Background:

Customers of all sizes need to understand how OpenShift will perform on the hardware they have. The best way to understand if the cluster will manage the expected load would be to run benchmark tests against the actual hardware.

The downside to benchmarking a cluster is the possibility that some of the benchmarking tests could result in an undesirable state from failed cleanup, unready nodes, etc. For this reason, it seems reasonable that the installer recognizes the users intent to only benchmark the cluster. A benchmark cluster should be considered ephemeral and nonsupported after benchmark testing.

internally the ability to run benchmark clusters will simplify baselining configurations and catching regressions.

Goal:

Provide standardized OpenShift driven benchmark tooling utilizing the existing installer workflow.

Functionality:

• based on the number of worker replicas run a performance test on the cluster after install complete from the bootstrap node utilizing the existing `OPENSHIFT_INSTALL_PRESERVE_BOOTSTRAP`

Previous Work (performance testing)

• https://github.com/tkashem/graceful
• https://github.com/cloud-bulldozer/kube-burner

Example Usage

openshift-install create benchmark