node $local_node kube-proxy-

Remove PATU

1. Remove PATU Daemon set, service account, cluster role+ binding:

   kubectl delete -f ./deploy/patu.yaml

Testing setup:

kubectl create -f https://k8s.io/examples/application/deployment.yaml
kubectl expose deployment nginx-deployment --type=ClusterIP
kubectl run tmp-shell --rm -i --tty --image nicolaka/netshoot

Expected: Curl to the exposed ip of the nginx-deployment service (obtained using kubectl get svc -A -o wide) should work from the tmp-shell. Coredns pods should be in ready state.