renderToReadableStream

renderToReadableStream renders a React tree to a Readable Web Stream.

const stream = await renderToReadableStream(reactNode, options?)

Not

This API depends on Web Streams. For Node.js, use renderToPipeableStream instead.


Reference

renderToReadableStream(reactNode, options?)

Call renderToReadableStream to render your React tree as HTML into a Readable Web Stream.

import { renderToReadableStream } from 'react-dom/server';

async function handler(request) {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js']
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

On the client, call hydrateRoot to make the server-generated HTML interactive.

See more examples below.

Parameters

  • reactNode: A React node you want to render to HTML. For example, a JSX element like <App />. It is expected to represent the entire document, so the App component should render the <html> tag.

  • optional options: An object with streaming options.

    • optional bootstrapScriptContent: If specified, this string will be placed in an inline <script> tag.
    • optional bootstrapScripts: An array of string URLs for the <script> tags to emit on the page. Use this to include the <script> that calls hydrateRoot. Omit it if you don’t want to run React on the client at all.
    • optional bootstrapModules: Like bootstrapScripts, but emits <script type="module"> instead.
    • optional identifierPrefix: A string prefix React uses for IDs generated by useId. Useful to avoid conflicts when using multiple roots on the same page. Must be the same prefix as passed to hydrateRoot.
    • optional namespaceURI: A string with the root namespace URI for the stream. Defaults to regular HTML. Pass 'http://www.w3.org/2000/svg' for SVG or 'http://www.w3.org/1998/Math/MathML' for MathML.
    • optional nonce: A nonce string to allow scripts for script-src Content-Security-Policy.
    • optional onError: A callback that fires whenever there is a server error, whether recoverable or not. By default, this only calls console.error. If you override it to log crash reports, make sure that you still call console.error. You can also use it to adjust the status code before the shell is emitted.
    • optional progressiveChunkSize: The number of bytes in a chunk. Read more about the default heuristic.
    • optional signal: An abort signal that lets you abort server rendering and render the rest on the client.

Returns

renderToReadableStream returns a Promise:

The returned stream has an additional property:

  • allReady: A Promise that resolves when all rendering is complete, including both the shell and all additional content. You can await stream.allReady before returning a response for crawlers and static generation. If you do that, you won’t get any progressive loading. The stream will contain the final HTML.

Usage

Rendering a React tree as HTML to a Readable Web Stream

Call renderToReadableStream to render your React tree as HTML into a Readable Web Stream:

import { renderToReadableStream } from 'react-dom/server';

async function handler(request) {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js']
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

Along with the root component, you need to provide a list of bootstrap <script> paths. Your root component should return the entire document including the root <html> tag.

For example, it might look like this:

export default function App() {
return (
<html>
<head>
<meta charSet="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link rel="stylesheet" href="/styles.css"></link>
<title>My app</title>
</head>
<body>
<Router />
</body>
</html>
);
}

React will inject the doctype and your bootstrap <script> tags into the resulting HTML stream:

<!DOCTYPE html>
<html>
<!-- ... HTML from your components ... -->
</html>
<script src="/main.js" async=""></script>

On the client, your bootstrap script should hydrate the entire document with a call to hydrateRoot:

import { hydrateRoot } from 'react-dom/client';
import App from './App.js';

hydrateRoot(document, <App />);

This will attach event listeners to the server-generated HTML and make it interactive.

Derinlemesine İnceleme

Reading CSS and JS asset paths from the build output

The final asset URLs (like JavaScript and CSS files) are often hashed after the build. For example, instead of styles.css you might end up with styles.123456.css. Hashing static asset filenames guarantees that every distinct build of the same asset will have a different filename. This is useful because it lets you safely enable long-term caching for static assets: a file with a certain name would never change content.

However, if you don’t know the asset URLs until after the build, there’s no way for you to put them in the source code. For example, hardcoding "/styles.css" into JSX like earlier wouldn’t work. To keep them out of your source code, your root component can read the real filenames from a map passed as a prop:

export default function App({ assetMap }) {
return (
<html>
<head>
<title>My app</title>
<link rel="stylesheet" href={assetMap['styles.css']}></link>
</head>
...
</html>
);
}

On the server, render <App assetMap={assetMap} /> and pass your assetMap with the asset URLs:

// You'd need to get this JSON from your build tooling, e.g. read it from the build output.
const assetMap = {
'styles.css': '/styles.123456.css',
'main.js': '/main.123456.js'
};

async function handler(request) {
const stream = await renderToReadableStream(<App assetMap={assetMap} />, {
bootstrapScripts: [assetMap['/main.js']]
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

Since your server is now rendering <App assetMap={assetMap} />, you need to render it with assetMap on the client too to avoid hydration errors. You can serialize and pass assetMap to the client like this:

// You'd need to get this JSON from your build tooling.
const assetMap = {
'styles.css': '/styles.123456.css',
'main.js': '/main.123456.js'
};

async function handler(request) {
const stream = await renderToReadableStream(<App assetMap={assetMap} />, {
// Careful: It's safe to stringify() this because this data isn't user-generated.
bootstrapScriptContent: `window.assetMap = ${JSON.stringify(assetMap)};`,
bootstrapScripts: [assetMap['/main.js']],
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

In the example above, the bootstrapScriptContent option adds an extra inline <script> tag that sets the global window.assetMap variable on the client. This lets the client code read the same assetMap:

import { hydrateRoot } from 'react-dom/client';
import App from './App.js';

hydrateRoot(document, <App assetMap={window.assetMap} />);

Both client and server render App with the same assetMap prop, so there are no hydration errors.


Streaming more content as it loads

Streaming allows the user to start seeing the content even before all the data has loaded on the server. For example, consider a profile page that shows a cover, a sidebar with friends and photos, and a list of posts:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Sidebar>
<Friends />
<Photos />
</Sidebar>
<Posts />
</ProfileLayout>
);
}

Imagine that loading data for <Posts /> takes some time. Ideally, you’d want to show the rest of the profile page content to the user without waiting for the posts. To do this, wrap Posts in a <Suspense> boundary:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Sidebar>
<Friends />
<Photos />
</Sidebar>
<Suspense fallback={<PostsGlimmer />}>
<Posts />
</Suspense>
</ProfileLayout>
);
}

This tells React to start streaming the HTML before Posts loads its data. React will send the HTML for the loading fallback (PostsGlimmer) first, and then, when Posts finishes loading its data, React will send the remaining HTML along with an inline <script> tag that replaces the loading fallback with that HTML. From the user’s perspective, the page will first appear with the PostsGlimmer, later replaced by the Posts.

You can further nest <Suspense> boundaries to create a more granular loading sequence:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Suspense fallback={<BigSpinner />}>
<Sidebar>
<Friends />
<Photos />
</Sidebar>
<Suspense fallback={<PostsGlimmer />}>
<Posts />
</Suspense>
</Suspense>
</ProfileLayout>
);
}

In this example, React can start streaming the page even earlier. Only ProfileLayout and ProfileCover must finish rendering first because they are not wrapped in any <Suspense> boundary. However, if Sidebar, Friends, or Photos need to load some data, React will send the HTML for the BigSpinner fallback instead. Then, as more data becomes available, more content will continue to be revealed until all of it becomes visible.

Streaming does not need to wait for React itself to load in the browser, or for your app to become interactive. The HTML content from the server will get progressively revealed before any of the <script> tags load.

Read more about how streaming HTML works.

Not

Only Suspense-enabled data sources will activate the Suspense component. They include:

  • Data fetching with Suspense-enabled frameworks like Relay and Next.js
  • Lazy-loading component code with lazy
  • Reading the value of a Promise with use

Suspense does not detect when data is fetched inside an Effect or event handler.

The exact way you would load data in the Posts component above depends on your framework. If you use a Suspense-enabled framework, you’ll find the details in its data fetching documentation.

Suspense-enabled data fetching without the use of an opinionated framework is not yet supported. The requirements for implementing a Suspense-enabled data source are unstable and undocumented. An official API for integrating data sources with Suspense will be released in a future version of React.


Specifying what goes into the shell

The part of your app outside of any <Suspense> boundaries is called the shell:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Suspense fallback={<BigSpinner />}>
<Sidebar>
<Friends />
<Photos />
</Sidebar>
<Suspense fallback={<PostsGlimmer />}>
<Posts />
</Suspense>
</Suspense>
</ProfileLayout>
);
}

It determines the earliest loading state that the user may see:

<ProfileLayout>
<ProfileCover />
<BigSpinner />
</ProfileLayout>

If you wrap the whole app into a <Suspense> boundary at the root, the shell will only contain that spinner. However, that’s not a pleasant user experience because seeing a big spinner on the screen can feel slower and more annoying than waiting a bit more and seeing the real layout. This is why usually you’ll want to place the <Suspense> boundaries so that the shell feels minimal but complete—like a skeleton of the entire page layout.

The async call to renderToReadableStream will resolve to a stream as soon as the entire shell has been rendered. Usually, you’ll start streaming then by creating and returning a response with that stream:

async function handler(request) {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js']
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

By the time the stream is returned, components in nested <Suspense> boundaries might still be loading data.


Logging crashes on the server

By default, all errors on the server are logged to console. You can override this behavior to log crash reports:

async function handler(request) {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
console.error(error);
logServerCrashReport(error);
}
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
}

If you provide a custom onError implementation, don’t forget to also log errors to the console like above.


Recovering from errors inside the shell

In this example, the shell contains ProfileLayout, ProfileCover, and PostsGlimmer:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Suspense fallback={<PostsGlimmer />}>
<Posts />
</Suspense>
</ProfileLayout>
);
}

If an error occurs while rendering those components, React won’t have any meaningful HTML to send to the client. Wrap your renderToReadableStream call in a try...catch to send a fallback HTML that doesn’t rely on server rendering as the last resort:

async function handler(request) {
try {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
console.error(error);
logServerCrashReport(error);
}
});
return new Response(stream, {
headers: { 'content-type': 'text/html' },
});
} catch (error) {
return new Response('<h1>Something went wrong</h1>', {
status: 500,
headers: { 'content-type': 'text/html' },
});
}
}

If there is an error while generating the shell, both onError and your catch block will fire. Use onError for error reporting and use the catch block to send the fallback HTML document. Your fallback HTML does not have to be an error page. Instead, you may include an alternative shell that renders your app on the client only.


Recovering from errors outside the shell

In this example, the <Posts /> component is wrapped in <Suspense> so it is not a part of the shell:

function ProfilePage() {
return (
<ProfileLayout>
<ProfileCover />
<Suspense fallback={<PostsGlimmer />}>
<Posts />
</Suspense>
</ProfileLayout>
);
}

If an error happens in the Posts component or somewhere inside it, React will try to recover from it:

  1. It will emit the loading fallback for the closest <Suspense> boundary (PostsGlimmer) into the HTML.
  2. It will “give up” on trying to render the Posts content on the server anymore.
  3. When the JavaScript code loads on the client, React will retry rendering Posts on the client.

If retrying rendering Posts on the client also fails, React will throw the error on the client. As with all the errors thrown during rendering, the closest parent error boundary determines how to present the error to the user. In practice, this means that the user will see a loading indicator until it is certain that the error is not recoverable.

If retrying rendering Posts on the client succeeds, the loading fallback from the server will be replaced with the client rendering output. The user will not know that there was a server error. However, the server onError callback and the client onRecoverableError callbacks will fire so that you can get notified about the error.


Setting the status code

Streaming introduces a tradeoff. You want to start streaming the page as early as possible so that the user can see the content sooner. However, once you start streaming, you can no longer set the response status code.

By dividing your app into the shell (above all <Suspense> boundaries) and the rest of the content, you’ve already solved a part of this problem. If the shell errors, your catch block will run which lets you set the error status code. Otherwise, you know that the app may recover on the client, so you can send “OK”.

async function handler(request) {
try {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
console.error(error);
logServerCrashReport(error);
}
});
return new Response(stream, {
status: 200,
headers: { 'content-type': 'text/html' },
});
} catch (error) {
return new Response('<h1>Something went wrong</h1>', {
status: 500,
headers: { 'content-type': 'text/html' },
});
}
}

If a component outside the shell (i.e. inside a <Suspense> boundary) throws an error, React will not stop rendering. This means that the onError callback will fire, but your code will continue running without getting into the catch block. This is because React will try to recover from that error on the client, as described above.

However, if you’d like, you can use the fact that something has errored to set the status code:

async function handler(request) {
try {
let didError = false;
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
didError = true;
console.error(error);
logServerCrashReport(error);
}
});
return new Response(stream, {
status: didError ? 500 : 200,
headers: { 'content-type': 'text/html' },
});
} catch (error) {
return new Response('<h1>Something went wrong</h1>', {
status: 500,
headers: { 'content-type': 'text/html' },
});
}
}

This will only catch errors outside the shell that happened while generating the initial shell content, so it’s not exhaustive. If knowing whether an error occurred for some content is critical, you can move it up into the shell.


Handling different errors in different ways

You can create your own Error subclasses and use the instanceof operator to check which error is thrown. For example, you can define a custom NotFoundError and throw it from your component. Then you can save the error in onError and do something different before returning the response depending on the error type:

async function handler(request) {
let didError = false;
let caughtError = null;

function getStatusCode() {
if (didError) {
if (caughtError instanceof NotFoundError) {
return 404;
} else {
return 500;
}
} else {
return 200;
}
}

try {
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
didError = true;
caughtError = error;
console.error(error);
logServerCrashReport(error);
}
});
return new Response(stream, {
status: getStatusCode(),
headers: { 'content-type': 'text/html' },
});
} catch (error) {
return new Response('<h1>Something went wrong</h1>', {
status: getStatusCode(),
headers: { 'content-type': 'text/html' },
});
}
}

Keep in mind that once you emit the shell and start streaming, you can’t change the status code.


Waiting for all content to load for crawlers and static generation

Streaming offers a better user experience because the user can see the content as it becomes available.

However, when a crawler visits your page, or if you’re generating the pages at the build time, you might want to let all of the content load first and then produce the final HTML output instead of revealing it progressively.

You can wait for all the content to load by awaiting the stream.allReady Promise:

async function handler(request) {
try {
let didError = false;
const stream = await renderToReadableStream(<App />, {
bootstrapScripts: ['/main.js'],
onError(error) {
didError = true;
console.error(error);
logServerCrashReport(error);
}
});
let isCrawler = // ... depends on your bot detection strategy ...
if (isCrawler) {
await stream.allReady;
}
return new Response(stream, {
status: didError ? 500 : 200,
headers: { 'content-type': 'text/html' },
});
} catch (error) {
return new Response('<h1>Something went wrong</h1>', {
status: 500,
headers: { 'content-type': 'text/html' },
});
}
}

A regular visitor will get a stream of progressively loaded content. A crawler will receive the final HTML output after all the data loads. However, this also means that the crawler will have to wait for all data, some of which might be slow to load or error. Depending on your app, you could choose to send the shell to the crawlers too.


Aborting server rendering

You can force the server rendering to “give up” after a timeout:

async function handler(request) {
try {
const controller = new AbortController();
setTimeout(() => {
controller.abort();
}, 10000);

const stream = await renderToReadableStream(<App />, {
signal: controller.signal,
bootstrapScripts: ['/main.js'],
onError(error) {
didError = true;
console.error(error);
logServerCrashReport(error);
}
});
// ...

React will flush the remaining loading fallbacks as HTML, and will attempt to render the rest on the client.