Page Contents

Overview

A thorough automated test suite is important because it:

  • Ensures your application works as expected.
  • Prevents regressions when new features are added and bugs are fixed.
  • Helps new and existing developers understand different parts of the codebase (knowledge sharing).
  • Speeds up development over the long run (the code writes itself!)

Types of tests

We encourage writing tests from a few perspectives, mainly black-box testing (acceptance) and white-box testing (integration and unit). Tests are usually written using typical patterns such as act/arrange/assert or given/when/then. While both styles work well, just pick one that you’re comfortable with and start writing tests!

For an introduction to automated testing, see Define your testing strategy; for a step-by-step tutorial, see Incrementally implement features.

This article is a reference guide for common types of tests and test helpers.

Project setup

An automated test suite requires a test runner to execute all the tests and produce a summary report. We use and recommend Mocha.

In addition to a test runner, the test suites generally requires:

  • An assertion library (we recommend Should.js).
  • A Library for making HTTP calls and verifying their results (we recommend supertest).
  • A library for creating test doubles (we recommend Sinon.JS).

The @loopback/testlab module integrates these packages and makes them easy to use together with LoopBack.

Set up testing infrastructure with LoopBack CLI

Setup testing infrastructure manually

If you have an existing application install mocha and @loopback/testlab:

npm install --save-dev mocha @loopback/testlab

Your package.json should then look something like this:

{
  // ...
  "devDependencies": {
    "@loopback/testlab": "^<current-version>",
    "mocha": "^<current-version>"
  },
  "scripts": {
    "test": "mocha"
  }
  // ...
}

Data handling

Tests accessing a real database often require existing data. For example, a method listing all products needs some products in the database; a method to create a new product instance must determine which properties are required and any restrictions on their values. There are various approaches to address this issue. Many of them unfortunately make the test suite difficult to understand, difficult to maintain, and prone to test failures unrelated to the changes made.

Based on our experience, we recommend the following approach.

Clean the database before each test

Always start with a clean database before each test. This may seem counter-intuitive: why not reset the database after the test has finished? When a test fails and the database is cleaned after the test has finished, then it’s difficult to observe what was stored in the database and why the test failed. When the database is cleaned in the beginning, then any failing test will leave the database in the state that caused the test to fail.

To clean the database before each test, set up a beforeEach hook to call a helper method; for example:

test/helpers/database.helpers.ts

export async function givenEmptyDatabase() {
  await new ProductRepository().deleteAll();
  await new CategoryRepository().deleteAll();
}

// in your test file
describe('ProductController (integration)', () => {
  before(givenEmptyDatabase);
  // etc.
});

Use test data builders

To avoid duplicating code for creating model data with all required properties filled in, use shared test data builders instead. This enables tests to provide a small subset of properties that are strictly required by the tested scenario, which is important because it makes tests:

  • Easier to understand, since it’s immediately clear what model properties are relevant to the test. If the test were setting all required properties, it would be difficult to tell whether some of those properties are actually relevant to the tested scenario.

  • Easier to maintain. As your data model evolves, you eventually need to add more required properties. If the tests were building model instance data manually, you would have to fix all tests to set the new required property. With a shared helper, there is only a single place where to add a value for the new required property.

See @loopback/openapi-spec-builder for an example of how to apply this design pattern for building OpenAPI Spec documents.

In practice, a rich method-based API is overkill and a simple function that adds missing required properties is sufficient.

export function givenProductData(data: Partial<Product>) {
  return Object.assign({
    name: 'a-product-name',
    slug: 'a-product-slug',
    price: 1,
    description: 'a-product-description',
    available: true,
  }, data);
}

export async function givenProduct(data: Partial<Product>) {
  return await new ProductRepository().create(
    givenProductData(data));
}

Avoid sharing the same data for multiple tests

It’s tempting to define a small set of data that’s shared by all tests. For example, in an e-commerce application, you might pre-populate the database with few categories, some products, an admin user and a customer. Such approach has several downsides:

  • When trying to understand any individual test, it’s difficult to tell what part of the pre-populated data is essential for the test and what’s irrelevant. For example, in a test checking the method counting the number of products in a category using a pre-populated category “Stationery”, is it important that “Stationery” contains nested sub-categories or is that fact irrelevant? If it’s irrelevant, then what are the other tests that depend on it?

  • As the application grows and new features are added, it’s easier to add more properties to existing model instances rather than create new instances using only properties required by the new features. For example, when adding a category image, it’s easier to add image to an existing category “Stationery” and perhaps keep another category “Groceries” without any image, rather than create two new categories “CategoryWithAnImage” and “CategoryMissingImage”. This further amplifies the previous problem, because it’s not clear that “Groceries” is the category that should be used by tests requiring a category with no image - the category name does not provide any hints on that.

  • As the shared dataset grows (together with the application), the time required to bring the database into initial state grows too. Instead of running a few “DELETE ALL” queries before each test (which is relatively fast), you can end up with running tens to hundreds different commands creating different model instances, triggering slow index rebuilds along the way, and considerably slowing the test suite.

Use the test data builders described in the previous section to populate your database with the data specific to your test only.

Using the e-commerce example described above, this is how integration tests for the CategoryRepository might look:

describe('Category (integration)', () => {
  beforeEach(givenEmptyDatabase);

  describe('countProducts()', () => {
    it('returns correct count for an empty', async () => {
      const category = await givenCategory();
      const count = await category.countProducts();
      expect(count).to.equal(0);
    });

    // etc.

    it('includes products in subcategories', async () => {
      const category = await givenCategory({
        products: [await givenProduct()],
        subcategories: [
          givenCategory({
            products: [await givenProduct()]
          })
        ],
      });

      const count = await category.countProducts();
      expect(count).to.equal(2);
    });
  });
});

Write higher-level helpers to share the code for re-creating common scenarios. For example, if your application has two kinds of users (admins and customers), then you may write the following helpers to simplify writing acceptance tests checking access control:

async function givenAdminAndCustomer() {
  return {
    admin: await givenUser({role: Roles.ADMIN}),
    customer: await givenUser({role: Roles.CUSTOMER}),
  };
}

Unit testing

Unit tests are considered “white-box” tests because they use an “inside-out” approach where the tests know about the internals and controls all the variables of the system being tested. Individual units are tested in isolation, their dependencies are replaced with Test doubles.

Use test doubles

Test doubles are functions or objects that look and behave like the real variants used in production, but are actually simplified versions giving the test more control of the behavior. For example, reproducing the situation where reading from a file failed because of a hard-drive error is pretty much impossible, unless we are using a test double that’s simulating file-system API and giving us control of how what each call returns.

Sinon.JS has become the de-facto standard for test doubles in Node.js and JavaScript/TypeScript in general. The @loopback/testlab package comes with Sinon preconfigured with TypeScript type definitions and integrated with Should.js assertions.

There are three kinds of test doubles provided by Sinon.JS:

  • Test spies are functions that record arguments, the return value, the value of this, and exceptions thrown (if any) for all its calls. There are two types of spies: Some are anonymous functions, while others wrap methods that already exist in the system under test.

  • Test stubs are functions (spies) with pre-programmed behavior. As spies, stubs can be either anonymous, or wrap existing functions. When wrapping an existing function with a stub, the original function is not called.

  • Test mocks (and mock expectations) are fake methods (like spies) with pre-programmed behavior (like stubs) as well as pre-programmed expectations. A mock will fail your test if it is not used as expected.

Create a stub Repository

When writing an application accessing data in a database, best practice is to use repositories to encapsulate all data-access/persistence-related code and let other parts of the application (typically controllers) to depend on these repositories for data access. To test Repository dependents (for example, Controllers) in isolation, we need to provide a test double, usually as a test stub.

In traditional object-oriented languages like Java or C#, to enable unit tests to provide a custom implementation of the repository API, the controller needs to depend on an interface describing the API, and the repository implementation needs to implement this interface. The situation is easier in JavaScript and TypeScript. Thanks to the dynamic nature of the language, it’s possible to mock/stub entire classes.

Creating a test double for a repository class is very easy using the Sinon.JS utility function createStubInstance. It’s important to create a new stub instance for each unit test in order to prevent unintended re-use of pre-programmed behavior between (unrelated) tests.

describe('ProductController', () => {
  let repository: ProductRepository;
  beforeEach(givenStubbedRepository);

  // your unit tests

  function givenStubbedRepository() {
    repository = sinon.createStubInstance(ProductRepository);
  }
});

In your unit tests, you will usually want to program the behavior of stubbed methods (what should they return) and then verify that the Controller (unit under test) called the right method with the correct arguments.

Configure stub’s behavior at the beginning of your unit test (in the “arrange” or “given” section):

// repository.find() will return a promise that
// will be resolved with the provided array
const findStub = repository.find as sinon.SinonStub;
findStub.resolves([{id: 1, name: 'Pen'}]);

Verify how was the stubbed method executed at the end of your unit test (in the “assert” or “then” section):

// expect that repository.find() was called with the first
// argument deeply-equal to the provided object
expect(findStub).to.be.calledWithMatch({where: {id: 1}});

See Unit test your controllers for a full example.

Create a stub Service

To be done. The initial beta release does not include Services as a first-class feature.

Unit test your Controllers

Unit tests should apply to the smallest piece of code possible to ensure other variables and state changes do not pollute the result. A typical unit test creates a controller instance with dependencies replaced by test doubles and directly calls the tested method. The example below gives the controller a stub implementation of its repository dependency, and then ensure the controller called repository’s find() method with a correct query and returned back the query results. See Create a stub repository for a detailed explanation.

test/controllers/product.controller.unit.ts

import {ProductController, ProductRepository} from '../..';
import {expect, sinon} from '@loopback/testlab';

describe('ProductController (unit)', () => {
  let repository: ProductRepository;
  beforeEach(givenStubbedRepository);

  describe('getDetails()', () => {
    it('retrieves details of a product', async () => {
      const controller = new ProductController(repository);
      const findStub = repository.find as sinon.SinonStub;
      findStub.resolves([{id: 1, name: 'Pen'}]);

      const details = await controller.getDetails(1);

      expect(details).to.containDeep({name: 'Pen'});
      expect(findStub).to.be.calledWithMatch({where: {id: 1}});
    });
  });

  function givenStubbedRepository() {
    repository = sinon.createStubInstance(ProductRepository);
  }
});

Unit test your models and repositories

In a typical LoopBack application, models and repositories rely on behavior provided by the framework (@loopback/repository package) and there is no need to test LoopBack’s built-in functionality. However, any additional application-specific API does need new unit tests.

For example, if the Person Model has properties firstname, middlename and surname and provides a function to obtain the full name, then you should write unit tests to verify the implementation of this additional method.

Remember to use Test data builders whenever you need valid data to create a new model instance.

test/unit/models/person.model.unit.ts

import {Person} from '../../models/person.model'
import {givenPersonData} from '../helpers/database.helpers'
import {expect} from '@loopback/testlab';

describe('Person (unit)', () => {
  // we recommend to group tests by method names
  describe('getFullName()', () => {
    it('uses all three parts when present', () => {
      const person = givenPerson({
        firstname: 'Jane',
        middlename: 'Smith',
        surname: 'Brown'
      }));

      const fullName = person.getFullName();
      expect(fullName).to.equal('Jane Smith Brown');
    });

    it('omits middlename when not present', () => {
      const person = givenPerson({
        firstname: 'Mark',
        surname: 'Twain'
      }));

      const fullName = person.getFullName();
      expect(fullName).to.equal('Mark Twain');
    });
  });

  function givenPerson(data: Partial<Person>) {
    return new Person(givenPersonData(data));
  }
});

Writing a unit test for a custom repository methods is not straightforward because CrudRepositoryImpl is based on legacy loopback-datasource-juggler that was not designed with dependency injection in mind. Instead, use integration tests to verify the implementation of custom repository methods; see Test your repositories against a real database in Integration Testing.

Unit test your Sequence

While it’s possible to test a custom Sequence class in isolation, it’s better to rely on acceptance-level tests in this exceptional case. The reason is that a custom Sequence class typically has many dependencies (which makes test setup too long and complex), and at the same time it provides very little functionality on top of the injected sequence actions. Bugs are much more likely to caused by the way how the real sequence action implementations interact together (which is not covered by unit tests), instead of the Sequence code itself (which is the only thing covered).

See Test Sequence customizations in Acceptance Testing.

Unit test your Services

To be done. The initial beta release does not include Services as a first-class feature.

See the following related GitHub issues:

  • Define services to represent interactions with REST APIs, SOAP Web Services, gRPC services, and more: #522
  • Guide: Services #451

Integration testing

Integration tests are considered “white-box” tests because they use an “inside-out” approach that tests how multiple units work together or with external services. You can use test doubles to isolate tested units from external variables/state that are not part of the tested scenario.

Test your repositories against a real database

There are two common reasons for adding repository tests:

  • Your models are using advanced configuration, for example, custom column mappings, and you want to verify this configuration is correctly picked up by the framework.
  • Your repositories have additional methods.

Integration tests are one of the places to put the best practices in Data handling to work:

  • Clean the database before each test
  • Use test data builders
  • Avoid sharing the same data for multiple tests

Here is an example showing how to write an integration test for a custom repository method findByName:

tests/integration/repositories/category.repository.integration.ts

import {givenEmptyDatabase} from '../../helpers/database.helpers.ts';

describe('CategoryRepository (integration)', () => {
  beforeEach(givenEmptyDatabase);

  describe('findByName(name)', () => {
    it('return the correct category', async () => {
      const stationery = await givenCategory({name: 'Stationery'});
      const groceries = await givenCategory({name: 'Groceries'});
      const repository = new CategoryRepository();

      const found = await repository.findByName('Stationery');

      expect(found).to.deepEqual(stationery);
    });
  });
});

Test Controllers and repositories together

Integration tests running controllers with real repositories are important to verify that the controllers use the repository API correctly, and the commands and queries produce expected results when executed on a real database. These tests are similar to repository tests: we are just adding controllers as another ingredient.

import {ProductController, ProductRepository, Product} from '../..';
import {expect} from '@loopback/testlab';
import {givenEmptyDatabase, givenProduct} from '../helpers/database.helpers';

describe('ProductController (integration)', () => {
  beforeEach(givenEmptyDatabase);

  describe('getDetails()', () => {
    it('retrieves details of the given product', async () => {
      const inkPen = await givenProduct({name: 'Pen', slug: 'pen'});
      const pencil = await givenProduct({name: 'Pencil', slug: 'pencil'});
      const controller = new ProductController(new ProductRepository());

      const details = await controller.getDetails('pen');

      expect(details).to.eql(pencil);
    });
  });
});

Test your Services against real backends

To be done. The initial beta release does not include Services as a first-class feature.

Acceptance (end-to-end) testing

Automated acceptance (end-to-end) tests are considered “black-box” tests because they use an “outside-in” approach that is not concerned about the internals of the system, just simply do the same actions (send the same HTTP requests) as the clients and consumers of your API will do, and verify the results returned by the system under test are matching the expectations.

Typically, acceptance tests start the application, make HTTP requests to the server, and verify the returned response. LoopBack uses supertest to make the test code that executes HTTP requests and verifies responses easier to write and read. Remember to follow the best practices from Data handling when setting up your database for tests:

  • Clean the database before each test
  • Use test data builders
  • Avoid sharing the same data for multiple tests

Validate your OpenAPI specification

The OpenAPI specification is a cornerstone of applications that provide REST APIs. It enables API consumers to leverage a whole ecosystem of related tooling. To make the spec useful, you must ensure it’s a valid OpenAPI Spec document, ideally in an automated way that’s an integral part of regular CI builds. LoopBack’s testlab module provides a helper method validateApiSpec that builds on top of the popular swagger-parser package.

Example usage:

// test/acceptance/api-spec.acceptance.ts
import {validateApiSpec} from '@loopback/testlab';
import {HelloWorldApp} from '../..';
import {RestServer} from '@loopback/rest';

describe('API specification', () => {
  it('api spec is valid', async () => {
    const app = new HelloWorldApp();
    const server = await app.getServer(RestServer);
    const spec = server.getApiSpec();
    await validateApiSpec(apiSpec);
  });
});

Perform an auto-generated smoke test of your REST API

The formal validity of your application’s spec does not guarantee that your implementation is actually matching the specified behavior. To keep your spec in sync with your implementation, you should use an automated tool like Dredd to run a set of smoke tests to verify conformance of your app with the spec.

Automated testing tools usually require little hints in your specification to tell them how to create valid requests or what response data to expect. Dredd in particular relies on response examples and request parameter x-example fields. Extending your API spec with examples is good thing on its own, since developers consuming your API will find them useful too.

Here is an example showing how to run Dredd to test your API against the spec:

describe('API (acceptance)', () => {
  let dredd: any;
  before(initEnvironment);

  it('conforms to the specification', done => {
    dredd.run((err: Error, stats: object) => {
      if (err) return done(err);
      expect(stats).to.containDeep({
        failures: 0,
        errors: 0,
        skipped: 0,
      });
      done();
    });
  });

  async function initEnvironment() {
    const app = new HelloWorldApp();
    const server = app.getServer(RestServer);
    // For testing, we'll let the OS pick an available port by setting
    // RestBindings.PORT to 0.
    server.bind(RestBindings.PORT).to(0);
    // app.start() starts up the HTTP server and binds the acquired port
    // number to RestBindings.PORT.
    await app.start();
    // Get the real port number.
    const port = await server.get(RestBindings.PORT);
    const baseUrl = `http://localhost:${port}`;
    const config: object = {
      server: baseUrl, // base path to the end points
      options: {
        level: 'fail', // report 'fail' case only
        silent: false, // false for helpful debugging info
        path: [`${baseUrl}/swagger.json`], // to download apiSpec from the service
      }
    };
    dredd = new Dredd(config);
  });
})

The user experience is not as great as we would like it, we are looking into better solutions; see GitHub issue #644. Let us know if you can recommend one!

Test your individual REST API endpoints

You should have at least one acceptance (end-to-end) test for each of your REST API endpoints. Consider adding more tests if your endpoint depends on (custom) sequence actions to modify the behavior when the corresponding controller method is invoked via REST, compared to behavior observed when the controller method is invoked directly via JavaScript/TypeScript API. For example, if your endpoint returns different response to regular users and to admin users, then you should have two tests: one test for each user role.

Here is an example of an acceptance test:

// test/acceptance/product.acceptance.ts
import {HelloWorldApp} from '../..';
import {RestBindings, RestServer} from '@loopback/rest';
import {expect, supertest} from '@loopback/testlab';
import {givenEmptyDatabase, givenProduct} from '../helpers/database.helpers';

describe('Product (acceptance)', () => {
  let app: HelloWorldApp;
  let request: supertest.SuperTest<supertest.Test>;

  before(givenEmptyDatabase);
  before(givenRunningApp);

  it('retrieves product details', async () => {
    // arrange
    const product = await givenProduct({
      name: 'Ink Pen',
      slug: 'ink-pen',
      price: 1,
      category: 'Stationery',
      description: 'The ultimate ink-powered pen for daily writing',
      label: 'popular',
      available: true,
      endDate: null,
    });

    // act
    const response = await request.get('/product/ink-pen')

    // assert
    expect(response.body).to.deepEqual({
      id: product.id,
      name: 'Ink Pen',
      slug: 'ink-pen',
      price: 1,
      category: 'Stationery',
      available: true,
      description: 'The ultimate ink-powered pen for daily writing',
      label: 'popular',
      endDate: null,
    });
  });

  async function givenRunningApp() {
    app = new HelloWorldApp();
    const server = await app.getServer(RestServer);
    server.bind(RestBindings.PORT).to(0);
    await app.start();

    const port: number = await server.get(RestBindings.PORT);
    request = supertest(`http://127.0.0.1:${port}`);
  }
});

Test Sequence customizations

Custom sequence behavior is best tested by observing changes in behavior of affected endpoints. For example, if your sequence has an authentication step that rejects anonymous requests for certain endpoints, then you can write a test making an anonymous request to such an endpoint to verify that it’s correctly rejected. These tests are essentially the same as the tests verifying implementation of individual endpoints as described in the previous section.