2nd/10_Wiki/Topics/Programming & Language/메모리 누수(Memory Leaks).md

---
id: wiki-2026-0508-메모리-누수-memory-leaks
title: 메모리 누수(Memory Leaks)
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Memory Leak, Heap Leak, Leak Detection]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [memory, debugging, performance, gc, profiling]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: C/C++/Java/JS/Python
  framework: Valgrind/ASan/Heaptrack/Chrome DevTools
---

# 메모리 누수(Memory Leaks)

## 매 한 줄
> **"매 alive 인데 매 unused, 매 unused 인데 매 free X"**. 매 manual mem (C/C++) 의 free 누락, 매 GC mem (Java/JS) 의 unintended retention. 매 long-running process 의 매 가장 흔한 incident, 매 2026 도구는 매 ASan/heaptrack/V8 heap snapshot/JFR + AI-assisted root cause.

## 매 핵심

### 매 두 의미
- **매 Manual leak**: 매 free X — 매 OS 가 매 process 종료 시 매 회수.
- **매 GC leak**: 매 root 에서 매 reachable 이지만 매 logically dead — 매 unintended retention.

### 매 흔한 원인
1. 매 Cache 의 unbounded growth.
2. 매 Listener / observer 의 unsubscribe 누락.
3. 매 Closure 의 unintended capture (JS).
4. 매 Static collection (HashMap) 의 누적.
5. 매 ThreadLocal 의 cleanup 누락 (JVM/web).
6. 매 Native handle (file/socket) 의 close 누락.
7. 매 Circular ref + RC (Python before GC, Swift, ObjC ARC).

### 매 detection 의 axes
- 매 Symptom: 매 RSS 증가, 매 OOM, 매 GC 빈도 ↑, 매 latency drift.
- 매 Tool: 매 sampling profiler, 매 heap snapshot, 매 allocation tracking.

## 패턴

### C++ Valgrind
```bash
g++ -g -O0 main.cpp -o app
valgrind --leak-check=full --show-leak-kinds=all --track-origins=yes ./app
# 매 "definitely lost" 의 우선 fix
```

### C++ AddressSanitizer + LeakSanitizer
```bash
g++ -fsanitize=address,leak -g main.cpp -o app
ASAN_OPTIONS=detect_leaks=1 ./app
```

### C++ RAII (matter of design)
```cpp
class FileHandle {
    FILE* f;
public:
    explicit FileHandle(const char* p) : f(fopen(p, "r")) {}
    ~FileHandle() { if (f) fclose(f); }
    FileHandle(const FileHandle&) = delete;
    FileHandle& operator=(const FileHandle&) = delete;
};
// 매 std::unique_ptr / std::shared_ptr 가 매 default
```

### JS — listener 누락
```javascript
// 매 BAD — node 가 떠도 listener 가 retain
function attach(el) {
  const handler = () => doStuff(el);
  el.addEventListener("click", handler);
}

// 매 GOOD
function attach(el) {
  const handler = () => doStuff(el);
  el.addEventListener("click", handler);
  return () => el.removeEventListener("click", handler);
}
```

### JS — closure 의 unintended capture
```javascript
// 매 BAD — 매 huge 가 매 outer scope 에 retained
function setup() {
  const huge = new Array(1e7).fill(0);
  const small = huge[0];
  return () => small;
}

// 매 GOOD
function setup() {
  const small = (() => {
    const huge = new Array(1e7).fill(0);
    return huge[0];
  })();
  return () => small;
}
```

### Java — static map leak
```java
// 매 BAD
public class Cache {
    static final Map<Key, Value> M = new HashMap<>();
}

// 매 GOOD — Caffeine size cap + eviction
import com.github.benmanes.caffeine.cache.*;
Cache<Key, Value> c = Caffeine.newBuilder()
    .maximumSize(10_000).expireAfterWrite(Duration.ofMinutes(10)).build();
```

### Java heap dump + MAT
```bash
jcmd <pid> GC.heap_dump /tmp/heap.hprof
# 매 Eclipse MAT 에서 dominator tree / leak suspect
```

### Chrome DevTools — heap snapshot diff
```text
1. baseline snapshot
2. suspect action 반복 (e.g. open/close modal x10)
3. second snapshot
4. "Comparison" mode, 매 retained 증가 식별
5. retainer chain 의 root 추적
```

### Python tracemalloc
```python
import tracemalloc
tracemalloc.start(25)
snap1 = tracemalloc.take_snapshot()
# ... 매 action 반복 ...
snap2 = tracemalloc.take_snapshot()
for stat in snap2.compare_to(snap1, 'lineno')[:10]:
    print(stat)
```

### Linux heaptrack (C/C++ low-overhead)
```bash
heaptrack ./app
heaptrack_gui heaptrack.app.<pid>.gz
```

## 매 결정 기준
| Stack | Tool |
|---|---|
| C/C++ dev | ASan + LeakSanitizer |
| C/C++ prod | heaptrack / jemalloc profiler |
| Java | JFR + heap dump + MAT |
| Node.js | --inspect + Chrome heap snapshot |
| Browser | DevTools Memory tab |
| Python | tracemalloc / objgraph / memray |
| Go | pprof (heap profile) |

**기본값**: 매 first symptom = RSS chart, 매 second = heap snapshot diff, 매 third = retainer/dominator analysis.

## Graph
- 부모: [[Memory Management]] · [[Performance]]
- 변형: [[Manual Leak]] · [[GC Retention]] · [[Native Handle Leak]]
- 응용: [[Heap Profiling]] · [[Cache Eviction]] · [[WeakRef]]
- Adjacent: [[Garbage Collection]] · [[Reference Counting]] · [[RAII]] · [[OOM]]

## LLM 활용
**언제**: 매 leak 패턴 식별, 매 fix 제안 (RAII, weak ref, eviction policy), 매 heap dump 분석 hint.
**언제 X**: 매 production heap 의 직접 access, 매 sensitive customer data dump.

## 안티패턴
- **매 OOM 시 -Xmx ↑**: 매 leak 을 매 늦출 뿐.
- **매 try/finally 없는 close**: 매 try-with-resources / RAII / context manager 사용.
- **매 Strong cache without eviction**: Caffeine / LRU.
- **매 Listener pattern 의 no unsubscribe**: weakRef / dispose.
- **매 GC tuning 으로 leak 감추기**: root cause X.

## 검증 / 중복
- Verified: Valgrind manual, ASan docs, V8 blog, JFR docs, tracemalloc PEP, heaptrack docs.
- 신뢰도 A.

## Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — leak detection per-stack/anti-patterns/RAII 작성 |