Skip to main content
Version: 11.x

HTTP Batch Stream Link

unstable_httpBatchStreamLink is a terminating link that batches an array of individual tRPC operations into a single HTTP request that's sent to a single tRPC procedure (equivalent to httpBatchLink), but doesn't wait for all the responses of the batch to be ready and streams the responses as soon as any data is available.

info

We have prefixed this as unstable_ as it's a new API, but you're safe to use it! Read more.

Usage

All usage and options are identical to httpBatchLink.

note

If you require the ability to change/set response headers (which includes cookies) from within your procedures, make sure to use httpBatchLink instead! This is due to the fact that unstable_httpBatchStreamLink does not support setting headers once the stream has begun. Read more.

You can import and add the httpBatchStreamLink to the links array as such:

client/index.ts
ts
import { createTRPCClient, httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from '../server';
const client = createTRPCClient<AppRouter>({
links: [
httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});
client/index.ts
ts
import { createTRPCClient, httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from '../server';
const client = createTRPCClient<AppRouter>({
links: [
httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});

After that, you can make use of batching by setting all your procedures in a Promise.all. The code below will produce exactly one HTTP request and on the server exactly one database query:

ts
const somePosts = await Promise.all([
trpc.post.byId.query(1),
trpc.post.byId.query(2),
trpc.post.byId.query(3),
]);
ts
const somePosts = await Promise.all([
trpc.post.byId.query(1),
trpc.post.byId.query(2),
trpc.post.byId.query(3),
]);

Streaming mode

When batching requests together, the behavior of a regular httpBatchLink is to wait for all requests to finish before sending the response. If you want to send responses as soon as they are ready, you can use httpBatchStreamLink instead. This is useful for long-running requests.

client/index.ts
ts
import { createTRPCClient, unstable_httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from '../server';
const client = createTRPCClient<AppRouter>({
links: [
unstable_httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});
client/index.ts
ts
import { createTRPCClient, unstable_httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from '../server';
const client = createTRPCClient<AppRouter>({
links: [
unstable_httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});

Compared to a regular httpBatchLink, a unstable_httpBatchStreamLink will:

  • Cause the requests to be sent with a trpc-accept: application/jsonl header
  • Cause the response to be sent with a transfer-encoding: chunked and content-type: application/jsonl
  • Remove the data key from the argument object passed to responseMeta (because with a streamed response, the headers are sent before the data is available)

Async generators and deferred promises

You can make your queries and mutations to return async generators and promises using the iterablesAndDeferreds feature flag.

You can try this out on the homepage of tRPC.io: https://trpc.io/?try=minimal#try-it-out

ts
// @filename: trpc.ts
import { initTRPC } from '@trpc/server';
 
const t = initTRPC.create({
experimental: {
iterablesAndDeferreds: true,
},
});
 
export const router = t.router;
export const publicProcedure = t.procedure;
 
// @filename: server.ts
import { publicProcedure, router } from './trpc';
 
const appRouter = router({
examples: {
iterable: publicProcedure.query(async function* () {
for (let i = 0; i < 3; i++) {
await new Promise((resolve) => setTimeout(resolve, 500));
yield i;
}
}),
},
});
 
export type AppRouter = typeof appRouter;
 
 
// @filename: client.ts
import { createTRPCClient, unstable_httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from './server';
 
const trpc = createTRPCClient<AppRouter>({
links: [
unstable_httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});
const iterable = await trpc.examples.iterable.query();
const iterable: AsyncIterable<number>
 
for await (const value of iterable) {
console.log('Iterable:', value);
const value: number
}
ts
// @filename: trpc.ts
import { initTRPC } from '@trpc/server';
 
const t = initTRPC.create({
experimental: {
iterablesAndDeferreds: true,
},
});
 
export const router = t.router;
export const publicProcedure = t.procedure;
 
// @filename: server.ts
import { publicProcedure, router } from './trpc';
 
const appRouter = router({
examples: {
iterable: publicProcedure.query(async function* () {
for (let i = 0; i < 3; i++) {
await new Promise((resolve) => setTimeout(resolve, 500));
yield i;
}
}),
},
});
 
export type AppRouter = typeof appRouter;
 
 
// @filename: client.ts
import { createTRPCClient, unstable_httpBatchStreamLink } from '@trpc/client';
import type { AppRouter } from './server';
 
const trpc = createTRPCClient<AppRouter>({
links: [
unstable_httpBatchStreamLink({
url: 'http://localhost:3000',
}),
],
});
const iterable = await trpc.examples.iterable.query();
const iterable: AsyncIterable<number>
 
for await (const value of iterable) {
console.log('Iterable:', value);
const value: number
}

Compatibility (client-side)

Browsers

Browser support should be identical to fetch support.

Node.js / Deno

For runtimes other than the browser ones, the fetch implementation should support streaming, meaning that the response obtained by await fetch(...) should have a body property of type ReadableStream<Uint8Array> | NodeJS.ReadableStream, meaning that:

  • either response.body.getReader is a function that returns a ReadableStreamDefaultReader<Uint8Array> object
  • or response.body is a Uint8Array Buffer

This includes support for undici, node-fetch, native Node.js fetch implementation, and WebAPI fetch implementation (browsers).

React Native

Receiving the stream relies on the TextDecoder and TextDecoderStream APIs, which is not available in React Native. If you still want to enable streaming, you need to polyfill those.

Compatibility (server-side)

⚠️ for aws lambda, unstable_httpBatchStreamLink is not supported (will simply behave like a regular httpBatchLink). It should not break anything if enabled, but will not have any effect.

⚠️ for cloudflare workers, you need to enable the ReadableStream API through a feature flag: streams_enable_constructors

Reference

You can check out the source code for this link on GitHub.