This repository contains some automation code to assist with performance testing for some aspects of the Kubernetes Operator SDK.
-
Install Kind.
-
Install Python dependencies:
pip3 install molecule openshift docker -
Run the benchmark with Molecule:
OPERATOR_UNDER_TEST=ansible molecule test -s kind
You can set OPERATOR_UNDER_TEST to one of the following:
ansiblegohelm
It is best to run each benchmark individually, in a clean Kind cluster, instead of running one after the other, to ensure the benchmark is as accurate as possible.
The kind Molecule test scenario manages a Kind instance, but the default scenario works with any functional Kubernetes cluster, as long as you can connect to it with a working KUBECONFIG file.
As an example, you can run the tests against Minikube:
-
Start a Minikube instance:
minikube start --cpus 4 --memory 8g -
By default, Minikube adds its configuration to
~/.kube/config, but if you override that path, make sure toexport KUBECONFIG=/path/to/overridden/kubeconfig. -
Run the benchmark with Molecule:
OPERATOR_UNDER_TEST=ansible molecule test
This project includes a set of playbooks to build an EC2 instance, install the prerequisites on it, then run the benchmarks on it.
See the ec2-testing README for details.
The Ansible playbook in main.yml uses a Kind cluster set up by Molecule to build and deploy the OPERATOR_UNDER_TEST into the cluster.
Then the playbook deploys the amount of Memcached CRs defined by memcached_cr_count in config.yml into the cluster.
Immediately after that, the playbook waits for all the expected Memcached instances to be up and running.
At the end of the playbook, Ansible's profile_tasks callback plugin reports back the timings of all the tasks. You can use these timings to benchmark the operators against each other.
This repository was created in 2020 by Jeff Geerling.