HTML Streaming with React. Batteries-included & easy.
Note
Unfamiliar with HTML Streaming? Check out Dan's article about SSR and Streaming.
Note
The React team is working on high-level APIs that will eventually make parts of react-streaming
obsolete, see @sebmarkbage comment at "RFC: injectToStream".
Contents
- Intro
- Why Streaming
- Usage
- Usage (Library Authors)
Features (for React users)
- Unlocks
<Suspense>
for SSR apps. useAsync()
: easily fetch data for SSR apps.- Two SEO strategies:
conservative
orgoogle-speed
. - Seamless support for Node.js (serverless) platforms (Vercel, AWS EC2, ...) and Edge platforms (Cloudflare Workers, Deno Deploy, Netlify Edge, Vercel Edge, ...).
- Easy error handling.
Features (for library authors)
useAsync()
: add data fetching capabilities to your library. High-level and easy to use.injectToStream()
: inject chunks to the stream for your library. Low-level and difficult to use, but highly flexible.
Easy
import { renderToStream } from 'react-streaming/server'
const {
pipe, // Node.js (Vercel, AWS EC2, ...)
readable // Edge (Cloudflare Workers, Deno Deploy, Netlify Edge, Vercel Edge, ...)
} = await renderToStream(<Page />)
React 18's new SSR streaming architecture unlocks many capabilities:
- Easily fetch data for SSR apps.
- Fundamentally improved mobile performance. (Mobile users can progressively load the page as data is fetched, before even a single line of JavaScript is loaded. Especially important for users with a low-end device and poor internet connection.)
- Progressive Hydration. (Page is interactive before even the page has finished loading.)
Problem: the current React Streaming architecture is low-level and difficult to use.
Solution: react-streaming
.
react-streaming
makes it easy to build the libraries of tomorrow, for example:
-
Install
npm install react-streaming
-
Server-side
import { renderToStream } from 'react-streaming/server' const { pipe, // Defined if running in Node.js, otherwise `null` readable // Defined if running on Edge (e.g. Cloudflare Workers), otherwise `null` } = await renderToStream(<Page />)
That's it.
const options = {
// ...
}
await renderToStream(<Page />, options)
-
options.disable?: boolean
: Disable streaming.<Page>
is still rendered to a stream, but the promiseconst promise = renderToStream()
resolves only after the stream has finished. (This effectively disables streaming from a user perspective, while unlocking React 18 Streaming capabilities such as SSR<Supsense>
.) -
options.seoStrategy?: 'conservative' | 'google-speed'
-
conservative
(default): Disable streaming if the HTTP request originates from a bot. (Ensuring bots to always see the whole HTML.) -
google-speed
: Don't disable streaming for the Google Bot.- Pro: Google may (to be researched) rank your website higher because the initial HTTP response is faster.
- Con: Google may (to be researched) not await the HTML stream (see Bots).
-
Custom SEO strategy: use
options.disable
. For example:// Always stream, even for bots: const disable = false // Disable streaming for bots, except for the Google Bot and some other bot: const disable = isBot(userAgent) && !['googlebot', 'some-other-bot'].some(n => userAgent.toLowerCase().includes(n)) await renderToStream(<Page />, { disable })
-
-
options.userAgent?: string
: The HTTP User-Agent request header. (Needed foroptions.seoStrategy
.) -
options.webStream?: boolean
: In Node.js, use a Web Stream instead of a Node.js Stream. (Node.js 18 released Web Streams support.) -
options.streamOptions
: Options passed to React'srenderToReadableStream()
andrenderToPipeableStream()
. Use this to passnonce
, bootstrap scripts, etc. It excludes error handling options, use Error Handling instead. -
options.timeout?: number | null
(seconds): Timeout after which the rendering stream is aborted, see Abort. Defaults to 20 seconds. Set tonull
to disable automatic timeout (we recommend to then implement a manual timeout as explained at Abort). -
options.onTimeout?: () => void
: Callback when the timeout is reached. -
options.onBoundaryError?: (err: unknown) => void
: Called when a<Suspense>
boundary fails. See Error Handling. -
const { streamEnd } = await renderToStream(<Page />) // β Page Shell succesfully rendered. const success: boolean = await streamEnd // Stream ended. if (success) { // β <Page> succesfully rendered } else { // β A <Suspense> boundary failed. }
Note that
streamEnd
never rejects.β οΈ Read Error Handling before usingstreamEnd
. In particular, do not usesuccess
to change the behavior of your app/stream (because React automatically takes care of gracefully handling<Suspense>
failures).
By default, react-streaming
disables streaming for bots and crawlers, such as:
- The Google Bot, which crawls the HTML of your pages to be able to show a preview of your website on Google's result pages.
- The bot of social sites (Twitter/Instagram/WhatsApp...), which crawl the HTML of your pages to be able to show a preview of your website when it's shared on Twitter/Instagram/WhatsApp/...
Note
These bots explore your website by navigating the HTML of your pages. It isn't clear what bots do when they encounter an HTML stream (to be researched); it's therefore safer to provide bots with a fully rendered HTML at once that contains all the content of your page (i.e. disable HTML streaming) instead of hoping that bots will await the HTML stream.
For react-streaming
to be able to determine whether a request comes from a bot or a real user, you need to provide options.userAgent
.
Note
If you use Vike with vike-react
, you can simply set renderPage({ headersOriginal })
instead. (The User-Agent request header will then automatically be passed to react-streaming
).
You can implement a custom strategy, see options.seoStrategy
.
You can use $ curl
to see the HTML response that bots and crawlers receive:
# What bots and crawls get: no HTML Streaming, just "classic SSR"
$ curl http://localhost:3000/star-wars
# What regular users get: HTML Streaming
$ curl http://localhost:3000/star-wars -N -H "User-Agent: chrome"
Note
By default curl
sets User-Agent: curl/8.5.0
, which react-streaming
interprets as bot.
The promise await renderToStream()
resolves after the page shell is rendered. This means that if an error occurs while rendering the page shell, then the promise rejects with that error.
π The page shell is the set of all components before
<Suspense>
boundaries.
try {
await renderToStream(<Page />)
// β
Page shell succesfully rendered and is ready in the stream buffer.
} catch(err) {
// β Something went wrong while rendering the page shell.
}
The stream returned by await renderToStream()
doesn't emit errors.
π If an error occurs during the stream, then that means that a
<Suspense>
boundary failed. Instead of emiting a stream error, React swallows the error on the server-side and retries to resolve the<Suspense>
boundary on the client-side. If the<Suspense>
fails again on the client-side, then the client-side throws the error.This means that errors occuring during the stream are handled by React and there is nothing for you to do on the server-side. That said, you may want to gracefully handle the error on the client-side e.g. with
react-error-boundary
.You can use
options.onBoundaryError()
for error tracking purposes.
After a default timeout of 20 seconds react-streaming
aborts the rendering stream, as recommended by React here and there.
When the timeout is reached react-streaming
ends the stream and tells React to stop rendering. Note that there isn't any thrown error: React merely stops server-side rendering and continues on the client-side, see explanation at Error Handling.
You can also manually abort:
const { abort } = await renderToStream(<Page />, { timeout: null })
abort()
In general, with React Streaming, all the content of your page is included in the HTML stream. This means you get all the benefits of SSR. However, it isn't clear whether crawlers fully wait the HTML stream to complete. It's therefore safer to disable HTML Streaming for crawlers and fall back to "classical SSR", see Bots.
Note
The order in which the content of your page is included in the HTML stream depends on which data comes first. For example, if you use a loading fallback component, the content of the loading component appears first, followed by the content of the main component after the <Suspense>
boundary resolves.
import { useAsync } from 'react-streaming'
function Page({ movieId }) {
return (
<Suspense fallback={<p>Loading...</p>}>
<Movie id={movieId}/>
</Suspense>
)
}
async function fetchMovie(id) {
const response = await fetch(`https://star-wars.brillout.com/api/films/${id}.json`)
return response.json()
}
// This component is isomorphic: it works on both the client-side and server-side. The
// data fetched during SSR is automatically passed and re-used on the client-side.
function Movie({ id }) {
const key = [
'star-wars-movies',
id // Re-run `fetchMovie()` if `id` changes
]
const movie = useAsync(key, () => fetchMovie(id))
return (
<ul>
<li>
Title: {movie.title}
</li>
<li>
Release Date: {movie.release_date}
</li>
</ul>
)
}
See useAsync()
(Library Authors) for more information.
react-streaming
enables you to suspend the React rendering and await for something to happen. (Usually data fetching.)
The novelty here is that it's isomorphic:
- It works on the client-side as well as on the server-side (while Serve-Side Rendering).
- For hydration, data is passed from the server to the client. (So that data isn't loaded twice.)
You have the choice between:
useAsync()
: High-level and easy.injectToStream()
: Low-level and highly flexible (useAsync()
is based on it). Easy & recommended for injecting script and style tags. Complex for data fetching (if possible, useuseAsync()
instead).
This section is a low-level description of
useAsync()
. For a high-level description, seeuseAsync()
instead.
import { useAsync } from 'react-streaming'
function SomeComponent() {
const someAsyncFunc = async function () {
const value = 'someData'
return value
}
const key = ['some', 'invalidating', 'values']
// useAsync() suspends rendering until the promise returned by someAsyncFunc() resolves
const value = useAsync(key, someAsyncFunc)
assert(value === 'someData')
}
When <SomeComponent>
is rendered on the server-side (SSR), it injects the
resolved value into the stream and the client-side picks up the injected value. This means that the
client-side doesn't call someAsyncFunc()
: instead, the client-side re-uses the value resolved on
the server-side.
If you want someAsyncFunc()
to be re-run, then change key
. The someAsyncFunc()
is only re-run if when the component is un-mounted and re-mounted, or if key
changes. For example, changing the state of your component (e.g. with useState()
) will not re-run someAsyncFunc()
if you provide the same key
.
Usually the key is set to ['name-of-the-function', ...functionArguments]
.
You can think of
key
to serve a similar purpose to React Queries's key, and to thedeps
argument of React'suseEffect(fn, deps)
.
type Chunk = string | Buffer
type Options = { flush?: boolean }
injectToStream(chunk: Chunk | Promise<Chunk>, options?: Options)`
The injectToStream()
function enables you to inject chunks to the stream.
There are two ways to access injectToStream()
:
- With
renderToStream()
:import { renderToStream } from 'react-streaming/server' const stream = await renderToStream(<Page />) const { injectToStream } = stream
- With
useStream()
:import { useStream } from 'react-streaming' function SomeComponent() { const stream = useStream() if (stream === null) { // No stream available. This is the case: // - On the client-side. // - When `option.disable === true`. // - When react-streaming is not installed. } const { injectToStream } = stream }
Usage examples:
// Inject JavaScript (e.g. for progressive hydration)
injectToStream('<script type="module" src="/main.js"></script>', { flush: true })
// Inject CSS (e.g. for CSS-in-JS)
injectToStream('<styles>.some-component { color: blue }</styles>', { flush: true })
// Pass data to client
injectToStream(`<script type="application/json">${JSON.stringify(someData)}</script>`)
For a full example of using injectToStream()
, have a look at useAsync()
's implementation.
If setting options.flush
to true
, then the stream will be flushed after chunk
has been written to the stream. This is only applicable for Node.js streams and only if you are using a compression library that makes a flush()
method available. For example, compression
adds a res.flush()
method. The option is ignored if there isn't a flush()
method available.
Typical usage:
const makeClosableAgain = stream.doNotClose()
// Ensure chunk is injected before the stream ends
injectToStream(chunk)
makeClosableAgain()
Like injectToStream()
, there are two ways to access it:
import { renderToStream } from 'react-streaming/server'
const stream = await renderToStream(<Page />)
const { doNotClose } = stream
import { useStream } from 'react-streaming'
function SomeComponent() {
const stream = useStream()
const { doNotClose } = stream
}
Check whether the stream has ended.
Like injectToStream()
, there are two ways to access it:
import { renderToStream } from 'react-streaming/server'
const stream = await renderToStream(<Page />)
const { hasStreamEnded } = stream
import { useStream } from 'react-streaming'
function SomeComponent() {
const stream = useStream()
const { hasStreamEnded } = stream
}