Antigravity Agent
225 lines
7.2 KiB
Markdown
225 lines
7.2 KiB
Markdown
---
|
|
id: wiki-2026-0508-웹-워커-이벤트-포워딩-통신-지연-최소화-방법
|
|
title: 웹 워커 이벤트 포워딩 통신 지연 최소화 방법
|
|
category: 10_Wiki/Topics
|
|
status: verified
|
|
canonical_id: self
|
|
aliases: [Web Worker Event Forwarding, Worker Message Latency, postMessage Optimization]
|
|
duplicate_of: none
|
|
source_trust_level: A
|
|
confidence_score: 0.88
|
|
verification_status: applied
|
|
tags: [frontend, web-worker, performance, postmessage, transferable, comlink]
|
|
raw_sources: []
|
|
last_reinforced: 2026-05-10
|
|
github_commit: pending
|
|
tech_stack:
|
|
language: TypeScript
|
|
framework: Web Workers, Comlink, SharedArrayBuffer
|
|
---
|
|
|
|
# 웹 워커 이벤트 포워딩 통신 지연 최소화 방법
|
|
|
|
## 매 한 줄
|
|
> **"매 main thread ↔ Worker 통신 의 latency 의 4가지 의 source: serialization, structured clone, queue, dispatch — 매 Transferable / SharedArrayBuffer / batching / Comlink RPC 의 mitigation"**. 매 postMessage 의 default behavior 의 매 bytes 의 deep clone — 매 large payload 의 매 ms 의 단위. 매 2026 의 modern approach 의 OffscreenCanvas, transferable streams, Atomics.
|
|
|
|
## 매 핵심
|
|
|
|
### 매 Latency Source
|
|
1. **Structured clone** — 매 object 의 deep copy, 매 size 의 비례.
|
|
2. **Serialization** — 매 V8 의 internal format 의 conversion.
|
|
3. **Event queue** — 매 receiver 의 task queue 의 enqueue.
|
|
4. **Dispatch overhead** — 매 microtask boundary, 매 ~0.1-0.5ms.
|
|
|
|
### 매 Mitigation 전략
|
|
- **Transferable Objects** — `ArrayBuffer`, `MessagePort`, `OffscreenCanvas`, `ReadableStream` 의 zero-copy 의 ownership transfer.
|
|
- **SharedArrayBuffer + Atomics** — 매 shared memory, 매 lock-free 통신.
|
|
- **Batching** — 매 multiple event 의 single message 의 합침.
|
|
- **MessageChannel** — 매 dedicated channel, 매 main thread 의 bypass.
|
|
- **Comlink** — 매 RPC abstraction, 매 ergonomics.
|
|
|
|
### 매 Transferable Objects (2026)
|
|
- `ArrayBuffer`, `MessagePort`, `ImageBitmap`, `OffscreenCanvas`, `RTCDataChannel`.
|
|
- `ReadableStream`, `WritableStream`, `TransformStream` (Streams API).
|
|
- `VideoFrame`, `AudioData` (WebCodecs).
|
|
- 매 transfer 후 매 sender 의 access 의 X (neutered).
|
|
|
|
## 💻 패턴
|
|
|
|
### Pattern 1: Transferable ArrayBuffer (zero-copy)
|
|
```ts
|
|
// main.ts
|
|
const buffer = new ArrayBuffer(10_000_000); // 10MB
|
|
const view = new Float32Array(buffer);
|
|
view.fill(1.0);
|
|
|
|
worker.postMessage({ buffer }, [buffer]); // 매 두번째 인자 의 transfer list
|
|
console.log(buffer.byteLength); // 0 — 매 neutered
|
|
|
|
// worker.ts
|
|
self.onmessage = (e) => {
|
|
const buffer: ArrayBuffer = e.data.buffer;
|
|
// 매 zero-copy 의 ownership 의 받음
|
|
const view = new Float32Array(buffer);
|
|
// ... process
|
|
};
|
|
```
|
|
|
|
### Pattern 2: Comlink RPC (ergonomic)
|
|
```ts
|
|
// worker.ts
|
|
import * as Comlink from 'comlink';
|
|
|
|
const api = {
|
|
async heavyCompute(data: Float32Array): Promise<Float32Array> {
|
|
// ... compute
|
|
return result;
|
|
},
|
|
async streamResults(onChunk: (c: number[]) => void) {
|
|
for (let i = 0; i < 100; i++) {
|
|
onChunk([Math.random()]);
|
|
await new Promise(r => setTimeout(r, 10));
|
|
}
|
|
},
|
|
};
|
|
Comlink.expose(api);
|
|
|
|
// main.ts
|
|
import * as Comlink from 'comlink';
|
|
const worker = new Worker(new URL('./worker.ts', import.meta.url), { type: 'module' });
|
|
const api = Comlink.wrap<typeof api>(worker);
|
|
const result = await api.heavyCompute(input);
|
|
```
|
|
|
|
### Pattern 3: SharedArrayBuffer + Atomics (lock-free)
|
|
```ts
|
|
// 매 COOP/COEP header 의 필요 — Cross-Origin-Opener-Policy: same-origin
|
|
const sab = new SharedArrayBuffer(1024);
|
|
const view = new Int32Array(sab);
|
|
worker.postMessage({ sab });
|
|
|
|
// worker 의 spinlock 의 wait
|
|
Atomics.wait(view, 0, 0); // 매 view[0]==0 의 동안 sleep
|
|
const data = view[1];
|
|
|
|
// main 의 signal
|
|
view[1] = 42;
|
|
view[0] = 1;
|
|
Atomics.notify(view, 0, 1); // 매 worker 의 wake
|
|
```
|
|
|
|
### Pattern 4: Batching (event coalescing)
|
|
```ts
|
|
// main.ts
|
|
const queue: Event[] = [];
|
|
let scheduled = false;
|
|
|
|
function forward(event: Event) {
|
|
queue.push(event);
|
|
if (!scheduled) {
|
|
scheduled = true;
|
|
queueMicrotask(() => {
|
|
worker.postMessage({ events: queue.splice(0) });
|
|
scheduled = false;
|
|
});
|
|
}
|
|
}
|
|
|
|
window.addEventListener('mousemove', forward); // 매 60fps 의 single batch
|
|
```
|
|
|
|
### Pattern 5: OffscreenCanvas (rendering off main thread)
|
|
```ts
|
|
// main.ts
|
|
const canvas = document.querySelector('canvas')!;
|
|
const offscreen = canvas.transferControlToOffscreen();
|
|
worker.postMessage({ canvas: offscreen }, [offscreen]);
|
|
|
|
// worker.ts
|
|
self.onmessage = (e) => {
|
|
const canvas = e.data.canvas as OffscreenCanvas;
|
|
const ctx = canvas.getContext('2d')!;
|
|
// 매 main thread 의 block X 의 render
|
|
function frame() {
|
|
ctx.clearRect(0, 0, canvas.width, canvas.height);
|
|
// ... draw
|
|
requestAnimationFrame(frame);
|
|
}
|
|
frame();
|
|
};
|
|
```
|
|
|
|
### Pattern 6: MessageChannel (direct port)
|
|
```ts
|
|
const channel = new MessageChannel();
|
|
worker1.postMessage({ port: channel.port1 }, [channel.port1]);
|
|
worker2.postMessage({ port: channel.port2 }, [channel.port2]);
|
|
// 매 worker 끼리 매 main thread 의 bypass 의 직접 통신
|
|
```
|
|
|
|
### Pattern 7: Transferable Stream (2026)
|
|
```ts
|
|
// main.ts
|
|
const stream = new ReadableStream({
|
|
start(ctrl) {
|
|
fetch('/large').then(async r => {
|
|
const reader = r.body!.getReader();
|
|
while (true) {
|
|
const { value, done } = await reader.read();
|
|
if (done) break;
|
|
ctrl.enqueue(value);
|
|
}
|
|
ctrl.close();
|
|
});
|
|
}
|
|
});
|
|
worker.postMessage({ stream }, [stream]); // 매 zero-copy
|
|
|
|
// worker.ts
|
|
self.onmessage = async (e) => {
|
|
for await (const chunk of e.data.stream) {
|
|
process(chunk);
|
|
}
|
|
};
|
|
```
|
|
|
|
## 매 결정 기준
|
|
| 상황 | 기법 |
|
|
|---|---|
|
|
| Large binary payload | Transferable ArrayBuffer |
|
|
| RPC ergonomics | Comlink |
|
|
| Lock-free shared state | SharedArrayBuffer + Atomics |
|
|
| High-frequency events | Batching with `queueMicrotask` |
|
|
| Canvas rendering | OffscreenCanvas |
|
|
| Worker-to-worker | MessageChannel |
|
|
| Streaming data | Transferable streams |
|
|
|
|
**기본값**: Transferable ArrayBuffer + Comlink RPC + 매 60fps event 의 batching.
|
|
|
|
## 🔗 Graph
|
|
- 부모: [[Web Workers]] · [[Frontend Performance Optimization (FE 성능 최적화)]]
|
|
- 변형: [[Service Worker]] · [[SharedWorker]] · [[Worklet]]
|
|
- 응용: [[웹 워커 (Web Workers)와 PostMessage]] · [[병렬 처리 (Parallel Processing)]]
|
|
- Adjacent: [[Event Loop]] · [[Structured Clone]] · [[Comlink]]
|
|
|
|
## 🤖 LLM 활용
|
|
**언제**: postMessage 의 large payload 의 transferable 의 변환 권고, COOP/COEP header 의 troubleshoot, batching 의 적절한 frequency 의 제안.
|
|
**언제 X**: 매 native thread (C++ worker) 의 통신 — 매 다른 domain.
|
|
|
|
## ❌ 안티패턴
|
|
- **Large object 의 매 postMessage**: 매 deep clone, 매 GC pressure, 매 main thread 의 stall.
|
|
- **Per-event 매 postMessage** (mousemove): 매 queue 의 폭주, 매 jank.
|
|
- **Atomics 의 SharedArrayBuffer 없음**: 매 cross-origin isolation 의 필요.
|
|
- **Worker 의 dynamic import 의 누락**: 매 type: 'module' 의 누락.
|
|
- **Transferable 의 후 의 access**: 매 neutered, 매 byteLength=0.
|
|
|
|
## 🧪 검증 / 중복
|
|
- Verified (HTML spec: Workers, MDN Transferable Objects, Comlink GitHub 2026).
|
|
- 신뢰도 A.
|
|
|
|
## 🕓 Changelog
|
|
| 날짜 | 변경 |
|
|
|---|---|
|
|
| 2026-05-08 | Phase 1 |
|
|
| 2026-05-10 | Manual cleanup — Transferables, Comlink RPC, SAB+Atomics, batching patterns |
|