Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 22 additions & 7 deletions components/mdx.tsx
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
// Components the MDX pages may use besides the Fumadocs ones.
import type { ReactNode } from "react";
import { run, scenarios, servers } from "@/data/bench";
import { run, scenarios, servers, widerRun } from "@/data/bench";

/**
* Two code blocks next to each other on a wide screen, one under the other on a narrow one:
Expand All @@ -12,9 +12,14 @@ export function Side({ children }: { children: ReactNode }) {

const REPORTS = "https://github.com/inkanjs/inkan/tree/main/bench/reports";

/** The numbers of data/bench.ts for the servers named, so the page never drifts from /benchmarks. */
export function BenchTable({ names }: { names: string[] }) {
const rows = names.map((name) => servers.find((s) => s.name === name)).filter((s) => s !== undefined);
/**
* The numbers of data/bench.ts for the servers named, so the page never drifts from /benchmarks.
* `wide` takes them from the run with every framework instead of the release bench.
*/
export function BenchTable({ names, wide = false }: { names: string[]; wide?: boolean }) {
const from = wide ? widerRun.servers : servers;
const meta = wide ? widerRun.run : run;
const rows = names.map((name) => from.find((s) => s.name === name)).filter((s) => s !== undefined);
return (
<>
<div className="not-prose my-6 overflow-x-auto rounded-xl border border-fd-border">
Expand Down Expand Up @@ -46,9 +51,19 @@ export function BenchTable({ names }: { names: string[] }) {
</table>
</div>
<p className="text-sm text-fd-muted-foreground">
Requests per second next to bare <code>node:http</code> as 100, on Node {run.node}, a {run.machine}, {run.tool},{" "}
{run.rounds} rounds of {run.seconds} seconds, measured {run.date}. The full runs, with the method and the raw numbers, are
in <a href={REPORTS}>bench/reports</a>; every scenario is on <a href="/benchmarks">/benchmarks</a>.
Requests per second next to bare <code>node:http</code> as 100, on Node {meta.node}, a {meta.machine}, {meta.tool},{" "}
{meta.rounds} rounds of {meta.seconds} seconds, measured {meta.date}
{wide ? ", all ten in one run" : ""}. The full runs, with the method and the raw numbers, are in{" "}
<a href={REPORTS}>bench/reports</a>
{wide ? (
<>
; this one is <a href={`${REPORTS}/frameworks-2026-11.md`}>frameworks-2026-11.md</a>.
</>
) : (
<>
; every scenario is on <a href="/benchmarks">/benchmarks</a>.
</>
)}
</p>
</>
);
Expand Down
82 changes: 78 additions & 4 deletions content/docs/compare.mdx
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
---
title: "inkan, Express, Fastify, Hono"
description: "What each of the four is best at, when I would pick which, and how fast they are on the same bench."
title: "inkan next to the other Node frameworks"
description: "Express, Fastify, Hono, Koa, hapi, h3, Elysia and NestJS: what each is best at, when I would pick which, and how fast they are on the same bench."
---

I write inkan, so take this page with that in mind. I have tried to put down where each of
the other three is the better choice, because for a lot of projects one of them is. If
the others is the better choice, because for a lot of projects one of them is. If
something here is wrong or out of date, [open an issue](https://github.com/inkanjs/inkan/issues)
and I will fix the page.

Expand All @@ -18,7 +18,7 @@ a rejected promise from an async handler to the error middleware, so the oldest
What it leaves to you is everything about the shape of the data. Validation, the types of
`req.body`, an OpenAPI document and the docs are separate tools (zod or express-validator,
swagger-jsdoc or a hand-written spec), and nothing makes them agree with the handler. It is
also the slowest of the four by a wide margin on the bench below, mostly in routing: with 400
also the slowest of Express, Fastify, Hono and inkan by a wide margin on the bench below, mostly in routing: with 400
routes or a 404 it does about a fifth of what bare `node:http` does.

**Pick Express when** the team knows it well, you depend on middleware that only exists for
Expand Down Expand Up @@ -58,6 +58,72 @@ is at home, and these numbers say nothing about that.
**Pick Hono when** your API runs at the edge or on Workers, or cold start and bundle size
matter more to you than anything on this page.

## Koa

Koa is Express done again by the same people, around `async`/`await`: one context object
instead of a request and a response, and middleware that wraps the rest of the chain, so it
can act both before and after the handler. The core is tiny and leaves almost everything to
packages: routing (`@koa/router`), body parsing, validation. Code written for it tends to be
clean and easy to follow.

On the bench below it comes last, and that needs a word. `@koa/router` tests each request
against every route, and the bench app has 800 of them. With two routes Koa is level with
Express. So the number is fair for a big Koa app on the standard router, not for Koa itself.

**Pick Koa when** you like the onion model of middleware, your app has tens of routes rather
than hundreds, and you want to choose every piece yourself.

## hapi

hapi is built for large teams and long-lived services. Routes are configuration objects
with validation (Joi), auth strategies, caching and lifecycle extensions next to the handler,
and plugins have explicit dependencies and their own config. It came out of Walmart, where it
carried Black Friday traffic, and it puts a lot of care into security defaults.

It is more ceremony than the others for a small API, and its ecosystem is mostly its own
(`@hapi/*`) rather than the wider middleware world. On the bench it is a little ahead of
Express overall, and clearly ahead on routing.

**Pick hapi when** a large team needs strong conventions, auth and validation configured per
route, and you value stable, careful releases over speed.

## h3

h3 is the server under Nitro and Nuxt. Like Hono it is built on web-standard requests and
responses and runs on Node, Bun, Deno, Workers and more, with small composable helpers
(`readBody`, `getQuery`) instead of a big framework object. On the bench it is the fastest of
the newcomers and not far behind Hono, strong on large bodies and answers.

**Pick h3 when** you are in the Nuxt or Nitro world, or want a small, standards-based server
that runs on several runtimes, and you bring your own validation and docs.

## Elysia

Elysia is made for Bun, and its idea is close to inkan's: a schema per route (TypeBox) gives
validation, the handler's types, an OpenAPI document and an end-to-end typed client (Eden).
It is fast and pleasant to write, and on Bun it is among the fastest servers in JavaScript.

On Node it runs through `@elysiajs/node`, and the bench only measures that. It does well on
answers without input, and loses about half wherever its schemas read params or a body,
because each request first goes through the adapter. Measure it on Bun before you judge it.

**Pick Elysia when** you run on Bun and want contracts, types and a typed client from one
place. On Node, inkan or Fastify give you the same idea for less.

## NestJS

NestJS is a full application framework: modules, dependency injection, decorators,
guards, pipes and interceptors, in the style of Angular or Spring. It brings an answer for
almost everything (OpenAPI, GraphQL, queues, microservices, testing) and a structure that
many developers already know, which is a lot when a big team builds a big app.

It sits on Express or Fastify and adds its own layer, so it is never faster than the one
beneath it. On its default Express platform it lands a little below Express on the bench.
Validation is usually `class-validator`, which the bench does not use; there it checks by hand.

**Pick NestJS when** the app is large, the team wants one structure and DI everywhere, and
the speed of a single request matters less than how the codebase holds together.

## inkan

inkan starts from one contract per route: params, query, headers, body, the answers per
Expand Down Expand Up @@ -90,6 +156,14 @@ schema check does not have to cost much: the two frameworks that validate everyt
two fastest. It measures one small route at a time on one machine; your database will
usually matter more than any of these differences.

All ten in one run, Koa, hapi, h3, Elysia and NestJS next to the ones above (Elysia checks
with its own schemas, the other four by hand):

<BenchTable wide names={["node:http, by hand", "Fastify", "inkan", "Hono", "h3", "Elysia", "hapi", "Express", "NestJS", "Koa"]} />

This is one run, not the median of three, so inkan and Fastify only differ by noise here;
the table above is the steadier one for them. Elysia is measured on Node, not on Bun.

## Moving over

If you have an app on one of the others, [from Express](/docs/migrate/express) and
Expand Down
2 changes: 1 addition & 1 deletion content/docs/index.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -79,7 +79,7 @@ server to what its docs say.
| Runtime dependencies | several | several | none | **none** |

That table is about features. Where each of the others is the better choice, and the
benchmark numbers next to each other, are in [inkan, Express, Fastify, Hono](/docs/compare).
benchmark numbers next to each other, are in [inkan next to the other Node frameworks](/docs/compare).
Coming from one of them: [from Express](/docs/migrate/express), [from Fastify](/docs/migrate/fastify).


Expand Down
18 changes: 18 additions & 0 deletions data/bench.ts
Original file line number Diff line number Diff line change
Expand Up @@ -47,3 +47,21 @@ export const servers: Server[] = [
{ name: "Hono", what: "checks by hand", score: 85.6, values: { hello: 94, params: 77, body: 93, answer: 98, router: 77, notFound: 85, invalid: 87 } },
{ name: "Express", what: "checks by hand", score: 48.9, values: { hello: 53, params: 52, body: 69, answer: 73, router: 23, notFound: 23, invalid: 53 } },
];

// One run with ten frameworks at once (bench/reports/frameworks-2026-11.md). Only the compare
// page shows it, next to the release numbers above, never mixed into them.
export const widerRun: { run: typeof run; servers: Server[] } = {
run: { ...run, date: "2026-11-13" },
servers: [
{ name: "node:http, by hand", what: "the baseline", base: true, score: 100, values: { hello: 100, params: 100, body: 100, answer: 100, router: 100, notFound: 100, invalid: 100 } },
{ name: "Fastify", what: "schemas, like inkan", score: 88.8, values: { hello: 88, params: 96, body: 95, answer: 80, router: 91, notFound: 83, invalid: 73 } },
{ name: "inkan", what: "on node:http", ours: true, score: 88.5, values: { hello: 89, params: 90, body: 89, answer: 88, router: 93, notFound: 88, invalid: 78 } },
{ name: "Hono", what: "checks by hand", score: 81.0, values: { hello: 85, params: 71, body: 88, answer: 97, router: 72, notFound: 71, invalid: 84 } },
{ name: "h3", what: "checks by hand", score: 75.8, values: { hello: 75, params: 72, body: 88, answer: 95, router: 72, notFound: 60, invalid: 80 } },
{ name: "Elysia", what: "on Node, its own schemas", score: 54.7, values: { hello: 76, params: 39, body: 41, answer: 93, router: 43, notFound: 59, invalid: 22 } },
{ name: "hapi", what: "checks by hand", score: 46.5, values: { hello: 45, params: 40, body: 52, answer: 76, router: 44, notFound: 45, invalid: 38 } },
{ name: "Express", what: "checks by hand", score: 40.1, values: { hello: 43, params: 39, body: 59, answer: 68, router: 18, notFound: 16, invalid: 43 } },
{ name: "NestJS", what: "on Express", score: 34.4, values: { hello: 36, params: 34, body: 54, answer: 63, router: 17, notFound: 14, invalid: 30 } },
{ name: "Koa", what: "with @koa/router", score: 25.2, values: { hello: 20, params: 20, body: 38, answer: 51, router: 22, notFound: 20, invalid: 24 } },
],
};
Loading