perf: reduce overhead

[tsctx]

Hi @yusukebe,
I saw your post on X (Twiiter) and tried to optimize it.
For performance improvement:

1. use writeHead instead of setHeader.
2. Quit changing ReadableStream to Readable. (Do not call Readable.fromWeb)
3. Do not use Readable.pipeline.
4. use Response#arrayBuffer instead of Response#text and Buffer.byteLength
5. There is no need to use Headers#getSetCookie since it is already split.

In other words, this change reduces the overhead of the handle portion of the body.

• main
  incomingMessage => Request => handling => Response => ReadableStream => Readable => outgoingMessage
• this PR
  incomingMessage => Request => handling => Response => ReadableStream => outgoingMessage

[tsctx]

@yusukebe
If you don't mind, could you measure the benchmark?

[yusukebe]

Hi @tsctx!

Super cool! The five changes you made all seem good. The tests passed, and some real-world examples with this version, including streaming responses, work fine.

The benchmark I'm taking is super simple, just a "Hello World" application:

import { serve } from '@hono/node-server'
import { Hono } from 'hono'

const app = new Hono()
app.get('/', (c) => c.text('Hi'))

serve(app)

And I access it with Bombardier using this command:

bombardier -d 10s --fasthttp http://localhost:3000

The results:

It really has become faster! In this case, it does not use streaming, so we may have to test another pattern, but currently, this simple pattern is a good reference.

I'd like to merge it into the "main". Do you have any more changes?

[tsctx]

Hi @tsctx!

Super cool! The five changes you made all seem good. The tests passed, and some real-world examples with this version, including streaming responses, work fine.

The benchmark I'm taking is super simple, just a "Hello World" application:

import { serve } from '@hono/node-server'
import { Hono } from 'hono'

const app = new Hono()
app.get('/', (c) => c.text('Hi'))

serve(app)

And I access it with Bombardier using this command:

bombardier -d 10s --fasthttp http://localhost:3000

The results:

It really has become faster! In this case, it does not use streaming, so we may have to test another pattern, but currently, this simple pattern is a good reference.

I'd like to merge it into the "main". Do you have any more changes?

You may merge them.
If you benchmark your storming, please let me know the results!

[yusukebe]

Thanks! Okay, let's ship it.

If you have any ideas for improvement, we'd be glad if you created an issue or PR!