Skip to content

nutanix-cloud-native/openshift-cluster-api-actuator-pkg

 
 

Repository files navigation

cluster-api-actuator-pkg

Shared packages for Cluster API actuators.

Running the cluster autoscaler operator e2e tests against an OpenShift cluster

This test suite is designed to run against a full OpenShift 4 cluster. The test suite is agnostic of the hosting environment and the choice of cloud provider is up to the reader.

These instructions are written for the cluster autoscaler operator though should work for any project using the cluster-api-actuator-pkg.

Create a cluster

The easiest way to get a cluster to test against is to use an installer that supports Installer-Provisioned Infrastructure (IPI).

Instructions for creating an IPI cluster are available for the following cloud providers:

Deploy the code to test

Before making any changes to the cluster components you wish to test, you must disable the Cluster-Version Operator (CVO). If you do not disable the CVO, when you try to deploy your test code, the CVO will revert the component back to the original version.

To disable the CVO, scale its deployment to 0 replicas:

oc scale --replicas 0 -n openshift-cluster-version deployments/cluster-version-operator

Now deploy the code either directly into the cluster, or by running it locally.

Deploy the code to test within the cluster

To deploy your test code within the cluster, you must first build and push a container image to a repository. Once pushed, override the image within the deployment to deploy your code for testing:

oc set image -n machine-api-operator deployment/cluster-autoscaler-operator cluster-autoscaler-operator=<YOUR CONTAINER IMAGE>

Deploy the code to test locally

To deploy your test code locally, you must first disable the existing operator running within the OpenShift cluster:

oc scale --replicas 0 -n openshift-machine-api deployments/cluster-autoscaler-operator

Once the operator has been disabled, build your code to test locally and run it on your machine, pointing it to the cluster by passing the appropriate --kubeconfig flag:

make build
./bin/cluster-autoscaler-operator --kubeconfig=<PATH/TO/YOUR/CLUSTERS/KUBECONFIG>

Build the e2e tests

make build-e2e 

Run the autoscaler e2e tests

NAMESPACE=kube-system ./hack/ci-integration.sh -focus "Autoscaler should" -v --dry-run

Adjust -focus as appropriate.

Some example expected output:

Running Suite: Machine Suite
============================
Random Seed: 1617813491
Will run 4 of 33 specs

SSSSSSSSSSSSSSSSSS
------------------------------
[Feature:Machines] Autoscaler should use a ClusterAutoscaler that has 100 maximum total nodes count 
  It scales from/to zero
  /cluster-api-actuator-pkg/pkg/autoscaler/autoscaler.go:306
•
------------------------------
[Feature:Machines] Autoscaler should use a ClusterAutoscaler that has 100 maximum total nodes count 
  cleanup deletion information after scale down [Slow]
  /cluster-api-actuator-pkg/pkg/autoscaler/autoscaler.go:370
•
------------------------------
[Feature:Machines] Autoscaler should use a ClusterAutoscaler that has 12 maximum total nodes count and balance similar nodes enabled 
  scales up and down while respecting MaxNodesTotal [Slow][Serial]
  /cluster-api-actuator-pkg/pkg/autoscaler/autoscaler.go:521
•
------------------------------
[Feature:Machines] Autoscaler should use a ClusterAutoscaler that has 12 maximum total nodes count and balance similar nodes enabled 
  places nodes evenly across node groups [Slow]
  /cluster-api-actuator-pkg/pkg/autoscaler/autoscaler.go:605
•SSSSSSSSSSS
Ran 4 of 33 Specs in 0.000 seconds
SUCCESS! -- 0 Passed | 0 Failed | 0 Pending | 29 Skipped
PASS

Ginkgo ran 1 suite in 1.887727166s
Test Suite Passed

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Go 99.1%
  • Other 0.9%