Skip to main content

Testing Backend Plugins and Modules

DISCLAIMER: The new backend system is in alpha, and still under active development. While we have reviewed the interfaces carefully, they may still be iterated on before the stable release.

Utilities for testing backend plugins and modules are available in @backstage/backend-test-utils. This section describes those facilities.

Testing Backend Plugins and Modules

To facilitate testing of backend plugins and modules, the @backstage/backend-test-utils package provides a startTestBackend function which starts up an entire backend harness, complete with a number of mock services. You can then provide overrides for services whose behavior you need to adjust for the test run. The function also accepts a number of features (a collective term for backend plugins and modules), that are the subjects of the test.

The function returns an HTTP server instance which can be used together with e.g. supertest to easily test the actual REST service surfaces of plugins who register routes with the HTTP router service API.

import { mockServices, startTestBackend } from '@backstage/backend-test-utils';
import request from 'supertest';
import { myPlugin } from './plugin.ts';

describe('myPlugin', () => {
it('can serve values from config', async () => {
const fakeConfig = { myPlugin: { value: 7 } };

const { server } = await startTestBackend({
features: [myPlugin()],
services: [mockServices.config.factory({ data: fakeConfig })],

const response = await request(server).get('/api/example/get-value');
expect(response.body).toEqual({ value: 7 });

This example shows how to access the mock service factories and pass options to them, which will override the default mock services.

The returned server also has a port() method which returns the dynamically bound listening port. You can use this to perform lower level network interactions with the running test service.

Testing Remote Service Interactions

If your backend plugin or service interacts with external services using HTTP calls, we recommend leveraging the msw package to intercept actual outgoing requests and return mock responses. This lets you stub out remote services rather than the local clients, leading to more thorough and robust tests. You can read more about how it works in their documentation.

The @backstage/backend-test-utils package exports a setupRequestMockHandlers function which ensures that the correct jest lifecycle hooks are invoked to set up and tear down your msw instance, and enables the option that completely rejects requests that don't match one of your mock rules. This ensures that your tests cannot accidentally leak traffic into production from tests.


import { setupRequestMockHandlers } from '@backstage/backend-test-utils';
import { rest } from 'msw';
import { setupServer } from 'msw/node';

describe('read from remote', () => {
const worker = setupServer();

it('should auth and read successfully', async () => {

rest.get('', (req, res, ctx) => {
expect(req.headers.get('authorization')).toBe('Bearer fake');
return res(
ctx.set('Content-Type', 'application/json'),
ctx.body(JSON.stringify({ value: 7 })),

// exercise your plugin or service as usual, with real clients

Testing Database Interactions

The @backstage/backend-test-utils package includes facilities for testing your plugins' interactions with databases, including spinning up testcontainers powered Docker images with real database engines to connect to.

The base setup for such a test could look as follows:

// MyDatabaseClass.test.ts
import { TestDatabaseId, TestDatabases } from '@backstage/backend-test-utils';
import { MyDatabaseClass, type FooTableRow } from './MyDatabaseClass';

describe('MyDatabaseClass', () => {
// Change this to the set of constants that you actually actively intend to
// support. This create call must be made inside a describe block. Make sure
// to create only one TestDatabases instance per file, since spinning up
// "physical" databases to test against is much costlier than creating the
// "logical" databases within them that the individual tests use.
const databases = TestDatabases.create({
ids: ['POSTGRES_13', 'POSTGRES_9', 'SQLITE_3'],

// Just an example of how to conveniently bundle up the setup code
async function createSubject(databaseId: TestDatabaseId) {
const knex = await databases.init(databaseId);
const subject = new MyDatabaseClass({ database: knex });
await subject.runMigrations();
return { knex, subject };

describe('foo', () => {
// Easily run the exact same test onto all supported databases
'should run foo on %p',
async databaseId => {
const { knex, subject } = await createSubject(databaseId);
// raw knex is available for underlying manipulation
await knex<FooTableRow>('foo').insert({ value: 2 });
// drive your system under test as usual
await expect(subject.foos()).resolves.toEqual([{ value: 2 }]);

If you want to pass the test database instance into backend plugins or services, you can supply it in the form of a mock instance of coreServices.database to your test database.

const { knex, subject } = await createSubject(databaseId);
const { server } = await startTestBackend({
features: [myPlugin()],
services: [[coreServices.database, { getClient: async () => knex }]],

When running locally, the tests only run against SQLite for the sake of speed. When the CI environment variable is set, all given database engines are used.

If you do not want or are unable to use docker based database engines, e.g. if your CI environment is able to supply databases natively, the TestDatabases support custom connection strings through the use of environment variables that it'll take into account when present.