testgrid/

Testgrid

Table of Contents

• Configuration
• Advanced Configuration
• Using the Client
• Unit Testing
• Merging Changes

The testgrid site is accessible at https://testgrid.k8s.io. The site is configured by config.yaml. Updates to the config are automatically tested and pushed to production.

Testgrid is composed of:

• A list of test groups that contain results for a job over time.
• A list of dashboards that are composed of tabs that display a test group
• A list of dashboard groups of related dashboards.

Configuration

Open config.yaml in your favorite editor and:

1. Configure the test groups
2. Add those testgroups to one or more tabs in one or more dashboards
3. Consider using dashboard groups if multiple dashboards are needed.

Test groups

Test groups contain a set of test results across time for the same job. Each group backs one or more dashboard tabs.

Add a new test group under test_groups:, specifying the group's name, and where the logs are located.

Ex:

test_groups:
- name: {test_group_name}
  gcs_prefix: kubernetes-jenkins/logs/{test_group_name}

See the TestGroup message in config.proto for additional fields to configure like days_of_results, tests_name_policy, notifications, etc.

Dashboards

Tabs

A dashboard tab is a particular view of a test group. Multiple dashboard tabs can view the same test group in different ways, via different configuration options. All dashboard tabs belong under a dashboard (see below).

Dashboards

A dashboard is a set of related dashboard tabs. The dashboard name shows up as the top-level link when viewing TestGrid.

Add a new dashboard under dashboards and a new dashboard tab under that.

Ex:

dashboards:
- name: {dashboard-name}
  dashboard_tab:
  - name: {dashboard-tab-name}
    test_group_name: {test-group-name}

See the Dashboard and DashboardTab messages in config.proto for additional configuration options, such as notifications, file_bug_template, description, code_search_url_template, etc.

Dashboard groups

A dashboard group is a set of related dashboards. When viewing a dashboard's tabs, you'll see the other dashboards in the Dashboard Group at the top of the client.

Add a new dashboard group, specifying names for all the dashboards that fall under this group.

Ex:

dashboard_groups:
- name: {dashboard-group-name}
  dashboard_names:
  - {dashboard-1}
  - {dashboard-2}
  - {dashboard-3}

Advanced configuration

See config.proto for an extensive list of configuration options. Here are some commonly-used ones.

More/Fewer Results

Specify days_of_results in a test group to increase or decrease the number of days of results shown.

test_groups:
- name: kubernetes-build
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-build
  days_of_results: 7

Tab descriptions

Add a short description to a dashboard tab describing its purpose.

  dashboard_tab:
  - name: gce
    test_group_name: ci-kubernetes-e2e-gce
    base_options: 'include-filter-by-regex=Kubectl%7Ckubectl'
    description: 'kubectl gce e2e tests for master branch'

Column headers

TestGrid shows date, build number, and k8s and test-infra commit shas above each run's results by default. To add your own custom column headers, add a key-value pair in your tests' metadata (see metadata for finished.json), and add the key for that pair as a configuration_value under column_header for your test group. Example:

test_groups:
- name: ci-kubernetes-e2e-gce-ubuntudev-k8sdev-default
  gcs_prefix:
  kubernetes-jenkins/logs/ci-kubernetes-e2e-gce-ubuntudev-k8sdev-default
  column_header:
  - configuration_value: node_os_image
  - configuration_value: master_os_image
  - configuration_value: Commit
  - configuration_value: infra-commit

Email alerts

In TestGroup, set num_failures_to_alert (alerts for consistent failures) and/or alert_stale_results_hours (alerts when tests haven't run recently). You can also set num_passes_to_disable_alert.

In DashboardTab, set alert_mail_to_addresses (comma-separated list of email addresses to send mail to).

These alerts will send whenever new failures are detected (or whenever the dashboard tab goes stale), and will stop when num_passes_to_disable_alert consecutive passes are found (or no failure is found in num_columns_recent runs).

# Send alerts to foo@bar.com whenever a test fails 3 times in a row, or tests
# haven't run in the last day.
test_groups:
- name: ci-kubernetes-e2e-gce
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-e2e-gce
  alert_stale_results_hours: 24
  num_failures_to_alert: 3

dashboards:
- name: google-gce
  dashboard_tab:
  - name: gce
    test_group_name: ci-kubernetes-e2e-gce
    alert_options:
      alert_mail_to_addresses: 'foo@bar.com'

Base options

Default to a set of client modifiers when viewing this dashboard tab.

# Show test cases from ci-kubernetes-e2e-gce, but only if the test has 'Kubectl' or 'kubectl' in the name.
  dashboard_tab:
  - name: gce
    test_group_name: ci-kubernetes-e2e-gce
    base_options: 'include-filter-by-regex=Kubectl%7Ckubectl'
    description: 'kubectl gce e2e tests for master branch'

More informative test names

If you run multiple versions of a test against different parameters, show which parameters they with after the test name.

# Show a test case as "{test_case_name} [{Context}]"
- name: ci-kubernetes-node-kubelet-benchmark
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-node-kubelet-benchmark
  test_name_config:
    name_elements:
    - target_config: Tests name
    - target_config: Context
    name_format: '%s [%s]'

Customize regression search

Narrow down where to search when searching for a regression between two builds/commits.

  dashboard_tab:
  - name: bazel
    description: Runs bazel test //... on the test-infra repo.
    test_group_name: ci-test-infra-bazel
    code_search_url_template:
      url: https://github.com/kubernetes/test-infra/compare/<start-custom-0>...<end-custom-0>

Notifications

Testgrid supports the ability to add notifications, which appears as a yellow butter bar / toast message at the top of the screen.

This is an effective way to broadcast system wide information (all FOO suites are failing due to blah, upgrade frobber to vX before the weekend, etc.)

Configure the list of notifications: under dashboard or testgroup: Each notification includes a summary: that defines the text displayed. Notifications benefit from including a context_link: url that can be clicked to provide more information.

Ex:

dashboards:
- name: k8s
  dashboard_tab:
  - name: build
    test_group_name: kubernetes-build
  notifications:  # Attach to a specific dashboard
  - summary: Hello world (first notification).
  - summary: Tests are failing to start (second notification).
    context_link: https://github.com/kubernetes/kubernetes/issues/123

or

test_groups:  # Attach to a specific test_group
- name: kubernetes-build
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-build
  notifications:
  - summary: Hello world (first notification)
  - summary: Tests are failing to start (second notification).
    context_link: https://github.com/kubernetes/kubernetes/issues/123

What Counts as 'Recent'

Configure num_columns_recent to change how many columns TestGrid should consider 'recent' for results. TestGrid uses this to calculate things like 'is this test stale?' (and hides the test).

test_groups:
- name: kubernetes-build
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-build
  num_columns_recent: 3

Ignore Pending Results

ignore_pending is false by default, which means that in-progress results will be shown if we have data for them. If you want to have these not show up, add:

test_groups:
- name: kubernetes-build
  gcs_prefix: kubernetes-jenkins/logs/ci-kubernetes-build
  ignore_pending: true

Using the client

Here are some quick tips and clarifications for using the TestGrid site!

Tab Statuses

TestGrid assigns dashboard tabs a status based on recent test runs.

• PASSING: No failures found in recent (num_columns_recent) test runs.
• FAILING: One or more consistent failures in recent test runs.
• FLAKY: The tab is neither PASSING nor FAILING. There is at least one recent failed result that is not a consistent failure.

Summary Widget

You can get a small widget showing the status of your dashboard tab, based on the tab statuses above! For example:

sig-testing-misc#bazel:

Inline it with:

<!-- Inline with a link to your tab -->
[![<dashboard_name>/<tab_name>](https://k8s-testgrid.appspot.com/q/summary/<dashboard_name>/<tab_name>/tests_status?style=svg)](https://k8s-testgrid.appspot.com/<dashboard_name>#<tab_name>)

Customizing Test Result Sizes

Change the size of the test result rectangles.

The three sizes are Standard, Compact, and Super Compact. You can also specify width=X in the URL (X > 3) to customize the width. For small widths, this may mean the date and/or changelist, or other custom headers, are no longer visible.

Filtering Tests

You can repeatedly add filters to include/exclude test rows. Under Options:

• Include/Exclude Filter by RegEx: Specify a regular expression that matches test names for rows you'd like to include/exclude.
• Exclude non-failed Tests: Omit rows with no failing results.

Grouping Tests

Grouped tests are summarized in a single row that is collapsible/expandable by clicking on the test name (shown as a triangle on the left). Under Options:

• Group by RegEx Mask: Specify a regular expression to mask a portion of the test name. Any test names that match after applying this mask will be grouped together.

• Group by Target: Any tests that contain the same target will be grouped together.

• Group by Hierarchy Pattern: Specify a regular expression that matches one or more parts of the tests' names and the tests will be grouped hierarchically. For example, if you have these tests in your dashboard:

  /test/dir1/target1
  /test/dir1/target2
  /test/dir2/target3
  

  By specifying regular expression "\w+", the tests will be organized into:

  ▼test
    ▼dir1
      target1
    ▼dir2
      target2
      target3
  

Sorting Tests

Under Options

• Sort by Failures: Tests with more recent failures will appear before other tests.
• Sort by Flakiness: Tests with a higher flakiness score will appear before tests with a lower flakiness score. The flakiness score, which is not reported, is based on the number of transitions from passing to failing (and vice versa) with more weight given to more recent transitions.
• Sort by Name: Sort alphabetically.

Unit testing

Run bazel test //testgrid/... to ensure the config is valid.

This finds common problems such as malformed yaml, a tab referring to a non-existent test group, a test group never appearing on any tab, etc.

Run bazel test //... for slightly more advanced testing, such as ensuring that every job in our CI system appears somewhere in testgrid, etc.

All PRs updating the configuration must pass prior to merging

Merging changes

Updates to the testgrid configuration are automatically pushed immediately when merging a change.

Manually convert the yaml file to the config proto with:

bazel run //testgrid/cmd/config -- \
  --yaml=testgrid/config.yaml \
  --print-text \
  --oneshot \
  --output=/tmp/config.pb \
  # Or push to gcs
  # --output=gs://my-bucket/config
  # --gcp-service-account=/path/to/foo.json