How It Works

The controller:

1. Monitors the creation, update, and deletion of PredictiveAutoscaler custom resources
2. On a schedule (every 5 minutes by default), collects current pod counts
3. Calculates required pods using the predictive formula
4. Updates the target HPA's minReplicas
5. Updates the historical data with the weighted formula
6. Stores this data persistently for future predictions

This implementation provides a clean separation of code and configuration, making it easier to develop, test, and maintain the controller.

Deployment Instructions

Build and Push the Controller Image

Navigate to the controller directory cd CSCI555/Scaler/controller

Build the Docker image docker build -t predictive-autoscaler:latest .

Tag and push to your container registry

    docker tag predictive-autoscaler:latest anirudhr120100/csci555-predictive-autoscaler:latest
    docker push anirudhr120100/csci555-predictive-autoscaler:latest

Update the controller deployment with your image registry

Edit /Scaler/deploy/controller-deployment.yaml

Apply CRD first

kubectl apply -f Scaler/crd/predictive-autoscaler-crd.yaml

Apply RBAC

kubectl apply -f Scaler/deploy/rbac.yaml

Apply controller deployment

kubectl apply -f Scaler/deploy/controller-deployment.yaml

Check if the controller is running

kubectl get pods -l app=predictive-autoscaler-controller

Apply the PredictiveAutoscaler instance

kubectl apply -f Scaler/deploy/predictive-autoscaler-instance.yaml

Check if the PredictiveAutoscaler custom resource was created

kubectl get predictiveautoscalers or kubectl get pa (using the short name)

Test the controller

Create the directory in your persistent volume if needed

kubectl exec -it predictive-autoscaler-controller-xxx -- mkdir -p /data

Copy the sample data file to your controller pod

kubectl cp Scaler/test-data/realistic-traffic.json predictive-autoscaler-controller-xxx:/data/default_simpleweb-predictor_history.json

Check the logs to see if the controller is using the data correctly

kubectl logs -f deployment/predictive-autoscaler-controller

Check if the HPA is being updated

kubectl get hpa

Check if the historical data is being updated

kubectl get predictiveautoscalers

Kubernetes Cluster Deployment instructions on Cloudlab

Cloudlabs has pre-existing profiles. Select the K8s

1. Click Experiment -> Start experiment
2. Change Profile to K8s profile from select window
3. Click confirm and next to the parameterize page.
4. Make edits to the parameters (optional)
5. Click next to the Finalize Page. -> Assign a name and cluster location
6. Click next to the schedule page.
7. Pick a time to deploy and click next

Once the cluster starts, click "extend" to extend the cluster expiration by 7 days.

Starting K8s and docker on the cluster

SSH to node-0

Click on the node in the node graph click on "Shell" from the pop-up menu

Install Docker

Only do the following if docker is not installed.

Dependencies: sudo apt install -y apt-transport-https ca-certificates curl software-properties-common

Docker GPG keys curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo gpg --dearmor -o /usr/share/keyrings/docker-archive-keyring.gpg

Create the docker repository echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/docker-archive-keyring.gpg] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable" | sudo tee /etc/apt/sources.list.d/docker.list

Install docker: sudo apt install -y docker-ce docker-ce-cli containerd.io

Start docker

    sudo systemctl enable docker
    sudo systemctl start docker

Install Kubectl

Only do the following if Kubernetes is not installed

    curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
    sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl
    kubectl version --client

Clone GitHub repo

Git installation

sudo apt install -y git

Clone the repo

git clone https://github.com/CSCI555-Spring25/Scaler.git

Building and registering docker container on the registry

1. navigate to "webserver" folder
2. Create docker container using docker build -t simpleweb:latest .
3. Find the IP of the node/registry using the following command:
   docker ps | grep "registry:2"
4. If the local registry is not running, run the following to start the registry and go back to step 3: docker run -d -p 5000:5000 --name registry registry:2
5. Once IP is known, for example 10.10.1.1 Tag and push the docker container to the registry

       docker tag simpleweb:latest 10.10.1.1:5000/simpleweb:latest
       docker push 10.10.1.1:5000/simpleweb:latest

6. If the ip address is not 10.10.1.1:5000, edit spec/template/spec/containers/image to have the correct IP address

Starting kubernetes service

1. navigate to root directory of Scaler github repo
2. Start the webserver with kubernetes: kubectl apply -f echo-server.yaml
3. Check status of the running pods with kubectl get pods

Exposing Service

kubectl expose deployment simpleweb-deployment --port=80 --type=LoadBalancer

kubectl not returning nodes

If kubectl is giving an error while returning nodes, make sure there is a K8s profile for your user on cloud labs. Use following commands to copy the admin profile into your user directory.

    sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config
    sudo chown $(id -u):$(id -g) $HOME/.kube/config