Skip to content

Latest commit

 

History

History
122 lines (91 loc) · 7.67 KB

TESTING.md

File metadata and controls

122 lines (91 loc) · 7.67 KB
Raw

Testing

Open MCT Testing is iterating and improving at a rapid pace. This document serves to capture and index existing testing documentation and house documentation which no other obvious location as our testing evolves.

General Testing Process

Documentation located here

Unit Testing

Unit testing is essential part of our test strategy and complements our e2e testing strategy.

Unit Test Guidelines

  • Unit Test specs should reside alongside the source code they test, not in a separate directory.
  • Unit test specs for plugins should be defined at the plugin level. Start with one test spec per plugin named pluginSpec.js, and as this test spec grows too big, break it up into multiple test specs that logically group related tests.
  • Unit tests for API or for utility functions and classes may be defined at a per-source file level.
  • Wherever possible only use and mock public API, builtin functions, and UI in your test specs. Do not directly invoke any private functions. ie. only call or mock functions and objects exposed by openmct.* (eg. openmct.telemetry, openmct.objectView, etc.), and builtin browser functions (fetch, requestAnimationFrame, setTimeout, etc.).
  • Where builtin functions have been mocked, be sure to clear them between tests.
  • Test at an appropriate level of isolation. Eg.
    • If you’re testing a view, you do not need to test the whole application UI, you can just fetch the view provider using the public API and render the view into an element that you have created.
    • You do not need to test that the view switcher works, there should be separate tests for that.
    • You do not need to test that telemetry providers work, you can mock openmct.telemetry.request() to feed test data to the view.
    • Use your best judgement when deciding on appropriate scope.
  • Automated tests for plugins should start by actually installing the plugin being tested, and then test that installing the plugin adds the desired features and behavior to Open MCT, observing the above rules.
  • All variables used in a test spec, including any instances of the Open MCT API should be declared inside of an appropriate block scope (not at the root level of the source file), and should be initialized in the relevant beforeEach block. beforeEach is preferable to beforeAll to avoid leaking of state between tests.
  • A afterEach or afterAll should be used to do any clean up necessary to prevent leakage of state between test specs. This can happen when functions on window are wrapped, or when the URL is changed. A convenience function is provided for resetting the URL and clearing builtin spies between tests.

Unit Test Examples

Unit Testing Execution

The unit tests can be executed in one of two ways: npm run test which runs the entire suite against headless chrome npm run test:debug for debugging the tests in realtime in an active chrome session.

e2e, performance, and visual testing

Documentation located here

Code Coverage

It's up to the individual developer as to whether they want to add line coverage in the form of a unit test or e2e test.

Line Code Coverage is generated by our unit tests and e2e tests, then combined by (Codecov.io Flags), and finally reported in GitHub PRs by Codecov.io's PR Bot. This workflow gives a comprehensive (if flawed) view of line coverage.

Karma-istanbul

Line coverage is generated by our karma-coverage-istanbul-reporter package as defined in our karma.conf.js file:

    coverageIstanbulReporter: {
      fixWebpackSourcePaths: true,
      skipFilesWithNoCoverage: true,
      dir: 'coverage/unit', //Sets coverage file to be consumed by codecov.io
      reports: ['lcovonly']
    },

Once the file is generated, it can be published to codecov with

    "cov:unit:publish": "codecov --disable=gcov -f ./coverage/unit/lcov.info -F unit",

e2e

The e2e line coverage is a bit more complex than the karma implementation. This is the general sequence of events:

  1. Each e2e suite will start webpack with the npm run start:coverage command with config webpack.coverage.mjs and the babel-plugin-istanbul plugin to generate code coverage during e2e test execution using our custom baseFixture.
  2. During testcase execution, each e2e shard will generate its piece of the larger coverage suite. This coverage file is not merged. The raw coverage file is stored in a .nyc_report directory.
  3. nyc converts this directory into a lcov file with the following command npm run cov:e2e:report
  4. Most of the tests focus on chrome/ubuntu at a single resolution. This coverage is published to codecov with npm run cov:e2e:ci:publish.
  5. The rest of our coverage only appears when run against persistent datastore (couchdb), non-ubuntu machines, and non-chrome browsers with the npm run cov:e2e:full:publish flag. Since this happens about once a day, we have leveraged codecov.io's carryforward flag to report on lines covered outside of each commit on an individual PR.

Limitations in our code coverage reporting

Our code coverage implementation has some known limitations:

Troubleshooting CI

The following is an evolving guide to troubleshoot CI and PR issues.

Github Checks failing

There are a few reasons that your GitHub PR could be failing beyond simple failed tests.

  • Required Checks. We're leveraging required checks in GitHub so that we can quickly and precisely control what becomes and informational failure vs a hard requirement. The only way to determine the difference between a required vs information check is check for the (Required) emblem next to the step details in GitHub Checks.
  • Not all required checks are run per commit. You may need to manually trigger addition GitHub checks with a pr:<label> label added to your PR.

Flaky tests

(CircleCI's test insights feature)[https://circleci.com/blog/introducing-test-insights-with-flaky-test-detection/] collects historical data about the individual test results for both unit and e2e tests. Note: only a 14 day window of flake is available.

Local=Pass and CI=Fail

Although rare, it is possible that your test can pass locally but fail in CI.

Reset your workspace

It's possible that you're running with dependencies or a local environment which is out of sync with the branch you're working on. Make sure to execute the following:

nvm use
npm run clean
npm install

Run tests in the same container as CI

In extreme cases, tests can fail due to the constraints of running within a container. To execute tests in exactly the same way as run in CircleCI.

// Replace {X.X.X} with the current Playwright version 
// from our package.json or circleCI configuration file
docker run --rm --network host --cpus="2" -v $(pwd):/work/ -w /work/ -it mcr.microsoft.com/playwright:v{X.X.X}-focal /bin/bash
npm install

At this point, you're running inside the same container and with 2 cpu cores. You can specify the unit tests:

npm run test

or e2e tests:

npx playwright test --config=e2e/playwright-ci.config.js --project=chrome --grep <the testcase name>