ownership-controller

ownership-controller

Kubernetes custom controller that controls all owned resources to be successfully created

Preface

This application is designed to familiarize myself with the CRD (custom resource definition) and CRD controller which are Kubernetes core features that allow make Kubernetes API extensions very easy. client-go library was used in main as it has all necessary tools for communication with Kubernetes cluster

Prerequisites

Ownership controller is designed to work inside of cluster so you should have a configured one. You can get acquainted how to configure your own Kubernetes cluster on Ubuntu 18 nice and easy using this article. Also you should have go installed on your workstation and $GOPATH must be set.

Building

Get the repo with:

go get github.com/kacejot/ownership-controller

Go to the project directory:

cd $GOPATH/src/github.com/kacejot/ownership-controller

Run codegen sript to generate helper types for controller work:

./scripts/update_codegen.sh

Build and push the docker image with it to Docker Hub:

docker build --no-cache -t "<user>/<tag>" . && docker push <user>/<tag>

Running controller

Let the project root dir be $PROJECT_ROOT. $PROJECT_ROOT/yaml/crd.yaml is a file that defines custom resource type called Owner. Simply register it in your cluster:

kubectl create -f $PROJECT_ROOT/yaml/crd.yaml

In $PROJECT_ROOT/yaml you have deploy-controller.yaml. This file is responsible for controller creation inside of the cluster. It has several resources: ServiceAccount, ClusterRoleBinding and Deployment. ClusterRoleBinding binds cluster-admin role to our ServiceAccount. That is necessary to allow controller to list Owner resource in cluster scope. Deployment is resource that pulls down image with controller and runs it in the pod. Write your image in image field and launch the controller inside the cluster:

kubectl create -f $PROJECT_ROOT/yaml/deploy-controller.yaml

Now controller works and CRD is registered on the cluster. The last thing that left is to test its work. Create resource with Owner type like described in owner-example.yaml and error-owner-example.yaml

You also can check controller work with logs and pod status. Check that pod is running:

kubectl get pods -n kube-system

Check controller logs:

kubectl logs -n kube-system controller-******

Owner usage

You can now use Owner to be sure that all resources that owner owns will be deleted if someone fails on its creation. Owner structure:Owner contains array of owned resource. Each resource must be represented with 3 fields: namespace, name and resource. Resource field is a plural form for your resource kind definded in its CRD. All the standard resource kinds has plural form by default.

Example of usage:

apiVersion: v1
kind: ServiceAccount
metadata:
  name: sample-service-account-1
  namespace: kube-system
---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: sample-service-account-2
  namespace: kube-system
---
apiVersion: myproject.com/v1alpha1
kind: Owner
metadata:
  name: accounts-owner
spec:
  ownedResources:
  - resource: serviceaccounts
    name: sample-service-account-1
    namespace: kube-system
  - resource: serviceaccounts
    name: sample-service-account-2
    namespace: kube-system