@atlaskit/feature-gate-js-client

FeatureGateJsClient

Atlassians wrapper for the Statsig js-lite client.

Usage

import FeatureGateJsClient from '@atlaskit/feature-gate-js-client';

Detailed docs and example usage can be found here.

What is this repository for?

The js-client covers frontend feature gate use cases. This client is modelled around bootstrapping feature gate values from the backend and does not receive live updates.

Client usage

Installation

The client can be pulled from the Artifactory NPM repository.

yarn add @atlaskit/feature-gate-js-client

Initialization

The client must be initialized before attempted usage or it will throw an error.

There are two ways to initialize your client:

1. Default initialization mechanism

This will initialize the client by calling out to feature-flag-service (fx3), and bootstrapping the client with the returned values. If the client fails to initialize for any reason, including taking longer than 2 seconds to fetch the values, default values will be used.

import FeatureGates, { FeatureGateEnvironment, FeatureGateProducts } from '@atlaskit/feature-gate-js-client';

try {
  await FeatureGates.initialize(
    {
      // This is an fx3 api key used to fetch the feature flag values. 
      // Supported keys found at go/fx3/resources/api-keys
      apiKey: 'client-test',
      // This is the environment that you are operating in, targeting rules can target specific environments
      environment: FeatureGateEnvironment.Production,
      // This will be used to filter data from Statsig to only one target app.
      // View [doc](https://hello.atlassian.net/wiki/spaces/MEASURE/pages/2955970231/How-to+Use+TargetApps+in+Statsig)  
      // for details on using targetApp.  
      targetApp: 'jira_web',
      // [Optional] Default is 2000ms
      // This is the fetch timeout used for requests to feature-flag-service to get values to bootstrap the client.
      // The higher this value the longer your application will need to wait to render if you block on client initialization
      // Too low and your application will be more likely to fallback on default values
      fetchTimeoutMs: 1000,
      // [Optional] Default is false
      // This is a boolean to indicate whether to use the `gateway/api` url for the request to feature flag service.
      // Note that this option takes precendence over the environment and perimeter options in building the url.
      // To be used for applications with strict cross-origin policies, as it will keep all requests to the same origin.
      useGatewayUrl: true,
      // [Optional] Must be one of the strings from the exported enum PerimeterType.
      // If provided, will build base url for the `feature-flag-service` based on environment and perimeter type, and
      // will disable all logging to Statsig in perimeters where it is prohibited..
      perimeter: 'fedramp-moderate'
    },
    {
      // These are expected identifiers, you must provide them if they are relevant to your product and will be added to exposures events
      analyticsAnonymousId: '<analyticsAnonymousId>',
      atlassianAccountId: '<aaid>',
      atlassianOrgId: '<orgid>',
      tenantId: '<tenantid>',
      transactionAccountId: '<transactionAccountId'>,
      trelloUserId: '<trelloUserId'>,
      trelloWorkspaceId: '<trelloWorkspaceId>'
    }, 
    {
      // These are additional custom attributes that can be used for targeting and will be added to exposure events
      exampleCustomAttribute: '<attributeValue>',
    }
  );
} catch (err) {
  console.error("Failed to initialize FeatureGates client.", err);
}

If your application has a log-in flow or other mechanism that makes it possible for the user to change during a session, then you can use the updateUser method to apply this change. The signature of this method is almost identical to initialize, except that it only requires options that relate to the network call it will perform to fetch the new set of values.

IMPORTANT: Calling this method will completely re-initialize the client with a new set of flags. You will need to re-render the entire page after this completes to ensure everything picks up the new flag values. You should avoid using this frequently as it has implications on the user experience.

import FeatureGates, { FeatureGateEnvironment, FeatureGateProducts } from '@atlaskit/feature-gate-js-client';

try {
  await FeatureGates.updateUser(
    {
      // This is an fx3 api key used to fetch the feature flag values. 
      // Supported keys found at go/fx3/resources/api-keys
      apiKey: 'client-test',
      // This is the environment that you are operating in, targeting rules can target specific environments
      environment: FeatureGateEnvironment.Production,
      // [Optional] Default is 2000ms
      // This is the fetch timeout used for requests to feature-flag-service to get values to bootstrap the client.
      fetchTimeoutMs: 1000,
      // [Optional] Default is false
      // This is a boolean to indicate whether to use the `gateway/api` url for the request to feature flag service.
      // Note that this option takes precendence over the environment and perimeter options in building the url.
      // To be used for applications with strict cross-origin policies, as it will keep all requests to the same origin.
      useGatewayUrl: true,
      // [Optional] Default is COMMERCIAL. Must be one of the values from the exported enum PerimeterType.
      // If provided, will build base url for the `feature-flag-service` based on environment and perimeter type, and
      // will disable all logging to Statsig in perimeters where it is prohibited.
      perimeter: PerimeterType.FEDRAMP_MODERATE
    },
    {
      // These are expected identifiers, you must provide them if they are relevant to your product and will be added to exposures events
      analyticsAnonymousId: '<analyticsAnonymousId>',
      atlassianAccountId: '<aaid>',
      atlassianOrgId: '<orgid>',
      tenantId: '<tenantid>',
      transactionAccountId: '<transactionAccountId'>,
      trelloUserId: '<trelloUserId'>,
      trelloWorkspaceId: '<trelloWorkspaceId>'
    }, 
    {
      // These are additional custom attributes that can be used for targeting and will be added to exposure events
      exampleCustomAttribute: '<attributeValue>',
    }
  );
} catch (err) {
  console.error("Failed to update the FeatureGates user.", err);
}

2. Initializing from values

You must fetch the values yourself using one of our wrapper backend clients (Also found in this repo) and providing them to this frontend client

import FeatureGates, { FeatureGateEnvironment, FeatureGateProducts } from '@atlaskit/feature-gate-js-client';

try {
  await FeatureGates.initializeFromValues(
    {
      // [Optional] This should come directly from a backend or service from the Statsig.getClientInitializeResponse(user) call 
      // Or supplied from the statsig UI
      // Should be provided if it is available. It is optional so initialisation can still occur if the request to fetch the key fails
      // If not provided, a default string will be used but exposure event data will not be sent back to Statsig
      sdkKey: 'client-test',
      // This is the environment that you are operating in, targeting rules can target specific environments
      environment: FeatureGateEnvironment.Production,
      // This will be used to filter config data relevant to the listed products
      products: [FeatureGateProducts.Jira]
    },
    {
      // These are expected identifiers, you must provide them if they are relevant to your product and will be added to exposures events
      analyticsAnonymousId: '<analyticsAnonymousId>',
      atlassianAccountId: '<aaid>',
      atlassianOrgId: '<orgid>',
      tenantId: '<tenantid>',
      transactionAccountId: '<transactionAccountId'>,
      trelloUserId: '<trelloUserId'>,
      trelloWorkspaceId: '<trelloWorkspaceId>'
    }, 
    {
      // These are additional custom attributes that will be added to exposure events
      exampleCustomAttribute: '<attributeValue>',
    },
    // This should come directly from a backend or service from the Statsig.getClientInitializeResponse(user) call
    initializeValues
  );
} catch (err) {
  console.error("Failed to initialize FeatureGates client.", err);
}

If your application has a log-in flow or other mechanism that makes it possible for the user to change during a session, then you can use the updateUserWithValues method to apply this change. The signature of this method is almost identical to initializeFromValues, except that it does not require any options.

IMPORTANT: Calling this method will completely re-initialize the client with a new set of flags. You will need to re-render the entire page after this completes to ensure everything picks up the new flag values. You should avoid using this frequently as it has implications on the user experience.

import FeatureGates, { FeatureGateEnvironment, FeatureGateProducts } from '@atlaskit/feature-gate-js-client';

try {
  await FeatureGates.updateWithValues(
    {
      // These are expected identifiers, you must provide them if they are relevant to your product and will be added to exposures events
      analyticsAnonymousId: '<analyticsAnonymousId>',
      atlassianAccountId: '<aaid>',
      atlassianOrgId: '<orgid>',
      tenantId: '<tenantid>',
      transactionAccountId: '<transactionAccountId'>,
      trelloUserId: '<trelloUserId'>,
      trelloWorkspaceId: '<trelloWorkspaceId>'
    }, 
    {
      // These are additional custom attributes that will be added to exposure events
      exampleCustomAttribute: '<attributeValue>',
    },
    // This should come directly from a backend or service from the Statsig.getClientInitializeResponse(user) call
    initializeValues
  );
} catch (err) {
  console.error("Failed to update FeatureGates user.", err);
}

---

If there are any issues during initialization, then the client will be put in a mode which always returns default values, and a rejected promise will be returned. You can catch this rejected promise if you wish to record your own logs and metrics, or if you wish to stop your application from loading with the defaults.

There is only once instance of the FeatureGates client, so only the first initialize call will start the initialization. Any subsequent calls will return the existing Promise for the first initialization, and the argument values will be ignored. In order to confirm whether the client has started to initialize already you can call FeatureGates.initializeCalled()

Evaluation

In order to evaluate a gate

import FeatureGates from '@atlaskit/feature-gate-js-client';

// Note: this checkGate call will automatically fire an exposure event
if (FeatureGates.checkGate('gateName')) {
    // do something here
}

In order to evaluate an experiment

import FeatureGates from '@atlaskit/feature-gate-js-client';

// Note: this call will automatically fire an exposure event
if (FeatureGates.getExperimentValue('myExperiment', 'myBooleanParameter', false)) {
    // do something here
}

In order to use more complex experiment configuration

import FeatureGates from '@atlaskit/feature-gate-js-client';

// You can provide an optional function like this to reject any incoming values that don't meet your expectations.
// If this function does not pass, the default value will be returned instead.
const isHexCode = (value: unknown) => typeof value === 'string' && value.startsWith('#') && value.length === 7;

// Note: this call will automatically fire an exposure event. You can provide "fireExposureEvent: false" in the options if you wish to suppress it.
const buttonColor: string = FeatureGates.getExperimentValue('myExperiment', 'myButtonColorStringParameter', '#000000', {
  typeGuard: isHexCode
});

Exposure Event Logging

Exposure events are batched and sent to Statsig every 10 seconds. Statsig's domain for their event logging API is blocked by some ad blockers, so by default we are proxying these requests through xp.atlassian.com to reduce exposure loss.

Testing

Jest

Testing initialization states

You can test the various initialization states by mocking the return values for initialize and updateUser. Note that you will also need to mock initializeCalled, as this is usually updated by the real initialize function.

import FeatureGates from '@atlaskit/feature-gate-js-client';

jest.mock('@atlaskit/feature-gate-js-client', () => ({
  ...jest.requireActual('@atlaskit/feature-gate-js-client'),
  initializeCalled: jest.fn(),
  initialize: jest.fn(),
  updateUser: jest.fn(),
}));

const MockedFeatureGates = jest.mocked(FeatureGates);

describe('with successful initialization', () => {
  beforeEach(() => {
    MockedFeatureGates.initializeCalled.mockReturnValue(true);
    MockedFeatureGates.initialize.mockResolvedValue();
    MockedFeatureGates.updateUser.mockResolvedValue();
  });

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

describe('with failed initialization', () => {
  beforeEach(() => {
    MockedFeatureGates.initializeCalled.mockReturnValue(true);
    MockedFeatureGates.initialize.mockRejectedValue();
    MockedFeatureGates.updateUser.mockRejectedValue();
  });

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

describe('with pending initialization', () => {
  // These can be called within your tests transition from the pending
  // initialization state into a successful/failed state.
  let resolveInitPromise;
  let rejectInitPromise;
  beforeEach(() => {
    const initPromise = new Promise((resolve, reject) => {
      resolveInitPromise = resolve;
      rejectInitPromise = reject;
    });
    MockedFeatureGates.initializeCalled.mockReturnValue(true);
    MockedFeatureGates.initialize.mockReturnValue(initPromise);
    MockedFeatureGates.updateUser.mockReturnValue(initPromise);
  });

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

Overriding values

There are two ways that you can override values in Jest tests:

1. Using mocks
2. Using the built-in override methods

Using mocks

import FeatureGates, { DynamicConfig, EvaluationReason } from '@atlaskit/feature-gate-js-client';

jest.mock('@atlaskit/feature-gate-js-client', () => ({
  ...jest.requireActual('@atlaskit/feature-gate-js-client'),
  getExperiment: jest.fn(),
  checkGate: jest.fn()
}));

const MockedFeatureGates = jest.mocked(FeatureGates);

describe("with mocked experiments and gates", () => {
  beforeEach(() => {
    const overrides = {
      configs: {
        'example-experiment': {
          cohort: 'variation'
        }
      },
      gates: {
        'example-gate': true
      }
    };

    MockedFeatureGates.getExperiment.mockImplementation(experimentName => {
      const values = overrides.configs[experimentName] || {};
      return new DynamicConfig(experimentName, values, {
        time: Date.now(),
        reason: EvaluationReason.LocalOverride
      });
    });

    MockedFeatureGates.checkGate.mockImplementation((gateName, defaultValue) => {
      return overrides.gates[gateName] || defaultValue;
    });
  });

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

Using overrides methods

import FeatureGates, { FeatureGateEnvironment } from '@atlaskit/feature-gate-js-client';

describe("with overridden gates and experiments", () => {
  beforeAll(async () => {
    // setOverrides can only be called if the client is _actually_ initialized. You can't mock the initialization, you will have invoke it properly.
    await FeatureGates.initializeWithValues({
      environment: FeatureGateEnvironment.Development,
      sdkKey: 'client-default-key',
      localMode: true 
    }, {}, {});
  });

  beforeEach(() => {
    const overrides = {
      configs: {
        'example-experiment': {
          cohort: 'variation'
        }
      },
      gates: {
        'example-gate': true
      }
    };

    FeatureGates.setOverrides(overrides);
  });

  afterEach(() => FeatureGates.clearAllOverrides());
});

Cypress

Overriding values

The .visit command in Cypress creates a new window with its own instance of FeatureGates, so you will not be able to simply import the module and apply stubs to it.

// ❌ This will not work!

import FeatureGates, { LocalOverrides } from '@atlaskit/feature-gate-js-client';

const overrides: LocalOverrides = {
  configs: {
    'example-experiment': {
      cohort: 'variation'
    }
  },
  gates: {
    'example-gate': true
  }
};

// These interact with the FeatureGates instance that your test framework is running in
cy.stub(FeatureGates, 'checkGate', (gateName, defaultValue) => overrides.gates[gateName] || defaultValue);

cy.stub(FeatureGates, 'getExperiment', (experimentName) => {
  const values = overrides.configs[experimentName] || {};
  return new DynamicConfig(experimentName, values, {
    time: Date.now(),
    reason: EvaluationReason.LocalOverride,
  });
});

// This will creates a new window, with its own FeatureGates instance.
cy.visit("http://localhost:3000/"); 

// The test feature will not exist since the stubs don't exist on the visited window.
cy.get("#test-feature").dblclick();

Instead, you will need to obtain a reference to the client that exists on the generated window, and apply your overrides to that instead.

We have exposed a window.__FEATUREGATES_JS__ variable which will contain the instance attached to the window.

// ✅ Do this instead!

import { LocalOverrides } from '@atlaskit/feature-gate-js-client';

const overrides: LocalOverrides = {
  configs: {
    'example-experiment': {
      cohort: 'variation'
    }
  },
  gates: {
    'example-gate': true
  }
};

cy.visit('http://localhost:3001', {
  // onLoad provides a reference to the generated window, and it is also only invoked when all scripts have finished loading, so the __FEATUREGATES_JS__
  // variable will be available at this point.
  onLoad: (contentWindow) => {
    const FeatureGates = contentWindow.__FEATUREGATES_JS__;

    // Note that the client would not have been initialized by this point, so we can't use the override* or setOverrides methods.
    // We also don't want to wait until the initialization completes, because then the page may have already started to render without these overrides.
    cy.stub(FeatureGates, 'checkGate', (gateName, defaultValue) => overrides.gates[gateName] || defaultValue);

    cy.stub(FeatureGates, 'getExperiment', (experimentName) => {
      const values = overrides.configs[experimentName] || {};
      return new DynamicConfig(experimentName, values, {
        time: Date.now(),
        reason: EvaluationReason.LocalOverride,
      });
    });
  }
});

You can also set up your own custom command which listens to the next window:load event to do the same thing, which you can invoke before any page visit:

import { LocalOverrides } from '@atlaskit/feature-gate-js-client';

Cypress.Commands.add('featureGateOverrides', (overrides: LocalOverrides) => {
  // Use cy.once instead of cy.on so that this only affects the next visit.
  cy.once('window:load', (contentWindow) => {
    const FeatureGates = contentWindow.__FEATUREGATES_JS__; 

    cy.stub(FeatureGates, 'checkGate', (gateName, defaultValue) => overrides?.gates?.[gateName] || defaultValue);

    cy.stub(FeatureGates, 'getExperiment', (experimentName) => {
      const values = overrides?.configs?.[experimentName] || {};

      return new DynamicConfig(experimentName, values, {
        time: Date.now(),
        reason: EvaluationReason.LocalOverride,
      });
    });
  });
});

// Once the command has been added, you can use it in any of your tests like this
cy.featureGateOverrides({
  configs: {
    'example-experiment': {
      cohort: 'variation'
    }
  },
  gates: {
    'example-gate': true
  }
}).visit('http://localhost:3001');

Storybook

Overriding values

Storybook does not have any mocking or stubbing APIs, but you can use the override* and setOverrides methods on this client as a replacement.

Please note that the client must be initialized before these methods can be called, and that the overrides will need to be cleared after each storybook is unmounted, since they are persisted to localStorage.

The easiest way to get set up is to use the FeatureGatesInitializationWithDefaults component in our React SDK (@atlassian/feature-gates-react) with the overrides prop set, since this manages the initialization and clean-up for you. Please see the component documentation for more information.

Development

How do I get set up?

• Summary of set up
  • This repo package uses yarn
  • Run yarn install in the root directory to set up your git hooks.
  • In order to get started run yarn to install the dependencies
• How to run tests
  • In order to run all tests simply run yarn test packages/measurement/feature-gate-js-client from the platform directory
  • In order to run jest tests in watch mode while doing development run yarn test:jest --watch
  • NOTE: You may need to run yarn build @atlaskit/feature-gate-js-client to create a version.ts file thats required for some tests

Contribution guidelines

• All new logic must be tested
  • There is no need to test direct pass through of Statsig APIs
  • Transformation of arguments counts as new logic
• Code review
  • Changes must go through a pull request to be merged
• Other guidelines

Releasing

This package is part of the AFP monorepo. Create a changeset using yarn changeset and commit. Documentation

Who do I talk to?

This repo is owned by the experimentation platform team, reach out to !disturbed in #help-switcheroo-statsig if you need a hand.