Reopen of: Problems with cooperation between http and pipeline/"for await"

Version

v20.5.1

Platform

Darwin Kernel Version 22.6.0

Subsystem

http

What steps will reproduce the bug?

https://github.com/nodejs/node/issues/38262
https://github.com/nodejs/node/issues/38262#issuecomment-991233190

This often stops after the first iteration:

let http = require("http");

let server = http.createServer(async function(req, res)
{
	try
	{
		let len = 0;
		for await (let chunk of req)
		{
			len += chunk.length;
			if(len > 2) throw "payload too large";
		}
		res.end("ok");
	}
	catch(err)
	{
		console.log("server log:", err);
		res.end(err);
	}
});

(async function()
{
	await new Promise(resolve => server.listen(8888, resolve));
	let agent = new http.Agent({ keepAlive: true });
	let data = Buffer.alloc(1000000);
	for(let i = 0; ; i++)
	{
		console.log(i);
		let req = http.request({ port: 8888, method: "post", agent }).end(data);
		let res = await new Promise((resolve, reject) => req.on("response", resolve).on("error", reject));
		let str = "";
		for await (let chunk of res) str += chunk;
		console.log(str);
	}
}());

How often does it reproduce? Is there a required condition?

Often.

What is the expected behavior? Why is that the expected behavior?

It should continue.

What do you see instead?

0
server log: payload too large
payload too large
1
^C

Comienza ejecutando la reproducción proporcionada con la versión de Node.js indicada y compara su comportamiento con el issue #38262 y el comentario enlazado. Traza el manejo de los iteradores asíncronos de solicitud/respuesta HTTP en torno a for await, el procesamiento de solicitudes rechazadas y el keep-alive agent. Se considera terminado cuando el bucle repetido continúa de forma fiable después de la carga útil sobredimensionada, en lugar de detenerse tras la primera iteración.