KubeCon CloudNativeCon EU 2024 - Saving the Planet One Cluster at a Time

For our experiment we have setup four scenarios

• Microservices using small nodes
• Microservices using big nodes
• A Monolith using Small nodes
• A monolith using big nodes

The idea behind this demonstartion is to show the affect of energy proportionality (i.e how utilized a node is) and the emissions that are generated from each scenario

This repository is for anyone that wants to replicate our demonstration

Repository Layout

Terraform

make sure the following APIs are enabled:

• Compute Engine API
• Cloud Resource Manager API
• Cloud DNS API
• Identity and Access Management (IAM) API
• Kubernetes Engine API

Manifests

Services for the cluster have been installed with manifests based on the instruction guides for each service. We have tried to make the installation as simple as possible, so in order to deploy the stack, you can simply run:

kubectl apply -k manifests

This should set up the experiment.

NOTE: we ran this demonstration on GKE therefore the node selection setup is based on how the cluster is generated in the terraform directory

Kepler

Installed Kepler the GKE cluster using manifests: https://sustainable-computing.io/installation/kepler/

1. Install monitoring stack

   • Prometheus with operator
   • kepler-exporter for grafana (requires yq) We have done this already for our installation, and running kubectl apply -f monitoring/ should deploy the monitoring stack and kepler dashboards.

2. Clone the kepler repo and run script to build manifests with the prometheus deployment from step 1: make build-manifest OPTS=PROMETHEUS_DEPLOY

Note: We have put our generated manifest files for the kepler deployment in the manifests/kepler dir

3. By default kepler uses eBPF to access power consumption values, but since we are running in GKE and the kernel is inaccessible, we need to enable the machine learning regression model. To do this edit the ConfigMap kepler-cfm in deployment.yaml file and update the MODEL_CONFIG value with the line MODEL_SERVER_ENABLE=true, so it looks like:

     MODEL_CONFIG: |
       CONTAINER_COMPONENTS_ESTIMATOR=false
       MODEL_SERVER_ENABLE=true
   

4. Make sure all of the kepler services deployed without issue to the kepler ns: kubectl get all -n kepler

5. Check the prometheus exporter is collecting metrics via kepler kubectl -n monitoring port-forward svc/prometheus-k8s 9090 Check the state is up for the kepler-exporter

6. Create Grafana dashboards based on collected data

Microservice-demo

The microservices-demo from GCP is a simple microservice boutique webshop.

The README.md in manifests/microservice-demo/ explains how to run it yourself Additionally, we added a kustomize overlay to specify a specific GKE nodepool to run the microservice on. More information on that can be found at manifests/microservice-demo/components/node-selector

Monolith Demo

We where not able to find a good example of a monolithic demo. Therefore as this is not a comparison between monoliths and microservices, but rather trying to understand how a single deployable large system contributes to emissions and where it performs better, we decided to take the previously described microservice demo and create a single pod with all the components in one. We understand that this is by no means an "optimized" setup, but it will be able to give us the data that we need.

The README for the Monlith Demo explains how to run it yourself Additionally, we added a kustomize overlay to specify a specific GKE nodepool to run the monolith on. More information on that can be found in the components README