Skip to content

Latest commit

 

History

History
149 lines (103 loc) · 6.52 KB

README.md

File metadata and controls

149 lines (103 loc) · 6.52 KB

E2E performance regression tests

This directory contains the scripts and configuration files for running the performance regression tests. These tests are called E2E tests because they run the app on a real device (physical or emulated).

Example of a e2e test run

To run the e2e tests:

  1. Connect an android device. The tests are currently designed to run only on android. It can be a physical device or an emulator.

  2. Make sure Fastlane was initialized by running bundle install

  3. Run the tests with npm run test:e2e.

💡 Tip: To run the tests locally faster, and you are only making changes to JS, it's recommended to build the app once with npm run android-build-e2e and from then on run the tests with npm run test:e2e -- --buildMode js-only. This will only rebuild the JS code, and not the whole native app!

Ideally you want to run these tests on your branch before you want to merge your new feature to main.

Available CLI options

The tests can be run with the following CLI options:

  • --config: Extend/Overwrite the default config with your values, e.g. --config config.local.js
  • --includes: Expects a string/regexp to filter the tests to run, e.g. --includes "login|signup"
  • --skipInstallDeps: Skips the npm install step, useful during development
  • --development: Applies some default configurations:
    • Sets the config to config.local.js, which executes the tests with fewer iterations
    • Runs the tests only on the current branch
  • --buildMode: There are three build modes, the default is full:
    1. full: rebuilds the full native app in (e2e) release mode
    2. js-only: only rebuilds the js bundle, and then re-packages the existing native app with the new package. If there is no existing native app, it will fallback to mode "full"
    3. skip: does not rebuild anything, and just runs the existing native app
  • --skipCheckout: Won't checkout any baseline or comparison branch, and will just run the tests

Available environment variables

The tests can be run with the following environment variables:

  • baseline: Change the baseline to run the tests again (default is main).

Performance regression testing

The output of the tests is a set of performance metrics (see video above). The test will tell you if the performance significantly worsened for any test case.

For this to work you need a baseline you test against. The baseline is set by default to the main branch.

The test suite will run each test-case twice, once on the baseline, and then on the branch you are currently on.

It will run the tests of a test case multiple time to average out the results.

Adding tests

To add a test checkout the designed guide.

Structure

For the test suite, no additional tooling was used. It is made of the following components:

  • The tests themselves :

    • The tests are located in src/libs/E2E/tests
    • As opposed to other test frameworks, the tests are inside the app, and execute logic using app code (e.g. navigationRef.navigate('Signin'))
    • For the tests there is a custom entry for react native, located in src/libs/E2E/reactNativeLaunchingTest.js
  • The test runner:

    • Orchestrates the test suite.
    • Runs the app with the tests on a device
    • Responsible for gathering and comparing results
    • Located in e2e/testRunner.js.
  • Test server:

    • A nodeJS application that starts an HTTP server.
    • Receives test results from the app.
    • Located in e2e/server.
  • Client:

    • Client-side code (app) for communication with the test server.
    • Located in src/libs/E2E/client.js.

How a test gets executed

There exists a custom android entry point for the app, which is used for the e2e tests. The entry file used is src/libs/E2E/reactNativeEntry.js, and here we can add our test case.

The test case should only execute its test once. The test runner is responsible for running the test multiple time to average out the results.

Any results of the test (which is usually a duration, like "Time it took to reopen chat", or "TTI") should be submitted to the test server using the client:

import E2EClient from './client';

// ... run you test logic
const someDurationWeCollected = // ...

E2EClient.submitTestResults({
    name: 'My test name',
    duration: someDurationWeCollected,
});

Submitting test results doesn't automatically finish the test. This enables you do submit multiple test results from one test (e.g. measuring multiple things at the same time).

To finish a test call E2EClient.submitTestDone().

Android specifics

The tests are designed to run on android (although adding support for iOS should be easy to add). To test under realistic conditions during the tests a release build is used.

However, to be able to call our local HTTP test server, we need to allow cleartext http traffic. Therefore, a customized release build type is needed, which is called e2eRelease. This build type has clear text traffic enabled but works otherwise just like a release build.

In addition to that, another entry file will be used (instead of index.js). The entry file used is src/libs/E2E/reactNativeEntry.js. By using a custom entry file we avoid bundling any e2e testing code into the actual release app.

For the app to detect that it is currently running e2e tests, an environment variable called E2E_TESTING=true must be set. There is a custom environment file in e2e/.env.e2e that contains the env setup needed. The build automatically picks this file for configuration.

It can be useful to debug the app while running the e2e tests (to catch errors durign development of a test). You can simply add the debuggable true property to the e2eRelease buildType config in android/app/build.gradle. Then rebuild the app. You can now monitor the app's logs using logcat (adb logcat | grep "ReactNativeJS").

Test the accuracy of the test suite

If you run the tests on the same branch, the result should ideally be a difference of 0%. However, as the tests get executed on an emulator, there will be some variance. This variance is mitigated by using statistical tools such as z-test. However, when running on the same branch, the results should be below 5% of change. You might want to tweak the values in e2e/config.js to adjust those values.