2nd/10_Wiki/Topics/Programming & Language/스트림Stream.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-스트림-stream	스트림(Stream)	10_Wiki/Topics	verified	self	stream streaming	none	A	0.95	applied	stream async io		2026-05-10	pending	language framework typescript nodejs

스트림(Stream)

매 한 줄

"매 데이터를 chunk 단위로 점진 전달한다". 매 entire payload 를 buffer 에 fully load 하지 않고, 매 사용 가능한 시점마다 chunk 를 yield → 매 memory footprint 감소 + first-byte latency 개선. 매 modern web (Server-Sent Events · LLM token streaming · HTTP/2 push) 의 backbone.

매 핵심

매 Stream 의 종류

• Readable: pull-based source (file read, HTTP response).
• Writable: push-based sink (file write, response.write).
• Duplex: read + write (TCP socket).
• Transform: read → mutate → write (gzip, JSON.parse).

매 동작 모델

• Chunk + backpressure: consumer slow → producer pause (highWaterMark).
• Object mode: chunk 가 raw bytes 가 아닌 object.
• Pipe: source.pipe(transform).pipe(sink) chain.

매 응용

1. LLM token streaming (Claude · GPT-5 의 SSE response).
2. Large file upload/download (Multipart, S3 multipart).
3. Real-time analytics (Kafka consumer).

💻 패턴

Web Streams API (cross-platform)

const response = await fetch("/api/llm");
const reader = response.body!.pipeThrough(new TextDecoderStream()).getReader();
while (true) {
  const { value, done } = await reader.read();
  if (done) break;
  process.stdout.write(value);
}

Async iteration over Node stream

import { createReadStream } from "node:fs";
const stream = createReadStream("big.log", { encoding: "utf-8" });
for await (const chunk of stream) {
  console.log(`got ${chunk.length} chars`);
}

Transform stream — gzip on the fly

import { pipeline } from "node:stream/promises";
import { createReadStream, createWriteStream } from "node:fs";
import { createGzip } from "node:zlib";
await pipeline(createReadStream("input.txt"), createGzip(), createWriteStream("output.gz"));

Server-Sent Events (LLM token stream)

// Hono / Bun
import { streamSSE } from "hono/streaming";
app.get("/chat", c => streamSSE(c, async stream => {
  for await (const tok of llm.generate(prompt)) {
    await stream.writeSSE({ data: tok });
  }
}));

ReadableStream from generator

function makeStream<T>(gen: AsyncGenerator<T>) {
  return new ReadableStream<T>({
    async pull(controller) {
      const { value, done } = await gen.next();
      done ? controller.close() : controller.enqueue(value);
    }
  });
}

Backpressure handling

const writable = createWriteStream("out.bin");
function writeChunk(buf: Buffer) {
  if (!writable.write(buf)) {
    return new Promise<void>(res => writable.once("drain", res));
  }
  return Promise.resolve();
}

매 결정 기준

상황	Approach
Browser fetch streaming	Web Streams API (response.body)
Node.js file/network	node:stream + async iter
LLM token stream	SSE or chunked transfer
Massive object pipeline	object mode + Transform
Cross-runtime (Bun/Deno)	Web Streams (standard)

기본값: Web Streams API + async iteration.

🔗 Graph

• 부모: Nodejs 성능 최적화 및 디버깅
• Adjacent: API 응답 및 에러 핸들링 아키텍처
• 응용: Server Architecture

🤖 LLM 활용

언제: streaming chat UI · large file async processing · backpressure debug. 언제 X: small payload (<1MB) — buffering 이 simpler.

❌ 안티패턴

• Concat all chunks before yielding: stream 의 의미 무효화.
• Ignore backpressure: producer 가 consumer 추월 → memory blow up.
• No error propagation in pipe: 한 stage error 가 silent.
• Sync transform on huge object: event loop block.

🧪 검증 / 중복

• Verified (Node.js docs · WHATWG Streams Standard).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Stream API 패턴 정리