performance

README

Performance test

We can use the perftool to do the performance test for the multicluster controlplane

Required

• One Kubernetes cluster, we will deploy the multicluster controlplane on that.
• Kind environment, we will create one kind cluster for performance test.

Run

1. export your Kubernetes cluster kubeconfig with KUBECONFIG.
2. run make test-performance to start the performance test.

Configuration

By default, this performance test creates 1000 clusters and generates 5 manifest works for each cluster, you can configure the --count and --work-count in the hack/performance.sh to modify the cluster and manifestwork conuts.

By default, we use the below manifestwork as the workload

apiVersion: work.open-cluster-management.io/v1
kind: ManifestWork
metadata:
  name: perftest-<cluster-name>-work-<work-number>
spec:
  workload:
    manifests:
    - apiVersion: v1
      kind: ConfigMap
      metadata:
        name: perftest-<cluster-name>-work-<work-number>
        namespace: default
      data:
        test-data: "I'm a test configmap"

You can customize your own workload by following steps

1. Define your manifestworks in a directory and use perftest.open-cluster-management.io/expected-work-count annotation to specify your expected manifestwork counts, e.g.

apiVersion: work.open-cluster-management.io/v1
kind: ManifestWork
metadata:
  name: work-1
  annotations:
    "perftest.open-cluster-management.io/expected-work-count": "10"
spec:	
  workload:
    manifests: <your-manifests>

---

apiVersion: work.open-cluster-management.io/v1
kind: ManifestWork
metadata:
  name: work-2
  annotations:
    "perftest.open-cluster-management.io/expected-work-count": "20"
spec:	
  workload:
    manifests: <your-manifests>

1. Use --work-template-dir flag to specify your manifestworks directory path, then the perftool will generate the manifest works with your expected work count on each cluster, and the generated work will use a configmap to save your manifestwork to simulate the same size with your customized workloads, for above example, the perftool will generate 10 manifest works for work-1 and 20 manifest works work-2 on each cluster, and the generated manifest work will be

apiVersion: work.open-cluster-management.io/v1
kind: ManifestWork
metadata:
  name: perftest-<cluster-name>-work-work-1-<work-number>
spec:	
  workload:
    - apiVersion: v1
      kind: ConfigMap
      metadata:
        name: perftest-<cluster-name>-work-1-<work-number>
        namespace: default
      data:
        test-data: |-
          <the content of work-1>

---

apiVersion: work.open-cluster-management.io/v1
kind: ManifestWork
metadata:
  name: <cluster-name>-work-work-2-<work-number>
spec:	
  workload:
    - apiVersion: v1
      kind: ConfigMap
      metadata:
        name: perftest-<cluster-name>-work-2-<work-number>
        namespace: default
      data:
        test-data: |-
          <the content of work-2>

Test Result

There is a doc to record our test result.