Skip to content

Kubernetes operator to measure TCP transmission speed between 2 pods

License

Notifications You must be signed in to change notification settings

piontec/netperf-operator

Repository files navigation

Go Report Card Build status

Netperf operator

This is a simple Kubernetes Operator, that uses the legendary netperf tool to run 2 pods in your cluster and measure real TCP connection throughput between them. This project is an example kubernetes operator based on Operator Framework SDK. I created it for two reasons: first one, to learn how to build an operator using the SDK, second to solve a network testing problem. This README.md file is about how to develop, build and run the project. If you wanto to learn more about how it works and how to write Kubernetes operators/controllers, head to my blog.

Installing

Note: for installation for development, check Developers guide

You need to deploy the controller, its Custom Resource Definition and RBAC resources:

kubectl create -f deploy/crd.yaml
kubectl create -f deploy/rbac.yaml
kubectl create -f deploy/operator.yaml

Users guide

The controller runs tests only in a single namespace, in which the controller is deployed. In this namespace, you have to create the following resource:

apiVersion: "app.example.com/v1alpha1"
kind: "Netperf"
metadata:
  name: "example"
spec:
  serverNode: "minikube"
  clientNode: "minikube"

Wait for the Netperf object to complete (status: Done) and check the measured throughput.

If you skip any of the serverNode or clientNode in spec:, they will be normally chosen and assigned by kube's scheduler. If you configure them, node affinity will be used to run on the specific node.

Developers guide

There are 2 ways you can build and run the operator:

  • for rapid development and testing: run the operator process outside of cluster, on your development machine, with kubectl configured to access your cluster
  • for deploying to cluster and testing full deployment and in cluster operation: you need to build a container image with the operator and deploy it to the cluster

Common tasks

In both cases, you need to start with following steps:

  • you need to have a functional golang build environment
  • you need to have dep installed
  • clone this repo
mkdir $GOPATH/src/github.com/piontec
cd $GOPATH/src/github.com/piontec
git clone https://github.com/piontec/netperf-operator.git
cd netperf-operator
  • install dependencies
dep ensure
  • create required Custom Resource Definition in the cluster
kubectl create -f deploy/crd.yaml

Running outside of cluster

To build the plugin on your machine, you have to build it with

go build cmd/netperf-operator/main.go

To run it, you have to set the following 2 environment variables that point to your .kube/config file and define the name of a Namespace to monitor:

export KUBERNETES_CONFIG=/your/path/.kube/config
export WATCH_NAMESPACE=default

Still, if you're using VS Code with go plugin (which I highly recommend), I incuded in the repo my launch.json file. Skip the manual steps above, open the project, set your environment variables in the launch.json file and hit F5 in VS Code - you're ready to go.

Running inside a cluster

The simplest way is through installing the operator sdk. You also need an image repository, where you can store the ready image:

export IMAGE=my.repo.url/netperf-operator:v0.0.1

Then, build it and push to registry:

operator-sdk build $IMAGE
docker push $IMAGE

As last step, deploy the RBAC definition for the controller and a Deployment that will run it:

kubectl create -f deploy/rbac.yaml
kubectl create -f deploy/operator.yaml

About

Kubernetes operator to measure TCP transmission speed between 2 pods

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published