End-to-end tests with Cypress

Metabase uses Cypress for “end-to-end testing”, that is, tests that are executed against the application as a whole, including the frontend, backend, and application database. These tests are essentially scripts written in JavaScript that run in the web browser: visit different URLs, click various UI elements, type text, and assert that things happen as expected (for example, an element appearing on screen, or a network request occuring).

Getting Started

Metabase’s Cypress tests are located in the e2e/test/scenarios source tree, in a structure that roughly mirrors Metabase’s URL structure. For example, tests for the admin “datamodel” pages are located in e2e/test/scenarios/admin/datamodel.

Our custom Cypress runner builds its own backend and creates a temporary H2 app db. Both are destroyed when this process is killed. The reserved default port is 4000 on the local host. There is nothing stopping you from running your local Metabase instance on localhost:3000 at the same time. This might even be helpful for debugging purposes.

Standard Development Flow

1. Continuously build the frontend

   a. If you need only the frontend, run yarn build-hot

   b. If you want to run a local Metabase instance alongside Cypress, the easiest way to achieve this is by using yarn dev or yarn dev-ee (both rely on frontend hot reloading under the hood)

2. In a separate terminal session (without killing the previous one) run yarn test-cypress-open. This will open a Cypress GUI that will let you choose which tests to run. Alterantively, take a look at more running options below.

Running Options

To run all Cypress tests programmatically in the terminal:

yarn run test-cypress-run

You can run a specific set of scenarios by using a custom --folder flag, which will pick up the chosen scenarios under e2e/test/scenarios/.

yarn run test-cypress-run --folder sharing

You can quickly test a single file only by using the official --spec flag.

yarn test-cypress-run --spec e2e/test/scenarios/question/new.cy.spec.js

You can specify a browser to execute Cypress tests in using the --browser flag. For more details, please consult the official documentation.

Specifying a browser makes most sense when running Cypress in a run mode. On the other hand, Cypress open mode (GUI) allows one to easily switch between all available browsers on the system. However, some people prefer to specify a browser even in this scenario. If you do this, keep in mind that you are merely preselecting an initial browser for Cypress but you still have the option to choose a different one.

Anatomy of the Test

Cypress test files are structured like Mocha tests, where describe blocks are used to group related tests, and it blocks are the tests themselves.

describe("homepage",() => {
  it('should load the homepage and...', () => {
    cy.visit("/metabase/url");
    // ...
  })
})

We strongly prefer using selectors like cy.findByText() and cy.findByLabelText() from @testing-library/cypress since they encourage writing tests that don’t depend on implementation details like CSS class names.

Try to avoid repeatedly testing pieces of the application incidentally. For example, if you want to test something about the query builder, jump straight there using a helper like openOrdersTable() rather than starting from the home page, clicking “New”, then “Question”, etc.

Cypress Documentation

• Introduction: https://docs.cypress.io/guides/core-concepts/introduction-to-cypress.html
• Commands: https://docs.cypress.io/api/api/table-of-contents.html
• Assertions: https://docs.cypress.io/guides/references/assertions.html

Tips/Gotchas

contains vs find vs get

Cypress has a set of similar commands for selecting elements. Here are some tips for using them:

• contains is (by default) case-sensitive to the text in the DOM. If it’s not matching text you’d expect, check that CSS hasn’t updated the case. You can explicitly tell it to ignore the case with the following option { matchCase: false }.
  • contains matches substrings. Given two strings “filter by” and “Add a filter”, cy.contains(“filter”); will match both. To avoid these issues, you can either pass a regexp that pins the start/end of the string or scope a string to a specific selector: cy.contains(selector, content);.
• find will let you search within your previous selection.
• get will search the entire page even if chained, unless you explicitly tweak the withinSubject option.

How to access Sample Database tables and field IDs?

The Sample Database that we use in E2E tests can change at any time, and with it the references to its tables and fields. Never ever use hard coded numeric references to those IDs. We provide a helpful mechanism to achieve this that is guaranteed to produce correct results. Every time you spin Cypress up, it fetches the information about the Sample Database, extracts table and field IDs and writes that to the e2e/support/cypress_sample_database JSON that we then re-export and make available to all tests.

// Don't
const query = {
  "source-table": 1,
  aggregation: [["count"]],
  breakout: [["field", 7, null]],
}

// Do this instead
import { SAMPLE_DATABASE } from "e2e/support/cypress_sample_database";
const { PRODUCTS, PRODUCTS_ID } = SAMPLE_DATABASE;

const query = {
  "source-table": PRODUCTS_ID,
  aggregation: [["count"]],
  breakout: [["field", PRODUCTS.CATEGORY, null]],
}

Increase viewport size to avoid scrolling

Sometimes Metabase views are a bit large for Cypress’ default 1280x800 viewport. This can require you to scroll for tests to work. For example, virtualized tables will not even render the contents outside of the viewport. To avoid these problems, increase the viewport size for a specific test. Unless you’re specifically testing how the application behaves on a window resize, please avoid using the cy.viewport(width, height); in the middle of the test. Set the viewport width/height using the optional Cypress test config instead. This config works with both describe and it blocks.

describe("foo", { viewportWidth: 1400 }, () => {});

it("bar", { viewportWidth: 1600, viewportHeight: 1200 }, () => {})

Code reloading vs test reloading

When you edit a Cypress test file, the tests will refresh and run again. However, when you edit a code file, Cypress won’t detect that change. If you’re running yarn build-hot, the code will rebuild and update within Cypress. You’ll have to manually click rerun after the new code has loaded.

Inspecting while the “contains helper” is open

One great feature of Cypress is that you can use the Chrome inspector after each step of a test. They also helpfully provide a helper that can test out contains and get calls. This helper creates new UI that prevents inspecting from targeting the correct elements. If you want to inspect the DOM in Chrome, you should close this helper.

Putting the wrong HTML template in the Uberjar

yarn build and yarn build-hot each overwrite an HTML template to reference the correct JavaScript files. If you run yarn build before building an Uberjar for Cypress tests, you won’t see changes to your JavaScript reflected even if you then start yarn build-hot.

Running Cypress on M1 machines

You might run into problems when running Cypress on M1 machine. This is caused by the @bahmutov/cypress-esbuild-preprocessor that is using esbuild as a dependency. The error might look like this. The solution is to install NodeJS using one of the Node version managers like nvm or n.

Another issue you will almost surely face is the inability to connect to our Mongo QA Database. You can solve it by providing the following env:

export EXPERIMENTAL_DOCKER_DESKTOP_FORCE_QEMU=1

Running tests that depend on Docker images

A subset of our tests depend on the external services that are available through the Docker images. At the time of this writing, those are the three supported external QA databases, Webmail, Snowplow and LDAP servers. It’s tedious to have all these Docker containers running locally. An escape hatch is provided for people who do not care about these tests, but still need to run specs containing them locally. Run this command:

yarn test-cypress-run --env grepTags="-@external" --spec path/to/spec/foo.cy.spec.js

Please note the minus sign before the @external tag. For more details, consult the official documentation.

If you want to or need to run these tests, there is a handy option that does the heavy lifting for you:

yarn test-cypress-open-qa

Running tests with Snowplow involved

Tests that depend on Snowplow expect a running server. To run them, you need to:

• pass env variables to the test run: MB_SNOWPLOW_AVAILABLE=true MB_SNOWPLOW_URL=http://localhost:9090 yarn test-cypress-open

Testing with Snowplow

Our end-to-end testing environment has been configured to run Snowplow Micro alongside the application.

To run Snowplow locally use the following commands:

docker-compose -f ./snowplow/docker-compose.yml up -d
export MB_SNOWPLOW_AVAILABLE=true
export MB_SNOWPLOW_URL=http://localhost:9090

1. You can use describeWithSnowplow (or describeWithSnowplowEE for EE edition) method to define tests that only run when a Snowplow instance is running
2. Use resetSnowplow() test helper before each test to clear the queue of processed events.
3. Use expectGoodSnowplowEvents(count) to assert that events have been sent and processed correctly. Use expectGoodSnowPlowEvent({ ...payload}) to assert on the content of a snowplow event
4. Use expectNoBadSnowplowEvents() after each test to assert that no invalid events have been sent.

Running tests that require SMTP server

Some of our tests, that depend on the email being set up, require a local SMTP server. We use maildev Docker image for that purpose. At the time of this writing the image we use is maildev/maildev:2.1.0. It should be safe to always use the :latest image in your local development. Run this command:

docker run -d -p 1080:1080 -p 1025:1025 maildev/maildev:latest

Cypress comes with Lodash for free

We don’t need to have Lodash in our direct dependencies to be able to use it with Cypress. It is aliased with an underscore and its methods can be accessed with Cypress._.method(). We can use _.times method to stress-test a certain test (or a set of tests) locally.

// Run the test N times
Cypress._.times(N, ()=> {
  it("should foo", ()=> {
    // ...
  });
});

Embedding SDK tests

Tests located in e2e/test/scenarios/embedding-sdk are used to run automated checks for the Embedding SDK.

Embedding SDK is a library, and not an application. We use Storybook to host public components, and we run tests against it.

In order to run stories used for tests locally, please check storybook setup docs

DB Snapshots

At the beginning of each test suite we wipe the backend’s db and settings cache. This ensures that the test suite starts in a predictable state.

Typically, we use the default snapshot by adding before(restore) inside the first describe block to restore before running the whole test suite. If you want to use a snapshot besides the default one, specify the name as an argument to restore like this: before(() => restore("blank")). You can also call restore() inside beforeEach() to reset before every test, or inside specific tests.

Snapshots are created with a separate set of Cypress tests. These tests start with a blank database and execute specific actions to put the database in predictable state. For example: signup as bob@metabase.com, add a question, turn on setting ABC.

These snapshot-generating tests have the extension .cy.snap.js. When these tests run they create db dumps in frontend/tests/snapshots/*.sql. They are run before the tests begin and don’t get committed to git.

Running in CI

Cypress records videos of each test run, which can be helpful in debugging. Additionally, failed tests have higher quality images saved.

These files can be found under the “Artifacts” section for each run’s summary in GitHub Actions. The example of the artifacts for a failed test in “Onboarding” directory:

Running Cypress tests against Metabase® Enterprise Edition™

Prior to running Cypress against Metabase® Enterprise Edition™, set MB_EDITION=ee environment variable. We have a special describe block called describeEE that will conditionally skip or run tests based on the edition.

Enterprise instance will start without a premium token!

If you want to test premium features (feature flags), valid tokens need to be available to all Cypress tests. We achieve this by prefixing environment variables with CYPRESS_. You must provide two tokens that correspond to the EE/PRO self-hosted (all features enabled) and STARTER Cloud (no features enabled) Metabase plans. For more information, please see Metabase pricing page.

• CYPRESS_ALL_FEATURES_TOKEN
• CYPRESS_NO_FEATURES_TOKEN

MB_EDITION=ee CYPRESS_ALL_FEATURES_TOKEN=xxxxxx CYPRESS_NO_FEATURES_TOKEN=xxxxxx yarn test-cypress-open

If you navigate to the /admin/settings/license page, the license input field should display the active token. Be careful when sharing screenshots!

• If tests under describeEE block are greyed out and not running, make sure you spun up Metabase® Enterprise Edition™.
• If tests start running but the enterprise features are missing: make sure that the token you use has corresponding feature flags enabled.
• If everything with the token seems to be okay, go nuclear and destroy all Java processes: run killall java and restart Cypress.

Tags

Cypress allows us to tag tests, to easily find certain categories of tags. For example, we can tag all tests that require an external database with @external and then run only those tests with yarn test-cypress-open --env grepTags="@external". Tags should start with @ just to make it easier to distinguish them from other strings in searches.

These are the tags currently in use:

• @external - tests that require an external docker container to run
• @actions - tests that use metabase actions and mutate data in a data source

How to stress-test a flake fix?

Fixing a flaky test locally doesn’t mean the fix works in GitHub’s CI environment. The only way to be sure the fix works is to stress-test it in CI. That’s what .github/workflows/e2e-stress-test-flake-fix.yml is made for. It allows you to quickly test the fix in your branch without waiting for the full build to complete.

Please follow these steps:

Prepare

• Create a new branch with your proposed fix and push it to the remote
• Either skip opening a PR altogether or open a draft pull request

Trigger the stress-test workflow manually

• Go to https://github.com/metabase/metabase/actions/workflows/e2e-stress-test-flake-fix.yml
• Click on Run workflow trigger next to “This workflow has a workflow_dispatch event trigger.”
  1. Choose your own branch in the first field “Use workflow from” (this part is crucial!)
  2. Copy and paste the relative path of the spec you want to test (e.g. e2e/test/scenarios/onboarding/urls.cy.spec.js) - you don’t have to wrap it in quotes
  3. Set the desired number of times to run the test
  4. Optionally provide a grep filter, according to the documentation
  5. Click the green “Run workflow” button and wait for the results

Things to keep in mind when using this workflow

• It will automatically try to find and download the previously built Metabase uberjar stored as an artifact from one of the past commits / CI runs.
• It was intended to be used for pure E2E fixes that don’t require new Metabase uberjar.
• If the fix required a source-code change (either backend of frontend), please open a regular PR instead and let the CI run all tests first. After this, you can trigger the stress-test workflow manually, as explained above, and it will automatically download newly built artifact from this CI run. Please, keep in mind that CI needs to fully finish running first. The workflow uses GitHub REST API which doesn’t see artifacts otherwise.