simple example of running netperf in k8s cluster

Simple project to get started with latency testing in kubernetes cluster. Similar concept can be used with different tools e.g. iperf, ping, ...

This is just a template project, feel free to update .yaml manifests to include node selectors, tolerations etc.

setup

You can skip this setup if you are going to run tests on an existing cluster

• start kind cluster kind create cluster --name netperf --config kind-config.yaml
• build netperf docker image docker build -t pete911/netperf:latest .
• load netperf docker image to the cluster kind load docker-image pete911/netperf:latest --name netperf

install

• install netperf deployments:
  • daemonset (one server, client on every node) kubectl apply -f k8s-netperf-ds.yaml
  • server and client as containers in a single pod kubectl apply -f k8s-netperf-single-pod.yaml

test

pod to pod

• get netperf server
  • pod name, node name and pod IP kubectl get pod -l app.kubernetes.io/name=netperf-server -o=custom-columns=NAME:.metadata.name,NODE:.spec.nodeName,IP:.status.podIP
  • or more info if needed kubectl get pod -l app.kubernetes.io/name=netperf-server -o wide
• get netperf clients
  • pod and node name kubectl get pod -l app.kubernetes.io/name=netperf-client -o=custom-columns=NAME:.metadata.name,NODE:.spec.nodeName
  • or more info if needed kubectl get pod -l app.kubernetes.io/name=netperf-client -o wide
• exec to a client pod (either on the same node as server or different) kubectl exec -it <pod-name> -- sh (replace <pod-name>)
• run netperf against the server netperf -H <server-pod-IP> -l 100 -t TCP_RR -- -o min_latency,mean_latency,max_latency (replace <server-pod-IP>)

container to container

• get netperf pod kubectl get pod -l app.kubernetes.io/name=netperf
• exec to a client container kubectl exec -it <pod-name> -c client -- sh (replace <pod-name>)
• run netperf against the server netperf -H 127.0.0.1 -l 100 -t TCP_RR -- -o min_latency,mean_latency,max_latency

cleanup

• daemonset
  • client kubectl delete ds netperf-client
  • server kubectl delete deploy netperf-server
• single pod deployment kubectl delete deploy netperf

If you used kind cluster (setup sections), run kind delete cluster --name netperf as well