Skip to content

Latest commit

 

History

History
52 lines (42 loc) · 4.67 KB

life_of_a_prow_job.md

File metadata and controls

52 lines (42 loc) · 4.67 KB
Raw

Life of a Prow Job

I comment /test all on a PR. GitHub creates a JSON object representing that action and sends a webhook to prow. Here are some examples of webhook payloads.

The webhook finds its way into the cluster with the help of an Ingress object. This object has a rule stating that /hook goes to the hook service which is backed by the hook deployment and finally into the hook program itself. The hook program deserializes the payload into the appropriate go struct, and then passes it along to each plugin.

To each plugin, hook passes two objects: a plugin client and the deserialized GitHub event. In this case, the event is an IssueCommentEvent. The plugin client has a Kubernetes client, a GitHub client, and the prow config:

type PluginClient struct {
	GitHubClient github.Client
	KubeClient   *kube.Client
	Config       *config.Config
	Logger       *logrus.Entry
}

The trigger plugin has a function called handleIC that it registered at init-time. This function checks the issue comment for any test requests by comparing the issue comment body to the list of jobs supplied in the Config object in the PluginClient. I requested that some tests run, so the trigger plugin talks to the Kubernetes API server and creates a ProwJob Custom Resource with the information from the issue comment. At this point, cmd/hook's role is done. It goes on and handles the next webhooks.

The ProwJob that it creates looks a little like this, although I elided some of the pod details:

apiVersion: prow.k8s.io/v1
kind: ProwJob
metadata:
  name: 32456927-35d9-11e7-8d95-0a580a6c1504
spec:
  job: pull-test-infra-bazel
  decorate: true
  pod_spec:
    containers:
    - image: gcr.io/k8s-testimages/bazelbuild:0.11
  refs:
    base_ref: master
    base_sha: 064678510782db5b382df478bb374aaa32e577ea
    org: kubernetes
    pulls:
    - author: ixdy
      number: 2716
      sha: dc32ccc9ea3672ccc523b7cbaa8b00360b4183cd
    repo: test-infra
  type: presubmit
status:
  startTime: 2017-05-10T23:34:22.567457715Z
  state: triggered

Every thirty seconds, cmd/plank runs a Sync function that lists ProwJobs and Kubernetes pods in the cluster. For each ProwJob, it runs syncKubernetesJob. This function notices that the above ProwJob does not have a corresponding Kubernetes pod, so it creates one in the test-pods namespace using the pod spec in the ProwJob. It also injects environment variables into the container such as PULL_NUMBER and JOB_NAME. It also updates the status line on the PR with a link to Gubernator, which will provide real-time logs and results.

Several minutes later, the syncKubernetesJob method notices that the Kubernetes pod in that ProwJob has finished and succeeded. It sets the ProwJob status to success and sets the status line on GitHub to success. A green check-mark shows up on the PR.

A day later, cmd/sinker notices that the job and pod are a day old and deletes them from the Kubernetes API server.