kubernetes-improvement-scope

Implementing a research paper to improve upon the existing kubernetes use case using an example kubernetes 5-node cluster System: Windows

Prerequisites

WSL 2, Docker
Kubectl:

 curl -LO https://storage.googleapis.com/kubernetes-release/release/v1.7.0/bin/windows/amd64/kubectl.exe

Kind: to easily set up multi-node clusters locally (Install go as well)

go install sigs.k8s.io/kind@v0.23.0

Create the cluster:

touch cluster-config.yaml

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
- role: worker
- role: worker
- role: worker
- role: worker
- role: worker

Setup cluster:

kind create cluster --config=cluster-config.yaml

Install WeaveNet CLI

kubectl apply -f https://github.com/weaveworks/weave/releases/download/v2.8.1/weave-daemonset-k8s.yaml

Verify weavenet configuration: kubectl get pods -n kube-system

Setup 5 microservices with varying resource requirements:
Create a separate YAML file for each microservice deployment. Each microservice will have different resource requirements specified.
Make a docker image: docker build -t hello-world-node:latest .
Deploy microservices:

kubectl apply -f microservice1-deployment.yaml
kubectl apply -f microservice1-service.yaml
# Repeat for other microservices

Check if the microservices are working: kubectl get pods -o wide

Apply network policy: kubectl apply -f network-policy.yaml
Check if the network policy is applied: kubectl get networkpolicy
Describe the NetworkPolicy to verify its details: kubectl describe networkpolicy allow-specific

Analyzing

Print kube scheduler logs:

kubectl logs -n kube-system kube-scheduler-kind-control-plane

Print events related to pods in the 'kube-system' namespace kubectl get events --namespace kube-system --field-selector involvedObject.kind=Pod

Improvements

Pod priority is not considered and a randomly selected pod is scheduled on the node with the highest node score, and hence the global optimal solution cannot be made
Solution: Queue Sort Plugin
The containers can be be distributed across different nodes - this can be done using Node Affinity

      spec:
        affinity:
          nodeAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
              nodeSelectorTerms:
              - matchExpressions:
                - key: kubernetes.io/hostname
                  operator: In
                  values:
                  - <node-name>  # Replace with your node name

and apply on that microservice: kubectl apply -f microservice-1.yaml
Solution: Split and Distribute Extender Plugin

Only CPU and RAM usage rates are considered in the service scheduling while latency or bandwidth usage rates are not considered at all.
Solution: Topology Aware Scoring Extender Plugin

Utilites

Verifying clusters and nodes:

kind get clusters
kubectl get nodes
Check Weavernet pods status: kubectl get pods -n kube-system -l name=weave-net
Check all pods are running: kubectl get pods -o wide
Check images on each node:

docker exec -it kind-control-plane crictl images | grep hello-world-node
docker exec -it kind-worker crictl images | grep hello-world-node
docker exec -it kind-worker2 crictl images | grep hello-world-node
docker exec -it kind-worker3 crictl images | grep hello-world-node
docker exec -it kind-worker4 crictl images | grep hello-world-node
docker exec -it kind-worker5 crictl images | grep hello-world-node

How to stop all services

kubectl delete deployment microservice1 microservice2 microservice3 microservice4 microservice5

kubectl delete service microservice1-service microservice2-service microservice3-service microservice4-service microservice5-service

kind delete cluster --name kind


- Verify cleanup

kubectl get all docker ps kind get clusters

valkyron / kubernetes-improvement-scope