Skip to content

Latest commit

 

History

History
237 lines (187 loc) · 10.7 KB

README.md

File metadata and controls

237 lines (187 loc) · 10.7 KB
Raw

Kubernetes Prow Job Configs

All prow job configs for prow.k8s.io live here.

They are tested and validated by tests in config/tests

Job Cookbook

This document attempts to be a step-by-step, copy-pastable guide to the use of prow jobs for the Kubernetes project. It may fall out of date. For more info, it was sourced from the following:

Job Types

There are three types of prow jobs:

  • Presubmits run against code in PRs
  • Postsubmits run after merging code
  • Periodics run on a periodic basis

Please see ProwJob docs for more info

Job Images

Where possible, we prefer that jobs use images that are pinned to a specific version, containing only what is needed.

Some good examples include:

Many jobs use gcr.io/k8s-testimages/foo images that are built from source in images/. Some of these have evolved organically, with way more dependencies than needed, and will be periodically bumped by PRs. These are sources of technical debt that are often not very well maintained. Use at your own risk, eg:

  • periodic-kubernetes-e2e-packages-pushed uses gcr.io/k8s-staging-test-infra/kubekins:latest-master to run ./tests/e2e/packages/verify_packages_published.sh which ends up running apt-get and yum commands. Perhaps a debian image would be better.

Job Presets

Prow supports Presets to define and patch in common env vars and volumes used for credentials or common job config. Some are defined centrally in [config/prow/config.yaml], while others can be defined in files here. eg:

Secrets

Prow jobs can use secrets located in the same namespace within the cluster where the jobs are executed, by using the same mechanism of podspec. The secrets used in prow jobs can be source controlled and synced from any major secret manager provider, such as google secret manager, see prow_secret for instructions.

Job Examples

A presubmit job named "pull-community-verify" that will run against all PRs to kubernetes/community's master branch. It will run make verify in a checkout of kubernetes/community at the PR's HEAD. It will report back to the PR via a status context named pull-kubernetes-community. Its logs and results are going to end up in GCS under kubernetes-jenkins/pr-logs/pull/community. Historical results will display in testgrid on the sig-contribex-community dashboard under the pull-verify tab

presubmits:
  kubernetes/community:
  - name: pull-community-verify  # convention: (job type)-(repo name)-(suite name)
    annotations:
      testgrid-dashboards: sig-contribex-community
      testgrid-tab-name: pull-verify
    branches:
    - master
    decorate: true
    always_run: true
    spec:
      containers:
      - image: public.ecr.aws/docker/library/golang:1.12.5
        command:
        - /bin/bash
        args:
        - -c
        # Add GOPATH/bin back to PATH to workaround #9469
        - "export PATH=$GOPATH/bin:$PATH && make verify"

A periodic job named "periodic-cluster-api-provider-aws-test-creds" that will run every 4 hours against kubernetes-sigs/cluster-api-provider-aws's master branch. It will run ./scripts/ci-aws-cred-test.sh in a checkout of the repo located at sigs.k8s.io/cluster-api-provider-aws. The presets it's using will ensure it has aws credentials and aws ssh keys in well known locations. Its logs and results are going to end up in GCS under kubernetes-jenkins/logs/periodic-cluster-api-provider-aws-test-creds. Historical results will display in testgrid on the sig-cluster-lifecycle-cluster-api-provider-aws dashboard under the test-creds tab

It's using the kubekins-e2e image which isn't recommended, but works for now.

periodics:
- name: periodic-cluster-api-provider-aws-test-creds
  annotations:
    testgrid-dashboards: sig-cluster-lifecycle-cluster-api-provider-aws
    testgrid-tab-name: test-creds
  decorate: true
  interval: 4h
  labels:
    preset-service-account: "true"
    preset-aws-ssh: "true"
    preset-aws-credential: "true"
  extra_refs:
  - org: kubernetes-sigs
    repo: cluster-api-provider-aws
    base_ref: master
    path_alias: "sigs.k8s.io/cluster-api-provider-aws"
  spec:
    containers:
    - image: gcr.io/k8s-staging-test-infra/kubekins-e2e:v20240923-c8645c1a17-master
      command:
      - "./scripts/ci-aws-cred-test.sh"

Adding or Updating Jobs

  1. Find or create the prowjob config file in this directory
    • In general jobs for github.com/org/repo use org/repo/filename.yaml
    • For kubernetes/kubernetes we prefer kubernetes/sig-foo/filename.yaml
    • Ensure filename.yaml is unique across the config subdir; prow uses this as a key in its configmap
  2. Ensure an OWNERS file exists in the directory for job, and has appropriate approvers/reviewers
  3. Write or edit the job config (please see How to configure new jobs)
  4. Ensure the job is configured to display its results in testgrid.k8s.io
  5. Open a PR with the changes; when it merges @k8s-ci-robot will deploy the changes automatically

Deleting Jobs

  1. Find the prowjob config file in this directory
  2. Remove the entry for your job; if that was the last job in the file, remove the file
  3. If the job had no testgrid annotations, ensure its [testgrid/config.yaml] entries are gone
  4. Open a PR with the changes; when it merges @k8s-ci-robot will deploy the changes automatically

Testing Jobs

You can read about how to test changes to ProwJobs both locally and remotely in the prow documentation.

Testing Jobs Remotely

This requires a running instance of prow. In general, we discourage the use of prow.k8s.io as a testbed for job development, and recommend the use of your own instance of prow for faster iteration. That said, an approach that people have used in the past with mostly-there jobs is to iterate via PRs; just recognize this is going to depend on review latency.

Running a Production Job

Normally prow will automatically schedule your job, however if for some reason you need to trigger it again and are a Prow administrator you have a few options:

  • you can use the rerun feature in prow.k8s.io to run the job again with the same config
  • you can use config/mkpj.sh to create a prowjob CR from your local config
  • you can use bazel run //prow/cmd/mkpj -- --job=foo ... to create a prowjob CR from your local config

For the latter two options you'll need to submit the resulting CR via kubectl configured against the prow services cluster.

Generated Jobs

There are some sets of jobs that are generated and should not be edited by hand. These specific instructions should probably just live adjacent to the jobs rather than in this central README, but here we are for now.

image-validation jobs

These test different master/node image versions against multiple k8s branches. If you want to change these, update releng/test_config.yaml and then run

# from test-infra root
$ ./hack/update-generated-tests.sh

release-branch jobs

When a release branch of kubernetes is first cut, the current set of master jobs must be forked to use the new release branch. Use releng/config-forker to accomplish this, eg:

# from test-infra root
$ go run ./releng/config-forker \
  --job-config $(pwd)/config/jobs \
  --version 1.27 \
  --go-version 1.20.2 \
  --output $(pwd)/config/jobs/kubernetes/sig-release/release-branch-jobs/1.27.yaml