2nd/10_Wiki/Topics/Architecture/Memory_Leaks.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-memory-leaks	Memory Leaks	10_Wiki/Topics	verified	self	메모리 누수 Mem Leak Heap Leak	none	A	0.9	applied	memory debugging performance gc		2026-05-10	pending	language framework multi node-jvm-python

Memory Leaks

매 한 줄

"매 reachable 한 garbage — 매 free 되지 않는 retained memory.". Memory leak 은 매 program 이 더 이상 필요 없는 memory 를 release 하지 못해서 매 heap 이 무한히 grow 하는 현상. 매 GC language (JS, Python, JVM) 에서도 매 reference 가 살아있으면 leak — 매 manual language (C/C++) 에서는 매 free() miss.

매 핵심

매 분류

• Unreachable leak (C/C++): 매 free() 누락 — pointer 잃어버림.
• Reachable leak (GC lang): 매 reference 가 살아있어서 GC 가 collect 안 함.
• Logical leak: 매 코드는 정상이지만 매 cache/list 가 무한 grow.

매 흔한 원인 (GC lang)

• Global variable: 매 process 끝까지 살아있음.
• Closure capture: 매 callback 이 매 큰 object 를 capture.
• Event listener: 매 removeEventListener 누락.
• Timer: 매 clearInterval 누락.
• DOM detach: 매 detached node 를 JS 가 still reference.
• Cache: 매 unbounded Map/dict.

매 진단 신호

1. RSS 가 매 시간/일 단위로 매 monotonically grow.
2. GC pause 가 매 frequent + long.
3. OOM crash 결국.

💻 패턴

Node.js heap snapshot diff

// Take 3 snapshots, diff 2 vs 3
const v8 = require('v8');
const fs = require('fs');

function snapshot(label) {
  const stream = v8.getHeapSnapshot();
  const file = fs.createWriteStream(`heap-${label}-${Date.now()}.heapsnapshot`);
  stream.pipe(file);
}

// Or via Chrome DevTools: chrome://inspect → Memory tab → Take snapshot
// Compare snapshots → "Allocations between snapshots" → find growing constructors

매 흔한 leak: closure capture

// BAD: closure captures huge `data` even though only `id` is used
function attach(data) {
  document.getElementById('btn').addEventListener('click', () => {
    console.log(data.id);  // captures entire `data` (10MB)
  });
}

// GOOD: extract only what's needed
function attach(data) {
  const id = data.id;  // only id is captured
  document.getElementById('btn').addEventListener('click', () => {
    console.log(id);
  });
}

매 EventEmitter leak

const emitter = new EventEmitter();
emitter.setMaxListeners(20);  // Node warns at >10

// BAD: each call adds a new listener
function onEachRequest(req) {
  emitter.on('data', () => process(req));  // listener never removed
}

// GOOD: once() or remove
function onEachRequest(req) {
  const handler = () => process(req);
  emitter.once('data', handler);
  // or: emitter.off('data', handler) when done
}

Python tracemalloc

import tracemalloc
tracemalloc.start(25)  # store 25 frames

# ... run workload ...

snap1 = tracemalloc.take_snapshot()
# ... run more workload ...
snap2 = tracemalloc.take_snapshot()

stats = snap2.compare_to(snap1, 'lineno')
for stat in stats[:10]:
    print(stat)
# → shows top 10 lines that allocated most new memory between snapshots

JVM heap dump

# Live process
jcmd <pid> GC.heap_dump /tmp/heap.hprof

# On OOM
java -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/heap.hprof MyApp

# Analyze with Eclipse MAT or VisualVM
# → "Leak Suspects Report" auto-identifies likely retainers

매 detached DOM leak

// BAD: cache holds reference to detached node
const cache = new Map();
function showModal(id) {
  const node = document.createElement('div');
  cache.set(id, node);  // never deleted
  document.body.appendChild(node);
  // later: document.body.removeChild(node) — but cache still holds it
}

// GOOD: WeakMap or explicit cleanup
const cache = new WeakMap();  // doesn't prevent GC
// or: cache.delete(id) when modal closes

매 unbounded cache

// BAD: grows forever
const cache = {};
function memoize(key, fn) {
  if (!(key in cache)) cache[key] = fn();
  return cache[key];
}

// GOOD: LRU with max size
import LRU from 'lru-cache';
const cache = new LRU({ max: 1000, ttl: 1000 * 60 * 5 });

매 결정 기준

증상	진단 도구
Node RSS 증가	Chrome DevTools heap snapshot diff
Python 증가	tracemalloc compare_to
JVM OOM	jcmd heap dump → Eclipse MAT
C/C++	Valgrind, AddressSanitizer (-fsanitize=address)
Browser tab	DevTools Memory → Allocation timeline

기본값: 매 3 snapshot diff — between 2nd & 3rd 사이의 retained 만 보면 매 startup noise 제거됨.

🔗 Graph

• 부모: Memory Management · Garbage Collection
• 변형: Mark-Sweep · Reference Counting
• 응용: Performance_Profiling_and_Memory
• Adjacent: WeakRef

🤖 LLM 활용

언제: long-running service (Node server, Python daemon, JVM app), browser SPA 가 시간 흐를수록 느려질 때. 언제 X: short-lived script — process exit 시 OS 가 다 회수.

❌ 안티패턴

• "GC lang 은 leak 없다": 매 reachable leak 이 매 가장 흔함.
• 단일 snapshot: 매 baseline 없으면 무엇이 leak 인지 모름 — 매 diff 가 정답.
• Premature optimization: 매 RSS 가 stable 한데 매 leak 추적 — 매 시간 낭비.
• try/catch 없는 cleanup: 매 error path 에서 listener removal 누락.

🧪 검증 / 중복

• Verified (V8 docs, Chrome DevTools docs, MDN, Node.js docs, Eclipse MAT).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — leak 분류 + Node/Python/JVM 진단 패턴