Testing Kibana Plugins

This document outlines best practices and patterns for testing Kibana Plugins.

Strategy

In general, we recommend three tiers of tests:

Unit tests: small, fast, exhaustive, make heavy use of mocks for external dependencies
Integration tests: higher-level tests that verify interactions between systems (eg. HTTP APIs, Elasticsearch API calls, calling other plugin contracts).
End-to-end tests (e2e): tests that verify user-facing behavior through the browser

These tiers should roughly follow the traditional "testing pyramid", where there are more exhaustive testing at the unit level, fewer at the integration level, and very few at the functional level.

Core Integrations

Core Mocks

When testing a plugin's integration points with Core APIs, it is heavily recommended to utilize the mocks provided in src/core/server/mocks and src/core/public/mocks. The majority of these mocks are dumb jest mocks that mimic the interface of their respective Core APIs, however they do not return realistic return values.

If the unit under test expects a particular response from a Core API, the test will need to set this return value explicitly. The return values are type checked to match the Core API where possible to ensure that mocks are updated when Core APIs changed.

Example

typescript

import { elasticsearchServiceMock } from 'src/core/server/mocks';

test('my test', async () => {
  // Setup mock and faked response
  const esClient = elasticsearchServiceMock.createScopedClusterClient();
  esClient.callAsCurrentUser.mockResolvedValue(/** insert ES response here */);

  // Call unit under test with mocked client
  const result = await myFunction(esClient);

  // Assert that client was called with expected arguments
  expect(esClient.callAsCurrentUser).toHaveBeenCalledWith(/** expected args */);
  // Expect that unit under test returns expected value based on client's response
  expect(result).toEqual(/** expected return value */)
});

Strategies for specific Core APIs

HTTP Routes

The HTTP API interface is another public contract of Kibana, although not every Kibana endpoint is for external use. When evaluating the required level of test coverage for an HTTP resource, make your judgment based on whether an endpoint is considered to be public or private. Public API is expected to have a higher level of test coverage. Public API tests should cover the observable behavior of the system, therefore they should be close to the real user interactions as much as possible, ideally by using HTTP requests to communicate with the Kibana server as a real user would do.

Preconditions

We are going to add tests for myPlugin plugin that allows to format user-provided text, store and retrieve it later. The plugin has thin route controllers isolating all the network layer dependencies and delegating all the logic to the plugin model.

typescript

class TextFormatter {
  public static async format(text: string, sanitizer: Deps['sanitizer']) {
    // sanitizer.sanitize throws MisformedTextError when passed text contains HTML markup
    const sanitizedText = await sanitizer.sanitize(text);
    return sanitizedText;
  }

  public static async save(text: string, savedObjectsClient: SavedObjectsClient) {
    const { id } = await savedObjectsClient.update('myPlugin-type', 'myPlugin', {
      userText: text
    });
    return { id };
  }

  public static async getById(id: string, savedObjectsClient: SavedObjectsClient) {
    const { attributes } = await savedObjectsClient.get('myPlugin-type', id);
    return { text: attributes.userText };
  }
}
router.get(
  {
    path: '/myPlugin/formatter',
    validate: {
      query: schema.object({
        text: schema.string({ maxLength: 100 }),
      }),
    },
  },
  async (context, request, response) => {
    try {
      const formattedText = await TextFormatter.format(request.query.text, deps.sanitizer);
      return response.ok({ body: formattedText });
    } catch(error) {
      if (error instanceof MisformedTextError) {
        return response.badRequest({ body: error.message })
      }

      throw e;
    }
  }
);
router.post(
  {
    path: '/myPlugin/formatter/text',
    validate: {
      body: schema.object({
        text: schema.string({ maxLength: 100 }),
      }),
    },
  },
  async (context, request, response) => {
    try {
      const { id } = await TextFormatter.save(request.query.text, context.core.savedObjects.client);
      return response.ok({ body: { id } });
    } catch(error) {
      if (SavedObjectsErrorHelpers.isConflictError(error)) {
        return response.conflict({ body: error.message })
      }
      throw e;
    }
  }
);

router.get(
  {
    path: '/myPlugin/formatter/text/{id}',
    validate: {
      params: schema.object({
        id: schema.string(),
      }),
    },
  },
  async (context, request, response) => {
    try {
      const { text } = await TextFormatter.getById(request.params.id, context.core.savedObjects.client);
      return response.ok({
        body: text
      });
    } catch(error) {
      if (SavedObjectsErrorHelpers.isNotFoundError(error)) {
        return response.notFound()
      }
      throw e;
    }
  }
);

Unit testing

Unit tests provide the simplest and fastest way to test the logic in your route controllers and plugin models. Use them whenever adding an integration test is hard and slow due to complex setup or the number of logic permutations. Since all external core and plugin dependencies are mocked, you don't have the guarantee that the whole system works as expected.

Pros:

fast
easier to debug

Cons:

doesn't test against real dependencies
doesn't cover integration with other plugins

Example

You can leverage existing unit-test infrastructure for this. You should add *.test.ts file and use dependencies mocks to cover the functionality with a broader test suit that covers:

input permutations
input edge cases
expected exception
interaction with dependencies

typescript

// src/platform/plugins/shared/my_plugin/server/formatter.test.ts
describe('TextFormatter', () => {
  describe('format()', () => {
    const sanitizer = sanitizerMock.createSetup();
    sanitizer.sanitize.mockImplementation((input: string) => `sanitizer result:${input}`);

    it('formats text to a ... format', async () => {
      expect(await TextFormatter.format('aaa', sanitizer)).toBe('...');
    });

    it('calls Sanitizer.sanitize with correct arguments', async () => {
      await TextFormatter.format('aaa', sanitizer);
      expect(sanitizer.sanitize).toHaveBeenCalledTimes(1);
      expect(sanitizer.sanitize).toHaveBeenCalledWith('aaa');
    });

    it('throws MisformedTextError if passed string contains banned symbols', async () => {
      sanitizer.sanitize.mockRejectedValueOnce(new MisformedTextError());
      await expect(TextFormatter.format('any', sanitizer)).rejects.toThrow(MisformedTextError);
    });
    // ... other tests
  });
});

Integration tests

Depending on the number of external dependencies, you can consider implementing several high-level integration tests. They would work as a set of smoke tests for the most important functionality.

Main subjects for tests should be:

authenticated / unauthenticated access to an endpoint.
endpoint validation (params, query, body).
main business logic.
dependencies on other plugins.

Functional Test Runner

If your plugin relies on the elasticsearch server to store data and supports additional configuration, you can leverage the Functional Test Runner(FTR) to implement integration tests. FTR bootstraps an elasticsearch and a Kibana instance and runs the test suite against it.

FTR development workflow:

bash

# Start servers once, keep them running
yarn test:ftr:server --config path/to/config.ts

# In another terminal, run tests against the running servers
yarn test:ftr:runner --config path/to/config.ts

Useful FTR flags:

bash

# Run a specific test by name
yarn test:ftr:runner --config path/to/config.ts --grep "test name"

# Debug with browser open, stop on first failure
yarn test:ftr --config path/to/config.ts --debug --bail

# Run against serverless ES
yarn test:ftr --config path/to/config.ts --esFrom serverless

Pros:

runs the whole Elastic stack
tests cross-plugin integration
emulates a real user interaction with the stack
allows adjusting config values

Cons:

slow start
hard to debug
brittle tests

Example

You can reuse existing api_integration setup by registering a test file within a test loader. More about the existing FTR setup in the contribution guide

The tests cover:

authenticated / non-authenticated user access (when applicable)

typescript

// src/platform/test/api_integration/apis/my_plugin/something.ts
export default function({ getService }: FtrProviderContext) {
  const supertest = getService('supertest');
  const security = getService('security');

  describe('myPlugin', () => {
    it('returns limited info when not authenticated', async () => {
      await security.logout();
      const response = await supertest
        .get('/myPlugin/health')
        .set('content-type', 'application/json')
        .expect(200);

      expect(response.body).to.have.property('basicInfo');
      expect(response.body).not.to.have.property('detailedInfo');
    });

    it('returns detailed info when authenticated', async () => {
      await security.loginAsSuperUser();
      const response = await supertest
        .get('/myPlugin/health')
        .set('content-type', 'application/json')
        .expect(200);

      expect(response.body).to.have.property('basicInfo');
      expect(response.body).to.have.property('detailedInfo');
    });
  });

request validation

typescript

// src/platform/test/api_integration/apis/my_plugin/something.ts
export default function({ getService }: FtrProviderContext) {
  const supertest = getService('supertest');

  describe('myPlugin', () => {
    it('validate params before to store text', async () => {
      const response = await supertest
        .post('/myPlugin/formatter/text')
        .set('content-type', 'application/json')
        .send({ text: 'aaa'.repeat(100) })
        .expect(400);

      expect(response.body).to.have.property('message');
      expect(response.body.message).to.contain('must have a maximum length of [100]');
    });
  });

the main logic of the plugin

typescript

export default function({ getService }: FtrProviderContext) {
  const supertest = getService('supertest');
  describe('myPlugin', () => {
    it('stores text', async () => {
      const response = await supertest
        .post('/myPlugin/formatter/text')
        .set('content-type', 'application/json')
        .send({ text: 'aaa' })
        .expect(200);

      expect(response.body).to.have.property('id');
      expect(response.body.id).to.be.a('string');
    });

    it('retrieves text', async () => {
      const { body } = await supertest
        .post('/myPlugin/formatter/text')
        .set('content-type', 'application/json')
        .send({ text: 'bbb' })
        .expect(200);

      const response = await supertest.get(`/myPlugin/formatter/text/${body.id}`).expect(200);
      expect(response.text).be('bbb');
    });

    it('returns NotFound error when cannot find a text', async () => {
      await supertest
        .get('/myPlugin/something/missing')
        .expect(404, 'Saved object [myPlugin-type/missing] not found');
    });
  });

TestUtils

It can be utilized if your plugin doesn't interact with the elasticsearch server or mocks the own methods doing so. Runs tests against real Kibana server instance.

Pros:

runs the real Kibana instance
tests cross-plugin integration
emulates a real user interaction with the HTTP resources

Cons:

faster than FTR because it doesn't run elasticsearch instance, but still slow
hard to debug
doesn't cover Kibana CLI logic

Example

To have access to Kibana TestUtils, you should create integration_tests folder and import test_utils within a test file:

typescript

// src/platform/plugins/my_plugin/server/integration_tests/formatter.test.ts
import { createRoot, request } from '@kbn/core-test-helpers-kbn-server';

describe('myPlugin', () => {
  describe('GET /myPlugin/formatter', () => {
    let root: ReturnType<typeof createRoot>;
    beforeAll(async () => {
      root = createRoot();
      await root.preboot();
      await root.setup();
      await root.start();
    }, 30000);

    afterAll(async () => await root.shutdown());
    it('validates given text', async () => {
      const response = await request
        .get(root, '/myPlugin/formatter')
        .query({ text: 'input string'.repeat(100) })
        .expect(400);

      expect(response.body).toHaveProperty('message');
    });

    it('formats given text', async () => {
      const response = await request
        .get(root, '/myPlugin/formatter')
        .query({ text: 'input string' })
        .expect(200);

      expect(response.text).toBe('...');
    });

    it('returns BadRequest if passed string contains banned symbols', async () => {
      await request
        .get(root, '/myPlugin/formatter')
        .query({ text: '<script>' })
        .expect(400, 'Text cannot contain unescaped HTML markup.');
    });
  });
});

Sometimes we want to test a route controller logic and don't rely on the internal logic of the platform or a third-party plugin. Then we can apply a hybrid approach and mock the necessary method of TextFormatter model to test how MisformedTextError handled in the route handler without calling sanitizer dependency directly.

typescript

jest.mock('../path/to/model');
import { TextFormatter } from '../path/to/model';
import { MisformedTextError } from '../path/to/sanitizer'

describe('myPlugin', () => {
  describe('GET /myPlugin/formatter', () => {
    let root: ReturnType<typeof createRoot>;
    beforeAll(async () => {
      root = createRoot();
      await root.preboot();
      await root.setup();
      await root.start();
    }, 30000);

    afterAll(async () => await root.shutdown());
    it('returns BadRequest if Sanitizer throws MisformedTextError', async () => {
      TextFormatter.format.mockRejectedValueOnce(new MisformedTextError());

      await request
        .get(root, '/myPlugin/formatter')
        .query({ text: 'any text' })
        .expect(400, 'bad bad request');
    });
  });
});

Applications

Kibana Platform applications have less control over the page than legacy applications did. It is important that your app is built to handle it's co-habitance with other plugins in the browser. Applications are mounted and unmounted from the DOM as the user navigates between them, without full-page refreshes, as a single-page application (SPA).

These long-lived sessions make cleanup more important than before. It's entirely possible a user has a single browsing session open for weeks at a time, without ever doing a full-page refresh. Common things that need to be cleaned up (and tested!) when your application is unmounted:

Subscriptions and polling (eg. uiSettings.get$())
Any Core API calls that set state (eg. core.chrome.setIsVisible).
Open connections (eg. a Websocket)

While applications do get an opportunity to unmount and run cleanup logic, it is also important that you do not depend on this logic to run. The browser tab may get closed without running cleanup logic, so it is not guaranteed to be run. For instance, you should not depend on unmounting logic to run in order to save state to localStorage or to the backend.

Example

By following the renderApp convention, you can greatly reduce the amount of logic in your application's mount function. This makes testing your application's actual rendering logic easier.

typescript

/** public/plugin.ts */
class Plugin {
  setup(core) {
    core.application.register({
      // id, title, etc.
      async mount(params) {
        const [{ renderApp }, [coreStart, startDeps]] = await Promise.all([
          import('./application'),
          core.getStartServices()
        ]);

        return renderApp(params, coreStart, startDeps);
      }
    })
  }
}

We could still write tests for this logic, but you may find that you're just asserting the same things that would be covered by type-checks.

<details> <summary>See example</summary>

typescript

/** public/plugin.test.ts */
jest.mock('./application', () => ({ renderApp: jest.fn() }));
import { coreMock } from '@kbn/core/public/mocks';
import { renderApp: renderAppMock } from './application';
import { Plugin } from './plugin';

describe('Plugin', () => {
  it('registers an app', () => {
    const coreSetup = coreMock.createSetup();
    new Plugin(coreMock.createPluginInitializerContext()).setup(coreSetup);
    expect(coreSetup.application.register).toHaveBeenCalledWith({
      id: 'myApp',
      mount: expect.any(Function)
    });
  });

  // Test the glue code from Plugin -> renderApp
  it('application.mount wires up dependencies to renderApp', async () => {
    const coreSetup = coreMock.createSetup();
    const [coreStartMock, startDepsMock] = await coreSetup.getStartServices();
    const unmountMock = jest.fn();
    renderAppMock.mockReturnValue(unmountMock);
    const params = coreMock.createAppMountParameters('/fake/base/path');

    new Plugin(coreMock.createPluginInitializerContext()).setup(coreSetup);
    // Grab registered mount function
    const mount = coreSetup.application.register.mock.calls[0][0].mount;

    const unmount = await mount(params);
    expect(renderAppMock).toHaveBeenCalledWith(params, coreStartMock, startDepsMock);
    expect(unmount).toBe(unmountMock);
  });
});

</details>

The more interesting logic is in renderApp:

typescript

/** public/application.ts */
import React from 'react';
import ReactDOM from 'react-dom';
import { switchMap } from 'rxjs';

import { AppMountParameters, CoreStart } from '@kbn/core/public';
import { AppRoot } from './components/app_root';

export const renderApp = (
  { element, history }: AppMountParameters,
  core: CoreStart,
  plugins: MyPluginDepsStart
) => {
  // Hide the chrome while this app is mounted for a full screen experience
  core.chrome.setIsVisible(false);

  // uiSettings subscription
  const uiSettingsClient = core.uiSettings.client;
  const pollingSubscription = uiSettingClient.get$('mysetting1').pipe(switchMap(async (mySetting1) => {
    const value = core.http.fetch(/** use `mySetting1` in request **/);
    // ...
  })).subscribe();

  // Render app
  ReactDOM.render(
    <AppRoot routerHistory={history} core={core} plugins={plugins} />,
    element
  );

  return () => {
    // Unmount UI
    ReactDOM.unmountComponentAtNode(element);
    // Close any subscriptions
    pollingSubscription.unsubscribe();
    // Make chrome visible again
    core.chrome.setIsVisible(true);
  };
};

In testing renderApp you should be verifying that:

Your application mounts and unmounts correctly
Cleanup logic is completed as expected

typescript

/** public/application.test.ts */
import { createMemoryHistory } from 'history';
import { ScopedHistory } from '@kbn/core/public';
import { coreMock } from '@kbn/core/public/mocks';
import { renderApp } from './application';

describe('renderApp', () => {
  it('mounts and unmounts UI', () => {
    const params = coreMock.createAppMountParameters('/fake/base/path');
    const core = coreMock.createStart();

    // Verify some expected DOM element is rendered into the element
    const unmount = renderApp(params, core, {});
    expect(params.element.querySelector('.some-app-class')).not.toBeUndefined();
    // Verify the element is empty after unmounting
    unmount();
    expect(params.element.innerHTML).toEqual('');
  });

  it('unsubscribes from uiSettings', () => {
    const params = coreMock.createAppMountParameters('/fake/base/path');
    const core = coreMock.createStart();
    // Create a fake Subject you can use to monitor observers
    const settings$ = new Subject();
    core.uiSettings.get$.mockReturnValue(settings$);

    // Verify mounting adds an observer
    const unmount = renderApp(params, core, {});
    expect(settings$.observers.length).toBe(1);
    // Verify no observers remaining after unmount is called
    unmount();
    expect(settings$.observers.length).toBe(0);
  });

  it('resets chrome visibility', () => {
    const params = coreMock.createAppMountParameters('/fake/base/path');
    const core = coreMock.createStart();

    // Verify stateful Core API was called on mount
    const unmount = renderApp(params, core, {});
    expect(core.chrome.setIsVisible).toHaveBeenCalledWith(false);
    core.chrome.setIsVisible.mockClear(); // reset mock
    // Verify stateful Core API was called on unmount
    unmount();
    expect(core.chrome.setIsVisible).toHaveBeenCalledWith(true);
  })
});

SavedObjectsClient

Unit Tests

To unit test code that uses the Saved Objects client mock the client methods and make assertions against the behaviour you would expect to see.

Since the Saved Objects client makes network requests to an external Elasticsearch cluster, it's important to include failure scenarios in your test cases.

When writing a view with which a user might interact, it's important to ensure your code can recover from exceptions and provide a way for the user to proceed. This behaviour should be tested as well.

Below is an example of a Jest Unit test suite that mocks the server-side Saved Objects client:

typescript

// src/platform/plugins/myplugin/server/lib/short_url_lookup.ts
import crypto from 'node:crypto';
import { SavedObjectsClientContract } from '@kbn/core/server';

export const shortUrlLookup = {
  generateUrlId(url: string, savedObjectsClient: SavedObjectsClientContract) {
    const id = crypto
      .createHash('md5')
      .update(url)
      .digest('hex');

    return savedObjectsClient
      .create(
        'url',
        {
          url,
          accessCount: 0,
          createDate: new Date().valueOf(),
          accessDate: new Date().valueOf(),
        },
        { id }
      )
      .then(doc => doc.id)
      .catch(err => {
        if (savedObjectsClient.errors.isConflictError(err)) {
          return id;
        } else {
          throw err;
        }
      });
  },
};

typescript

// src/platform/plugins/myplugin/server/lib/short_url_lookup.test.ts
import { shortUrlLookup } from './short_url_lookup';
import { savedObjectsClientMock } from '@kbn/core/server/mocks';

describe('shortUrlLookup', () => {
  const ID = 'bf00ad16941fc51420f91a93428b27a0';
  const TYPE = 'url';
  const URL = 'http://elastic.co';

  const mockSavedObjectsClient = savedObjectsClientMock.create();

  beforeEach(() => {
    jest.resetAllMocks();
  });

  describe('generateUrlId', () => {
    it('provides correct arguments to savedObjectsClient', async () => {
      const ATTRIBUTES = {
        url: URL,
        accessCount: 0,
        createDate: new Date().valueOf(),
        accessDate: new Date().valueOf(),
      };
      mockSavedObjectsClient.create.mockResolvedValueOnce({
        id: ID,
        type: TYPE,
        references: [],
        attributes: ATTRIBUTES,
      });
      await shortUrlLookup.generateUrlId(URL, mockSavedObjectsClient);

      expect(mockSavedObjectsClient.create).toHaveBeenCalledTimes(1);
      const [type, attributes, options] = mockSavedObjectsClient.create.mock.calls[0];
      expect(type).toBe(TYPE);
      expect(attributes).toStrictEqual(ATTRIBUTES);
      expect(options).toStrictEqual({ id: ID });
    });

    it('ignores version conflict and returns id', async () => {
      mockSavedObjectsClient.create.mockRejectedValueOnce(
        mockSavedObjectsClient.errors.decorateConflictError(new Error())
      );
      const id = await shortUrlLookup.generateUrlId(URL, mockSavedObjectsClient);
      expect(id).toEqual(ID);
    });

    it('rejects with passed through savedObjectsClient errors', () => {
      const error = new Error('oops');
      mockSavedObjectsClient.create.mockRejectedValueOnce(error);
      return expect(shortUrlLookup.generateUrlId(URL, mockSavedObjectsClient)).rejects.toBe(error);
    });
  });
});

The following is an example of a public saved object unit test. The biggest difference with the server-side test is the slightly different Saved Objects client API which returns SimpleSavedObject instances which needs to be reflected in the mock.

typescript

// src/platform/plugins/myplugin/public/saved_query_service.ts
import {
  SavedObjectsClientContract,
  SimpleSavedObject,
} from '@kbn/core/public';

export type SavedQueryAttributes = {
  title: string;
  description: 'bar';
  query: {
    language: 'kuery';
    query: 'response:200';
  };
};

export const createSavedQueryService = (savedObjectsClient: SavedObjectsClientContract) => {
  const saveQuery = async (
    attributes: SavedQueryAttributes
  ): Promise<SimpleSavedObject<SavedQueryAttributes>> => {
    try {
      return await savedObjectsClient.create<SavedQueryAttributes>('query', attributes, {
        id: attributes.title as string,
      });
    } catch (err) {
      throw new Error('Unable to create saved query, please try again.');
    }
  };

  return {
    saveQuery,
  };
};

typescript

// src/platform/plugins/myplugin/public/saved_query_service.test.ts
import { createSavedQueryService, SavedQueryAttributes } from './saved_query_service';
import { savedObjectsServiceMock } from '@kbn/core/public/mocks';
import { SavedObjectsClientContract, SimpleSavedObject } from '@kbn/core/public';

describe('saved query service', () => {
  const savedQueryAttributes: SavedQueryAttributes = {
    title: 'foo',
    description: 'bar',
    query: {
      language: 'kuery',
      query: 'response:200',
    },
  };

  const mockSavedObjectsClient = savedObjectsServiceMock.createStartContract()
    .client as jest.Mocked<SavedObjectsClientContract>;

  const savedQueryService = createSavedQueryService(mockSavedObjectsClient);

  afterEach(() => {
    jest.resetAllMocks();
  });

  describe('saveQuery', function() {
    it('should create a saved object for the given attributes', async () => {
      // The public Saved Objects client returns instances of
      // SimpleSavedObject, so we create an instance to return from our mock.
      const mockReturnValue = new SimpleSavedObject(mockSavedObjectsClient, {
        type: 'query',
        id: 'foo',
        attributes: savedQueryAttributes,
        references: [],
      });
      mockSavedObjectsClient.create.mockResolvedValue(mockReturnValue);

      const response = await savedQueryService.saveQuery(savedQueryAttributes);
      expect(mockSavedObjectsClient.create).toHaveBeenCalledWith('query', savedQueryAttributes, {
        id: 'foo',
      });
      expect(response).toBe(mockReturnValue);
    });

    it('should reject with an error when saved objects client errors', async done => {
      mockSavedObjectsClient.create.mockRejectedValue(new Error('timeout'));

      try {
        await savedQueryService.saveQuery(savedQueryAttributes);
      } catch (err) {
        expect(err).toMatchInlineSnapshot(
          `[Error: Unable to create saved query, please try again.]`
        );
        done();
      }
    });
  });
});

Integration Tests

To get the highest confidence in how your code behaves when using the Saved Objects client, you should write at least a few integration tests which loads data into and queries a real Elasticsearch database.

To do that we'll write a Jest integration test using TestUtils to start Kibana and esArchiver to load fixture data into Elasticsearch.

Create the fixtures data you need in Elasticsearch
Create a fixtures archive with node scripts/es_archiver save <path> [index patterns...]
Load the fixtures in your test using esArchiver esArchiver.load('path from root of repo');

todo: fully worked out example

Saved Objects model versions

Also see <DocLink id="kibDevTutorialSavedObject" section="model-versions" text="Defining model versions"/>.

Model versions definitions are more structured than the legacy migration functions, which makes them harder to test without the proper tooling. This is why a set of testing tools and utilities are exposed from the @kbn/core-test-helpers-model-versions package, to help properly test the logic associated with model version and their associated transformations.

Tooling for unit tests

For unit tests, the package exposes utilities to easily test the impact of transforming documents from a model version to another one, either upward or backward.

Model version test migrator

The createModelVersionTestMigrator helper allows to create a test migrator that can be used to test model version changes between versions, by transforming documents the same way the migration algorithm would during an upgrade.

Example:

import {
  createModelVersionTestMigrator,
  type ModelVersionTestMigrator
} from '@kbn/core-test-helpers-model-versions';

const mySoTypeDefinition = someSoType();

describe('mySoTypeDefinition model version transformations', () => {
  let migrator: ModelVersionTestMigrator;

  beforeEach(() => {
    migrator = createModelVersionTestMigrator({ type: mySoTypeDefinition });
  });

  describe('Model version 2', () => {
    it('properly backfill the expected fields when converting from v1 to v2', () => {
      const obj = createSomeSavedObject();

      const migrated = migrator.migrate({
        document: obj,
        fromVersion: 1,
        toVersion: 2,
      });

      expect(migrated.properties).toEqual(expectedV2Properties);
    });

    it('properly removes the expected fields when converting from v2 to v1', () => {
      const obj = createSomeSavedObject();

      const migrated = migrator.migrate({
        document: obj,
        fromVersion: 2,
        toVersion: 1,
      });

      expect(migrated.properties).toEqual(expectedV1Properties);
    });
  });
});

Tooling for integration tests

During integration tests, we can boot a real Elasticsearch cluster, allowing us to manipulate SO documents in a way almost similar to how it would be done on production runtime. With integration tests, we can even simulate the cohabitation of two Kibana instances with different model versions to assert the behavior of their interactions.

Model version test bed

The package exposes a createModelVersionTestBed function that can be used to fully setup a test bed for model version integration testing. It can be used to start and stop the ES server, and to initiate the migration between the two versions we're testing.

Example:

import {
  createModelVersionTestBed,
  type ModelVersionTestKit
} from '@kbn/core-test-helpers-model-versions';

describe('myIntegrationTest', () => {
  const testbed = createModelVersionTestBed();
  let testkit: ModelVersionTestKit;

  beforeAll(async () => {
    await testbed.startES();
  });

  afterAll(async () => {
    await testbed.stopES();
  });

  beforeEach(async () => {
    // prepare the test, preparing the index and performing the SO migration
    testkit = await testbed.prepareTestKit({
      savedObjectDefinitions: [{
        definition: mySoTypeDefinition,
        // the model version that will be used for the "before" version
        modelVersionBefore: 1,
        // the model version that will be used for the "after" version
        modelVersionAfter: 2,
      }]
    })
  });

  afterEach(async () => {
    if(testkit) {
      // delete the indices between each tests to perform a migration again
      await testkit.tearsDown();
    }
  });

  it('can be used to test model version cohabitation', async () => {
    // last registered version is `1` (modelVersionBefore)
    const repositoryV1 = testkit.repositoryBefore;
    // last registered version is `2` (modelVersionAfter)
    const repositoryV2 = testkit.repositoryAfter;

    // do something with the two repositories, e.g
    await repositoryV1.create(someAttrs, { id });
    const v2docReadFromV1 = await repositoryV2.get('my-type', id);
    expect(v2docReadFromV1.attributes).toEqual(whatIExpect);
  });
});

Limitations:

Because the test bed is only creating the parts of Core required to instantiate the two SO repositories, and because we're not able to properly load all plugins (for proper isolation), the integration test bed currently has some limitations:

no extensions are enabled
- no security
- no encryption
- no spaces
all SO types will be using the same SO index

Limitations and edge cases in serverless environments

The serverless environment, and the fact that upgrade in such environments are performed in a way where, at some point, the old and new version of the application are living in cohabitation, leads to some particularities regarding the way the SO APIs works, and to some limitations / edge case that we need to document

Using the `fields` option of the `find` savedObjects API

By default, the find API (as any other SO API returning documents) will migrate all documents before returning them, to ensure that documents can be used by both versions during a cohabitation (e.g an old node searching for documents already migrated, or a new node searching for documents not yet migrated).

However, when using the fields option of the find API, the documents can't be migrated, as some model version changes can't be applied against a partial set of attributes. For this reason, when the fields option is provided, the documents returned from find will not be migrated.

Which is why, when using this option, the API consumer needs to make sure that all the fields passed to the fields option were already present in the prior model version. Otherwise, it may lead to inconsistencies during upgrades, where newly introduced or backfilled fields may not necessarily appear in the documents returned from the search API when the option is used.

(note: both the previous and next version of Kibana must follow this rule then)

Using `bulkUpdate` for fields with large `json` blobs

The savedObjects bulkUpdate API will update documents client-side and then reindex the updated documents. These update operations are done in-memory, and cause memory constraint issues when updating many objects with large json blobs stored in some fields. As such, we recommend against using bulkUpdate for savedObjects that:

use arrays (as these tend to be large objects)
store large json blobs in some fields

Elasticsearch

How to test ES clients

Plugin integrations

In the Kibana Platform, all plugin's dependencies to other plugins are explicitly declared in their kibana.json manifest. As for core, the dependencies setup and start contracts are injected in your plugin's respective setup and start phases. One of the upsides with testing is that every usage of the dependencies is explicit, and that the plugin's contracts must be propagated to the parts of the code using them, meaning that isolating a specific logical component for unit testing is way easier than in legacy.

The approach to test parts of a plugin's code that is relying on other plugins is quite similar to testing code using core APIs: it's expected to mock the dependency, and make it return the value the test is expecting.

Most plugins are defining mocks for their contracts. The convention is to expose them in a mocks file in my_plugin/server and/or my_plugin/public. For example for the data plugin, the client-side mocks are located in src/plugins/data/public/mocks.ts. When such mocks are present, it's strongly recommended to use them when testing against dependencies. Otherwise, one should create it's own mocked implementation of the dependency's contract (and should probably ping the plugin's owner to ask them to add proper contract mocks).

Preconditions

For these examples, we are going to see how we should test the myPlugin plugin.

This plugin declares the data plugin as a required dependency and the usageCollection plugin as an optional one. It also exposes a getSpecialSuggestions API in it's start contract, which relies on the data plugin to retrieve data.

MyPlugin plugin definition:

typescript

// src/platform/plugins/myplugin/public/plugin.ts
import { METRIC_TYPE } from '@kbn/analytics';
import { CoreSetup, CoreStart, Plugin } from '@kbn/core/public';
import { DataPublicPluginSetup, DataPublicPluginStart } from '../../data/public';
import { UsageCollectionSetup } from '../../usage_collection/public';
import { SuggestionsService } from './suggestions';

interface MyPluginSetupDeps {
  data: DataPublicPluginSetup;
  usageCollection?: UsageCollectionSetup;
}

interface MyPluginStartDeps {
  data: DataPublicPluginStart;
}

export class MyPlugin implements Plugin<MyPluginSetup, MyPluginStart, MyPluginSetupDeps, MyPluginStartDeps> {
  private suggestionsService = new SuggestionsService();

  public setup(core: CoreSetup, { data, usageCollection }: MyPluginSetupDeps) {
    // setup our internal service
    this.suggestionsService.setup(data);

    // an example on using an optional dependency that will be tested
    if (usageCollection) {
      usageCollection.reportUiCounter('my_plugin', METRIC_TYPE.LOADED, 'my_event');
    }
    // or in a shorter version
    usageCollection?.reportUiCounter('my_plugin', METRIC_TYPE.LOADED, 'my_event');

    return {};
  }

  public start(core: CoreStart, { data }: MyPluginStartDeps) {
    const suggestions = this.suggestionsService.start(data);
    return {
      getSpecialSuggestions: (query: string) => suggestions.getSuggestions(query),
    };
  }

  public stop() {}
}

export type MyPluginSetup = ReturnType<MyPlugin['setup']>;
export type MyPluginStart = ReturnType<MyPlugin['start']>;

The underlying SuggestionsService implementation:

typescript

// src/platform/plugins/myplugin/public/suggestions/suggestion_service.ts
import { DataPublicPluginSetup, DataPublicPluginStart } from '@kbn/data-plugin/public';

// stubs for testing purposes
const suggestDependingOn = (...args: any[]) => [];
const baseOptions = {} as any;
export const defaultSuggestions = [
  {
    text: 'a default suggestion',
  },
] as any[];

export class SuggestionsService {
  public setup(data: DataPublicPluginSetup) {
    // register a suggestion provider to the `data` dependency plugin
    data.autocomplete.addQuerySuggestionProvider('fr', async args => {
      return suggestDependingOn(args);
    });
  }

  public start(data: DataPublicPluginStart) {
    return {
      getSuggestions: async (query: string) => {
        // use the `data` plugin contract to retrieve arbitrary data
        // note: this logic does not really make any sense and is only here to introduce a behavior to test
        const baseSuggestions = await data.autocomplete.getQuerySuggestions({
          ...baseOptions,
          query,
        });
        if (!baseSuggestions || baseSuggestions.length === 0) {
          return defaultSuggestions;
        }
        return baseSuggestions.filter(suggestion => suggestion.type !== 'conjunction');
      },
    };
  }
}

Testing dependencies usages

A plugin should test expected usage and calls on it's dependency plugins' API.

Some calls, such as 'registration' APIs exposed from dependency plugins, should be checked, to ensure both that they are actually executed, and performed with the correct parameters.

For our example plugin's SuggestionsService, we should assert that the suggestion provider is correctly registered to the data plugin during the setup phase, and that getSuggestions calls autocomplete.getQuerySuggestions with the correct parameters.

typescript

// src/platformplugins/myplugin/public/suggestions/suggestion_service.test.ts
import {
  dataPluginMock,
  Setup as DataPluginSetupMock,
  Start as DataPluginStartMock,
} from '@kbn/data-plugin/public/mocks';
import { SuggestionsService } from './suggestion_service';

describe('SuggestionsService', () => {
  let service: SuggestionsService;
  let dataSetup: DataPluginSetupMock;
  let dataStart: DataPluginStartMock;

  beforeEach(() => {
    service = new SuggestionsService();
    dataSetup = dataPluginMock.createSetupContract();
    dataStart = dataPluginMock.createStartContract();
  });

  describe('#setup', () => {
    it('registers the query suggestion provider to the data plugin', () => {
      service.setup(dataSetup);

      expect(dataSetup.autocomplete.addQuerySuggestionProvider).toHaveBeenCalledTimes(1);
      expect(dataSetup.autocomplete.addQuerySuggestionProvider).toHaveBeenCalledWith(
        'fr',
        expect.any(Function)
      );
    });
  });

  describe('#start', () => {
    describe('#getSuggestions', () => {
      it('calls getQuerySuggestions with the correct query', async () => {
        service.setup(dataSetup);
        const serviceStart = service.start(dataStart);

        await serviceStart.getSuggestions('some query');

        expect(dataStart.autocomplete.getQuerySuggestions).toHaveBeenCalledTimes(1);
        expect(dataStart.autocomplete.getQuerySuggestions).toHaveBeenCalledWith(
          expect.objectContaining({
            query: 'some query',
          })
        );
      });
    });
  });
});

Testing components consuming the dependencies

When testing parts of your plugin code that depends on the dependency plugin's data, the best approach is to mock the dependency to be able to get the behavior expected for the test.

In this example, we are going to mock the results of autocomplete.getQuerySuggestions to be able to test the service's getSuggestions method.

typescript

// src/platform/plugins/myplugin/public/suggestions/suggestion_service.ts

describe('#start', () => {
  describe('#getSuggestions', () => {
    it('returns the default suggestions when autocomplete returns no results', async () => {
      dataStart.autocomplete.getQuerySuggestions.mockResolvedValue([]);

      service.setup(dataSetup);
      const serviceStart = service.start(dataStart);

      const results = await serviceStart.getSuggestions('some query');
      expect(results).toEqual(defaultSuggestions);
    });

    it('excludes conjunctions from the autocomplete results', async () => {
      dataStart.autocomplete.getQuerySuggestions.mockResolvedValue([
        {
          type: 'field',
          text: 'field suggestion',
        },
        {
          type: 'conjunction',
          text: 'conjunction suggestion',
        },
      ]);

      service.setup(dataSetup);
      const serviceStart = service.start(dataStart);

      const results = await serviceStart.getSuggestions('some query');

      expect(results).toEqual([
        {
          type: 'field',
          text: 'field suggestion',
        },
      ]);
    });
  });
});

Testing optional plugin dependencies

Plugins should test that their behavior remains correct when their optional dependencies are either available or not.

A basic test would be to ensure that the plugin properly initialize without error when the optional dependency is missing:

typescript

// src/platform/plugins/myplugin/public/plugin.test.ts
import { coreMock } from '@kbn/core/public/mocks';
import { dataPluginMock } from '@kbn/data-plugin/public/mocks';
import { MyPlugin } from './plugin';

describe('Plugin', () => {
  it('initializes correctly if usageCollection is disabled', () => {
    const plugin = new MyPlugin(coreMock.createPluginInitializerContext());
    const coreSetup = coreMock.createSetup();
    const setupDeps = {
      data: dataPluginMock.createSetupContract(),
      // optional usageCollector dependency is not available
    };

    const coreStart = coreMock.createStart();
    const startDeps = {
      data: dataPluginMock.createStartContract(),
    };

    expect(() => {
      plugin.setup(coreSetup, setupDeps);
    }).not.toThrow();
    expect(() => {
      plugin.start(coreStart, startDeps);
    }).not.toThrow();
  });
});

Then we should test that when optional dependency is properly used when present:

typescript

// src/platform/plugins/myplugin/public/plugin.test.ts
import { coreMock } from '@kbn/core/public/mocks';
import { dataPluginMock } from '@kbn/data-plugin/public/mocks';
import { usageCollectionPluginMock } from '@kbn/usage-collection-plugin/public/mocks';

import { METRIC_TYPE } from '@kbn/analytics';

import { MyPlugin } from './plugin';

describe('Plugin', () => {
  // [...]

  it('enables trackUserAgent when usageCollection is available', async () => {
    const plugin = new MyPlugin(coreMock.createPluginInitializerContext());
    const coreSetup = coreMock.createSetup();
    const usageCollectionSetup = usageCollectionPluginMock.createSetupContract();
    const setupDeps = {
      data: dataPluginMock.createSetupContract(),
      usageCollection: usageCollectionSetup,
    };

    plugin.setup(coreSetup, setupDeps);

    expect(usageCollectionSetup.reportUiCounter).toHaveBeenCalledTimes(2);
    expect(usageCollectionSetup.reportUiCounter).toHaveBeenCalledWith('my_plugin', METRIC_TYPE.LOADED, 'my_event');
  });
});

RXJS testing

Testing RXJS observables with marble

Testing observable based APIs can be challenging, specially when asynchronous operators or sources are used, or when trying to assert against emission's timing.

Fortunately, RXJS comes with it's own marble testing module to greatly facilitate that kind of testing.

See the official doc for more information about marble testing.

Preconditions

The following examples all assume that the following snippet is included in every test file:

typescript

import { TestScheduler } from 'rxjs/testing';

const getTestScheduler = () =>
  new TestScheduler((actual, expected) => {
    expect(actual).toEqual(expected);
  });

getTestScheduler creates a TestScheduler that is wired on jest's expect statement when comparing an observable's time frame.

Examples

Testing an interval based observable

Here is a very basic example of an interval-based API:

typescript

class FooService {
  setup() {
    return {
      getUpdate$: () => {
        return interval(100).pipe(map((count) => `update-${count + 1}`));
      },
    };
  }
}

If we were to be adding a test that asserts the correct behavior of this API without using marble testing, it would probably be something like:

typescript

it('getUpdate$ emits updates every 100ms', async () => {
  const service = new FooService();
  const { getUpdate$ } = service.setup();
  expect(await getUpdate$().pipe(take(3), toArray()).toPromise()).toEqual([
    'update-1',
    'update-2',
    'update-3',
  ]);
});

Note that if we are able to test the correct value of each emission, we don't have any way to assert that the interval of 100ms was respected. Even using a subscription based test to try to do so would result in potential flakiness, as the subscription execution could trigger on the 101ms time frame for example.

It also may be important to note:

as we need to convert the observable to a promise and wait for the result, the test is async
we need to perform observable transformation (take + toArray) in the test to have an usable value to assert against.

Marble testing would allow to get rid of these limitations. An equivalent and improved marble test could be:

typescript

  describe('getUpdate$', () => {
    it('emits updates every 100ms', () => {
      getTestScheduler().run(({ expectObservable }) => {
        const { getUpdate$ } = service.setup();
        expectObservable(getUpdate$(), '301ms !').toBe('100ms a 99ms b 99ms c', {
          a: 'update-1',
          b: 'update-2',
          c: 'update-3',
        });
      });
    });
  });

Notes:

the test is now synchronous
the second parameter of expectObservable ('301ms !') is used to perform manual unsubscription to the observable, as interval never ends.
an emission is considered a time frame, meaning that after the initial a emission, we are at the frame 101, not 100 which is why we are then only using a 99ms gap between a->b and b->c.

Testing observable completion

Let's 'improve' our getUpdate$ API by allowing the consumer to manually terminate the observable chain using a new abort$ option:

typescript

class FooService {
  setup() {
    return {
      // note: using an abortion observable is usually an anti-pattern, as unsubscribing from the observable
      // is, most of the time, a better solution. This is only used for the example purpose.
      getUpdate$: ({ abort$ = EMPTY }: { abort$?: Observable<undefined> } = {}) => {
        return interval(100).pipe(
            takeUntil(abort$),
            map((count) => `update-${count + 1}`)
        );
      },
    };
  }
}

We would then add a test to assert than this new option usage is respected:

typescript

it('getUpdate$ completes when `abort$` emits', () => {
  const service = new FooService();
  getTestScheduler().run(({ expectObservable, hot }) => {
    const { getUpdate$ } = service.setup();
    const abort$ = hot('149ms a', { a: undefined });
    expectObservable(getUpdate$({ abort$ })).toBe('100ms a 48ms |', {
      a: 'update-1',
    });
  });
});

Notes:

the | symbol represents the completion of the observable.
we are here using the hot testing utility to create the abort$ observable to ensure correct emission timing.

Testing observable errors

Testing errors thrown by the observable is very close to the previous examples and is done using the third parameter of expectObservable.

Say we have a service in charge of processing data from an observable and returning the results in a new observable:

typescript

interface SomeDataType {
  id: string;
}

class BarService {
  setup() {
    return {
      processDataStream: (data$: Observable<SomeDataType>) => {
        return data$.pipe(
          map((data) => {
            if (data.id === 'invalid') {
              throw new Error(`invalid data: '${data.id}'`);
            }
            return {
              ...data,
              processed: 'additional-data',
            };
          })
        );
      },
    };
  }
}

We could write a test that asserts the service properly emit processed results until an invalid data is encountered:

typescript

it('processDataStream throw an error when processing invalid data', () => {
  getTestScheduler().run(({ expectObservable, hot }) => {
    const service = new BarService();
    const { processDataStream } = service.setup();

    const data = hot('--a--b--(c|)', {
      a: { id: 'a' },
      b: { id: 'invalid' },
      c: { id: 'c' },
    });

    expectObservable(processDataStream(data)).toBe(
      '--a--#',
      {
        a: { id: 'a', processed: 'additional-data' },
      },
      `'[Error: invalid data: 'invalid']'`
    );
  });
});

Notes:

the - symbol represents one virtual time frame.
the # symbol represents an error.
when throwing custom Error classes, the assertion can be against an error instance, but this doesn't work with base errors.

Testing promise based observables

In some cases, the observable we want to test is based on a Promise (like of(somePromise).pipe(...)). This can occur when using promise-based services, such as core's http, for instance.

typescript

export const callServerAPI = (
  http: HttpStart,
  body: Record<string, any>,
  { abort$ }: { abort$: Observable<undefined> }
): Observable<SomeResponse> => {
  let controller: AbortController | undefined;
  if (abort$) {
    controller = new AbortController();
    abort$.subscribe(() => {
      controller!.abort();
    });
  }
  return from(
    http.post<SomeResponse>('/api/endpoint', {
      body,
      signal: controller?.signal,
    })
  ).pipe(
    takeUntil(abort$ ?? EMPTY),
    map((response) => response.results)
  );
};

Testing that kind of promise based observable does not work out of the box with marble testing, as the asynchronous promise resolution is not handled by the test scheduler's 'sandbox'.

Fortunately, there are workarounds for this problem. The most common one being to mock the promise-returning API to return an observable instead for testing, as of(observable) also works and returns the input observable.

Note that when doing so, the test suite must also include tests using a real promise value to ensure correct behavior in real situation.

typescript


// NOTE: test scheduler do not properly work with promises because of their asynchronous nature.
// we are cheating here by having `http.post` return an observable instead of a promise.
// this still allows more finely grained testing about timing, and asserting that the method
// works properly when `post` returns a real promise is handled in other tests of this suite

it('callServerAPI result observable emits when the response is received', () => {
  const http = httpServiceMock.createStartContract();
  getTestScheduler().run(({ expectObservable, hot }) => {
    // need to cast the observable as `any` because http.post.mockReturnValue expects a promise, see previous comment
    http.post.mockReturnValue(hot('---(a|)', { a: { someData: 'foo' } }) as any);

    const results = callServerAPI(http, { query: 'term' }, {});

    expectObservable(results).toBe('---(a|)', {
      a: { someData: 'foo' },
    });
  });
});

it('completes without returning results if aborted$ emits before the response', () => {
  const http = httpServiceMock.createStartContract();
  getTestScheduler().run(({ expectObservable, hot }) => {
    // need to cast the observable as `any` because http.post.mockReturnValue expects a promise, see previous comment
    http.post.mockReturnValue(hot('---(a|)', { a: { someData: 'foo' } }) as any);
    const aborted$ = hot('-(a|)', { a: undefined });
    const results = callServerAPI(http, { query: 'term' }, { aborted$ });

    expectObservable(results).toBe('-|');
  });
});

Strategy

Core Integrations

Core Mocks

Example

Strategies for specific Core APIs

HTTP Routes

Preconditions

Unit testing

Example

Integration tests

Functional Test Runner

Example

TestUtils

Example

Applications

Example

SavedObjectsClient

Unit Tests

Integration Tests

Saved Objects model versions

Tooling for unit tests

Model version test migrator

Tooling for integration tests

Model version test bed

Limitations and edge cases in serverless environments

Using the fields option of the find savedObjects API

Using bulkUpdate for fields with large json blobs

Elasticsearch

Plugin integrations

Preconditions

Testing dependencies usages

Testing components consuming the dependencies

Testing optional plugin dependencies

RXJS testing

Testing RXJS observables with marble

Preconditions

Examples

Testing an interval based observable

Testing observable completion

Testing observable errors

Testing promise based observables

Using the `fields` option of the `find` savedObjects API

Using `bulkUpdate` for fields with large `json` blobs