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.
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
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.constunsubscribe=onTTVC((measurement)=>{console.log('TTVC:',measurement.duration);});import{init,onTTVC}from'@dropbox/ttvc';init();letmeasurements=[];// capture measurements in clientonTTVC((measurement)=>{measurements.append(measurement);});// flush data to server when page is hiddendocument.addEventListener('visibilitychange',()=>{if(document.visibilityState==='hidden'){navigator.sendBeacon('/log',JSON.stringify(measurements));measurements=[];}});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,});});@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.jsimport{init,onTTVC}from'@dropbox/ttvc';init();onTTVC((measurement)=>{console.log('TTVC:',measurement.duration);});// app.jsimport{start}from'@dropbox/ttvc';importReact,{useEffect}from'react';importReactDOMfrom'react-dom';import{BrowserRouter,useLocation}from'react-router-dom';ReactDOM.render(<BrowserRouter><App/></BrowserRouter>,document.getElementById('root'));constApp=()=>{constlocation=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(newEvent('locationchange'));},[location]);return(<divclassName="App"><h1>Welcome to React Router!</h1><Routes><Routepath="/"element={<Home/>}/>{/* ... more routes */}</Routes></div>);};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);});exporttypeMetric={// 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 navigationend: number;// the difference between start and end; this is the value of "TTVC"duration: number;// additional metadata related to the current navigationdetail: {// if ttvc ignored a stalled network request, this value will be truedidNetworkTimeOut: 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/typenavigationType: // 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';};};exporttypeCancellationError={// time since timeOrigin that the navigation was triggeredstart: number;// time since timeOrigin that cancellation occurredend: number;// the difference between start and endduration: number;// reason for cancellationcancellationReason: CancellationReason;// Optional type of event that triggered cancellationeventType?: string;// Optional target of event that triggered cancellationeventTarget?: EventTarget;// the most recent visual update; this can be either a mutation or a load event targetlastVisibleChange?: HTMLElement|TimestampedMutationRecord;navigationType: NavigationType;};exportenumCancellationReason{// navigation has occurredNEW_NAVIGATION='NEW_NAVIGATION',// page was put in backgroundVISIBILITY_CHANGE='VISIBILITY_CHANGE',// user interaction occurredUSER_INTERACTION='USER_INTERACTION',// measurement was cancelled because a new one was startedNEW_MEASUREMENT='NEW_MEASUREMENT',// manual cancellation via API happenedMANUAL_CANCELLATION='MANUAL_CANCELLATION',}exporttypeTtvcOptions={// decide whether to log debug messagesdebug?: boolean;// the duration in ms to wait before declaring the page completely idleidleTimeout?: number;// a duration in ms to wait before assuming that a single network request// was not instrumented correctlynetworkTimeout?: number;};typeinit=(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).
typestart=()->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.
typeonTTVC=(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.
typecancel=(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.
typeincrementAjaxCount=()->void;typedecrementAjaxCount=()->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.fetchconstnativeFetch=window.fetch;window.fetch=(...args)=>{TTVC.incrementAjaxCount();returnnativeFetch(...args).finally(TTVC.decrementAjaxCount);};@dropbox/ttvc relies on the following browser features.
See browsers that support all above features on caniuse.com:
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.
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
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
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
You can run all tests together with:
$ yarn test
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)
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/
NOTE: This package should be released by a repository administrator. New versions are expected to be pushed directly to
main.
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 npfrom 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 -- --previewDo a dry run to test that your release will behave as expectednpm run np -- patch|minor|majorbump the version to the next patch/minor/major incrementnpm run np -- prerelease --tag=betacreate an "prerelease" version which will not automatically be marked aslatest
See the full documentation for all np CLI options.