Kubernetes Test Infrastructure

The test-infra repository contains a collection of tools for testing Kubernetes and displaying Kubernetes tests results. See also CONTRIBUTING.md.

See the architecture diagram for an overview of how the different services interact.

Viewing test results

• The Kubernetes TestGrid shows historical test results
  • Configure your own testgrid dashboard at testgrid/config.yaml
  • Gubernator formats the output of each run
• PR Dashboard finds PRs that need your attention
• Prow schedules testing and updates issues
  • Prow responds to GitHub events, timers and manual commands given in GitHub comments.
  • The prow dashboard shows what it is currently testing
  • Configure prow to run new tests at config/jobs
• Triage Dashboard aggregates failures
  • Triage clusters together similar failures
  • Search for test failures across jobs
  • Filter down failures in a specific regex of tests and/or jobs
• Velodrome metrics track job and test health.
  • Kettle does collection, metrics does reporting, and velodrome is the frontend.

E2E Testing

Our e2e testing uses kubetest to build/deploy/test kubernetes clusters on various providers. Please see those documents for additional details about this tool as well as e2e testing generally.

Anyone can reconfigure our CI system with a test-infra PR that updates the appropriate files. Detailed instructions follow:

CI Job management

Create a new job

Bootstrap is deprecated, Please utilize the podutils to create new prowjobs.

Create a PR in this repo to add/update/remove a job or suite. Specifically you'll need to do the following:

• Add the job to the appropriate section in config/jobs

  • Directory Structure:
    • In general for jobs for github.com/org/repo use config/jobs/org/repo/filename.yaml
    • For Kubernetes repos we also allow config/jobs/kubernetes/sig-foo/filename.yaml
    • We use basename of the config name as a key in the prow configmap, so the name of your config file need to be unique across the config subdir
  • Type of jobs:
    • Presubmit jobs run on unmerged code in PRs
    • Postsubmit jobs run after merging code
    • Periodic job run on a timed basis
    • You can find more prowjob definitions at how-to-add-new-jobs
  • A simple sample job uses podutil looks like:

    - name: foo-repo-presubmit-test
      decorate: true
      spec:
        containers:
        - image: gcr.io/k8s-testimages/kubekins-e2e:latest-master
          command:
          - /path/to/cmd
          args:
          - positional
          - --and
          - flags

• Add the job name to the test_groups list in testgrid/config.yaml

  • Also the group to at least one dashboard_tab

The configs need to be sorted and kubernetes must be in sync with the security repo, or else presubmit will fail. You can run the script below to keep them valid:

hack/update-config.sh

Local testing

docker run your image locally, and mount in the repos you depend on.

Release branch jobs & Image validation jobs

Release branch jobs and image validation jobs are defined in test_config.yaml. We test different master/node image versions against multiple k8s branches on different features.

Those jobs are using channel based versions, current supported testing map is:

• k8s-dev : master
• k8s-beta : release-1.12
• k8s-stable1 : release-1.11
• k8s-stable2 : release-1.10
• k8s-stable3 : release-1.9

Our build job will generate a ci/(channel-name) file pointer in gcs.

After you update test_config.yaml, please run

bazel run //experiment:generate_tests -- --yaml-config-path=experiment/test_config.yaml

to regenerate the job configs.

We are moving towards making more jobs to fit into the generated config.

Presubmit will tell you if you forget to do any of this correctly.

Merge your PR and @k8s-ci-robot will deploy your change automatically.

Update an existing job

Largely similar to creating a new job, except you can just modify the existing entries rather than adding new ones.

Update what a job does by editing its definition in config/jobs.

Update where the job appears on testgrid by changing testgrid/config.yaml.

Delete a job

The reverse of creating a new job: delete the appropriate entries in config/jobs and testgrid/config.yaml.

Merge your PR and @k8s-ci-robot will deploy your change automatically.

Building and testing the test-infra

We use Bazel to build and test the code in this repo. The commands bazel build //... and bazel test //... should be all you need for most cases. If you modify Go code, run ./hack/update-bazel.sh to keep BUILD.bazel files up-to-date.

Contributing Test Results

The Kubernetes project encourages organizations to contribute execution of e2e test jobs for a variety of platforms (e.g., Azure, rktnetes). For information about how to contribute test results, see Contributing Test Results.

Other Docs

• kubernetes/test-infra dependency management