ttvc

• Overview
• Get started
• Usage
  • Basic usage
  • Report metrics to a collection endpoint
  • Record a PerformanceTimeline entry
  • Client-side navigation with React Router
  • Attributing TTVC measurement cancellations
• API
  • Types
  • Functions
• Browser Support
• How does it work?
• Developing
  • Building
  • Testing
  • Releasing

Overview

ttvc provides an in-browser implementation of the VisuallyComplete metric suitable for field data collection (real-user monitoring).

Visually Complete measures the moment in time when users perceive that all the visual elements of a page have completely loaded. Once the page has reached visually complete, nothing else should change in the viewport without the user’s input.

Get started

This library is available from npm. Add it to your project using the npm or yarn package managers.

$ npm install @dropbox/ttvc

$ yarn add @dropbox/ttvc

Usage

Basic usage

import {init, onTTVC} from '@dropbox/ttvc';

// Call this as early in pageload as possible to setup instrumentation.
init({
  debug: false,
  idleTimeout: 2000,
  networkTimeout: 0,
});

// Reports the last visible change for each navigation that
// occurs during the life of this document.
const unsubscribe = onTTVC((measurement) => {
  console.log('TTVC:', measurement.duration);
});

Report metrics to a collection endpoint

import {init, onTTVC} from '@dropbox/ttvc';

init();

let measurements = [];

// capture measurements in client
onTTVC((measurement) => {
  measurements.append(measurement);
});

// flush data to server when page is hidden
document.addEventListener('visibilitychange', () => {
  if (document.visibilityState === 'hidden') {
    navigator.sendBeacon('/log', JSON.stringify(measurements));
    measurements = [];
  }
});

Record a PerformanceTimeline entry

Capture a span using the Performance Timeline API.

NOTE: Setting arbitrary start and end times with performance.measure relies on the User Timing Level 3 specification. This is not yet adopted by all major browsers.

import {init, onTTVC} from '@dropbox/ttvc';

init();

onTTVC(({start, end, duration, detail}: Metric) => {
  window.performance.measure('TTVC', {
    start,
    end,
    duration,
    detail,
  });
});

Client-side navigation with React Router

@dropbox/ttvc supports measuring client-side navigations!

What counts as navigation may be different in each application, but as long as you signal that a navigation has begun, this library can figure out the rest.

To trigger a new navigation measurement, call start() or dispatch a "locationchange" event on the window object.

// analytics.js
import {init, onTTVC} from '@dropbox/ttvc';

init();

onTTVC((measurement) => {
  console.log('TTVC:', measurement.duration);
});

// app.js
import {start} from '@dropbox/ttvc';
import React, {useEffect} from 'react';
import ReactDOM from 'react-dom';
import {BrowserRouter, useLocation} from 'react-router-dom';

ReactDOM.render(
  <BrowserRouter>
    <App />
  </BrowserRouter>,
  document.getElementById('root')
);

const App = () => {
  const location = useLocation();

  useEffect(() => {
    // Option 1: If you have access to the ttvc library, import it and
    // call start().
    start();

    // Option 2: Dispatch a custom 'locationchange' event. TTVC subscribes to
    // this and will call start() for you.
    window.dispatchEvent(new Event('locationchange'));
  }, [location]);

  return (
    <div className="App">
      <h1>Welcome to React Router!</h1>
      <Routes>
        <Route path="/" element={<Home />} />
        {/* ... more routes */}
      </Routes>
    </div>
  );
};

Attributing TTVC measurement cancellations

In certain cases, @dropbox/ttvc might discard the measurement before it is captured.

This can happen if a user interacts with or navigates away from the page, or the page is put in the background before it has reached a state ot visual completeness.

This is done to obtain a higher confidence of the measurement's accuracy, as interaction with a page can cause it to change in ways that invalidate the measurement.

However, @dropbox/ttvc provides a way to monitor these cancellations and attribute them to a specific cause. A second callback function provided to onTTVC will be called when the measurement is cancelled.

import {init, onTTVC} from '@dropbox/ttvc';

init();

onTTVC(
  (measurement) => {
    console.log('TTVC measurement captured:', measurement.duration);
  },
  (error) => {
    console.log('TTVC measurement cancelled:', error.cancellationReason);
  }
);

API

Types

Metric

export type Metric = {
  // time since timeOrigin that the navigation was triggered
  // (this will be 0 for the initial pageload)
  start: number;

  // time since timeOrigin that ttvc was marked for the current navigation
  end: number;

  // the difference between start and end; this is the value of "TTVC"
  duration: number;

  // additional metadata related to the current navigation
  detail: {
    // if ttvc ignored a stalled network request, this value will be true
    didNetworkTimeOut: boolean;

    // the most recent visible update
    // (this can be either a mutation or a load event target, whichever
    // occurred last)
    lastVisibleChange?: HTMLElement | TimestampedMutationRecord;

    // describes how the navigation being measured was initiated
    // NOTE: this extends the navigation type values defined in the W3 spec;
    // "script" is usually reported as "navigation" by the browser, but we
    // report that distinctly
    // @see https://developer.mozilla.org/en-US/docs/Web/API/PerformanceNavigationTiming/type
    navigationType: // Navigation started by clicking a link, by entering the
    // URL in the browser's address bar, or by form submission.
    | 'navigate'
      // Navigation is through the browser's reload operation.
      | 'reload'
      // Navigation is through the browser's history traversal operation.
      | 'back_forward'
      // Navigation is initiated by a prerender hint.
      | 'prerender'
      // Navigation is triggered with a script operation, e.g. in a single page application.
      | 'script';
  };
};

CancellationError

export type CancellationError = {
  // time since timeOrigin that the navigation was triggered
  start: number;

  // time since timeOrigin that cancellation occurred
  end: number;

  // the difference between start and end
  duration: number;

  // reason for cancellation
  cancellationReason: CancellationReason;

  // Optional type of event that triggered cancellation
  eventType?: string;

  // Optional target of event that triggered cancellation
  eventTarget?: EventTarget;

  // the most recent visual update; this can be either a mutation or a load event target
  lastVisibleChange?: HTMLElement | TimestampedMutationRecord;

  navigationType: NavigationType;
};

export enum CancellationReason {
  // navigation has occurred
  NEW_NAVIGATION = 'NEW_NAVIGATION',

  // page was put in background
  VISIBILITY_CHANGE = 'VISIBILITY_CHANGE',

  // user interaction occurred
  USER_INTERACTION = 'USER_INTERACTION',

  // measurement was cancelled because a new one was started
  NEW_MEASUREMENT = 'NEW_MEASUREMENT',

  // manual cancellation via API happened
  MANUAL_CANCELLATION = 'MANUAL_CANCELLATION',
}

TtvcOptions

export type TtvcOptions = {
  // decide whether to log debug messages
  debug?: boolean;

  // the duration in ms to wait before declaring the page completely idle
  idleTimeout?: number;

  // a duration in ms to wait before assuming that a single network request
  // was not instrumented correctly
  networkTimeout?: number;
};

Functions

init()

type init = (options?: TtvcOptions) -> void;

Sets up instrumentation for the current page and begins monitoring. For the most accurate results, call this as early in pageload as possible.

Accepts an optional options argument (see above).

start()

type start = () -> void;

Start a new TTVC measurement.

You do not need to call this for the initial page load. Use this to notify @dropbox/ttvc that a new client-side navigation is about to take place.

onTTVC()

type onTTVC = (subscriber: (metric: Metric) -> void) -> () => void;

Register a callback function as a subscriber to new TTVC metric measurements. Returns an "unsubscribe" function which may be called to unregister the subscribed callback function.

The callback function may be called more than once if in-page navigation occurs.

onTTVC may be called more than once to register more than one subscriber.

cancel()

type cancel = (eventType?: string) => void;

Abort the current TTVC measurement.

This method is provided as an escape hatch. Consider using cancel to notify @dropbox/ttvc that a user interaction has occurred and continuing the measurement may produce an invalid result.

An optional argument can be passed specifying the event that triggered the cancellation. Type of that event will be logged along with the cancellation to the error callback.

incrementAjaxCount() & decrementAjaxCount()

type incrementAjaxCount = () -> void;
type decrementAjaxCount = () -> void;

Use these functions to instrument AJAX requests in your application. Try to ensure incrementAjaxCount and decrementAjaxCount are called exactly once for each request, regardless of success or failure.

e.g.

// patch window.fetch
const nativeFetch = window.fetch;

window.fetch = (...args) => {
  TTVC.incrementAjaxCount();
  return nativeFetch(...args).finally(TTVC.decrementAjaxCount);
};

Browser Support

@dropbox/ttvc relies on the following browser features.

• ES6 syntax
• async/await syntax
• window.MutationObserver
• window.IntersectionObserver
• performance.now()

See browsers that support all above features on caniuse.com:

https://caniuse.com/mutationobserver,async-functions,mdn-api_intersectionobserver,es6,high-resolution-time

If you would like to use this library and support a browser that does not support one of these features, you will be responsible for supplying polyfills and/or recompiling the library to remove unsupported syntax.

How does it work?

Browsers today don't automatically report much about their paint cycles. It would definitely be an improvement if browser developers decided to report this value for us.

In the meantime, we approximate TTVC using a heuristic based on a combination of observing mutations and "load" events.

For a more detailed write-up of the architecture, check out our announcement blog post!

https://dropbox.tech/frontend/measuring-ttvc-web-performance-metric-open-source-library

Developing

To develop this package, you will need node version 16 or greater, and the yarn package manager. Once you have these prerequisites, install project dependencies with:

yarn install

Building

This project is developed with TypeScript. You can compile the TypeScript source files to JavaScript with:

$ yarn build

While testing locally, you may find it useful to build the rollup bundle in watch mode.

$ yarn build:rollup --watch

Testing

You can run all tests together with:

$ yarn test

Individual tests

There are four individual test scripts

$ yarn test:lint // runs eslint
$ yarn test:typecheck // runs typescript with --noEmit
$ yarn test:unit // runs jest unit tests
$ yarn test:e2e // runs playwright tests (requires yarn build to have been run)

Testing with Playwright

Before running any playwright tests, you will need to install the default set of browsers:

$ npx playwright install

Run test suite:

$ yarn build // build the ttvc package before testing
$ yarn test:e2e

To manually test pages, start the test server.

$ yarn express

You can launch tests in the browser by opening http://localhost:3000/test/[test-folder]. e.g. http://localhost:3000/test/ajax1/

Releasing

This package is configured to publish new releases using np.

To cut a new release, switch to the main branch and run npm run release.

NOTE: Despite the fact that this package is managed with yarn, there is currently a bug in np preventing yarn np from running correctly.

This should automate the process of releasing by doing the following:

• prompt you to select a new version
• install all dependencies
• run tests
• build a new compiled artifact
• bump the version in package.json
• push the new version to npm
• create a new tag and associated "release" on github.com

np supports a number of cli flags which may also be useful for you:

• npm run np -- --preview Do a dry run to test that your release will behave as expected
• npm run np -- patch|minor|major bump the version to the next patch/minor/major increment
• npm run np -- prerelease --tag=beta create an "prerelease" version which will not automatically be marked as latest

See the full documentation for all np CLI options.