gRPC Load Balance Test

This repo holds instructions and code to demonstrate a gRPC service that is load balanced in Kubernetes using a Gateway, implemented with traefik.

Minikube is used to run the demonstration locally.

By using the new Gateway API, GRPCRoute, and Traefik, gRPC services can be deployed and balanced. This gets past the need for tears.

Prereqs

• go
• kubectl
• minikube
• helm
• Docker

Running the demo

Create a cluster using minikube

1. Start the cluster. This will build a kubernetes cluster using a single node, locally. The method (Docker, VM, etc) is dependent on config and OS.

   This will set your kubectl context to the minikube cluster.

   minikube start
   

2. Enable addons

   minikube addons enable dashboard
   minikube addons enable metrics-server
   

3. Start the dashboard. This will block the terminal it is run in and open a web browser.

   minikube dashboard
   

4. Open a tunnel. This is used so the Gateway is accessible from the host machine. This also blocks the terminal it is run in.

   minikube tunnel
   

Install Traefik to the Cluster

1. Follow the instructions here to install in the cluster.
2. Configure Traefik to enable Gateway and not Ingress.

   helm upgrade --install --version 36.2.0 --namespace default traefik traefik/traefik -f values.yaml
   

Build and Deploy Image to Cluster

1. Build an image for the server

   docker build -f Dockerfile -t test-server .
   

2. Add the image to the cluster (no remote downloads).

   eval $(minikube docker-env -u)
   minikube image load test-server:latest
   

Deploy the Kubernetes manifests

1. Deploy

   kubectl apply -f manifests.yml
   

Configure /etc/hosts

1. Get the IP of the Gateway

   kubectl get Gateway traefik
   

2. Add to /etc/hosts

   123.123.123.123 echo.test
   

Ping the Server

1. Run the client

   go run client/main.go
   

Observe the Balance

1. In the browser tab that minikube dashboard opened, open "Pods" and watch as requests are balanced across all 5 pods. It may take a couple minutes for the metrics to update.

   

Observe the Lack of Balance

1. Port forward directly to the service

   kubectl port-forward svc/my-service 8000:8000
   

2. Run the client, pointed at the port forwarded.

   go run client/main.go "localhost:8000"
   

   Only one pod will get load now.

   

Cleaning Up

1. Delete the cluster

   minikube delete
   

Further Reading

• Gateway API
• GRPCRoute