Files

T

Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization

10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-20 23:52:15 +09:00

6.6 KiB

Raw Blame History

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

title

Distributed Systems Fallacies

매 한 줄

"매 network 의 invisible 한 assumption 의 매 production failure 의 source". 1994년 Peter Deutsch (Sun) 가 매 7 fallacies 의 articulate, 1997년 James Gosling 가 8th 의 add. 매 2026 cloud-native 시대 에도 매 microservices / serverless / edge compute 의 매 매 valid.

매 핵심

매 8 Fallacies

Network 의 reliable: 매 packet drop / partition / DNS failure 의 inevitable.
Latency 의 zero: 매 LAN ~0.5ms, 매 cross-region ~150ms, 매 satellite ~600ms.
Bandwidth 의 infinite: 매 video / ML model weights / log shipping 의 saturate.
Network 의 secure: 매 default 의 insecure — 매 zero-trust assume.
Topology 의 안 변함: 매 autoscaling / k8s pod reschedule / failover 의 매 second.
Administrator 의 single: 매 multi-cloud / multi-region 의 매 다른 policy.
Transport cost 의 zero: 매 serialization / TLS handshake / egress fee 의 real.
Network 의 homogeneous: 매 IPv4/IPv6, 매 protocol versions, 매 MTU mismatch.

매 왜 fallacy 인가

매 dev 의 localhost / monolith mental model 의 distributed 에 적용 시 fail.
매 "happy path" coding 의 매 timeout / retry / circuit breaker 의 lack.
매 50ms RTT 의 매 100 calls 의 5 second user-facing latency.

매 응용

Microservices design — 매 call graph 의 latency budget 산정.
Cross-region replication — 매 split-brain / eventual consistency 의 plan.
Edge computing — 매 intermittent connectivity 의 first-class.

💻 패턴

Pattern 1: Timeout + Retry with Exponential Backoff

async function callWithRetry<T>(
  fn: () => Promise<T>,
  opts = { maxRetries: 3, baseMs: 100, timeoutMs: 2000 }
): Promise<T> {
  for (let attempt = 0; attempt <= opts.maxRetries; attempt++) {
    try {
      return await Promise.race([
        fn(),
        new Promise<never>((_, reject) =>
          setTimeout(() => reject(new Error("timeout")), opts.timeoutMs)
        ),
      ]);
    } catch (err) {
      if (attempt === opts.maxRetries) throw err;
      const jitter = Math.random() * opts.baseMs;
      await new Promise(r => setTimeout(r, opts.baseMs * 2 ** attempt + jitter));
    }
  }
  throw new Error("unreachable");
}

Pattern 2: Circuit Breaker (Resilience4j-style)

class CircuitBreaker {
  private failures = 0;
  private state: "closed" | "open" | "half-open" = "closed";
  private openedAt = 0;
  constructor(private threshold = 5, private cooldownMs = 30_000) {}

  async exec<T>(fn: () => Promise<T>): Promise<T> {
    if (this.state === "open" && Date.now() - this.openedAt < this.cooldownMs)
      throw new Error("circuit open");
    if (this.state === "open") this.state = "half-open";
    try {
      const r = await fn();
      this.failures = 0; this.state = "closed";
      return r;
    } catch (e) {
      if (++this.failures >= this.threshold) {
        this.state = "open"; this.openedAt = Date.now();
      }
      throw e;
    }
  }
}

Pattern 3: Bulkhead (concurrency limiter)

import pLimit from "p-limit";
const dbLimit = pLimit(20);   // 매 DB pool 의 isolate
const apiLimit = pLimit(50);  // 매 external API 의 separate

async function getUser(id: string) {
  return dbLimit(() => db.users.findOne({ id }));
}

Pattern 4: Idempotency Key

async function chargeCard(req: ChargeReq, idemKey: string) {
  const cached = await redis.get(`idem:${idemKey}`);
  if (cached) return JSON.parse(cached);
  const result = await stripe.charges.create(req);
  await redis.setex(`idem:${idemKey}`, 86400, JSON.stringify(result));
  return result;
}

Pattern 5: Latency Budget

// 매 user-facing 200ms 의 budget
// API gateway:    20ms
// auth check:     10ms (cached)
// service call:   50ms (timeout 100ms)
// DB query:       30ms (timeout 80ms)
// serialization:  10ms
// buffer:         80ms
// 매 each hop 의 explicit budget — over 시 fail fast.

Pattern 6: Chaos Testing (Toxiproxy)

# 매 latency injection
toxiproxy-cli toxic add api -t latency -a latency=500 -a jitter=100
# 매 packet loss
toxiproxy-cli toxic add db -t timeout -a timeout=2000

Pattern 7: Health check with deep probe

app.get("/health/deep", async (_, res) => {
  const checks = await Promise.allSettled([
    db.raw("SELECT 1").then(() => ({ db: "ok" })),
    redis.ping().then(() => ({ redis: "ok" })),
    fetch(upstream + "/health", { signal: AbortSignal.timeout(500) }),
  ]);
  const failed = checks.filter(c => c.status === "rejected");
  res.status(failed.length ? 503 : 200).json({ checks });
});

매 결정 기준

상황	Approach
LAN microservice call	timeout 1-2s, retry 2x
Cross-region call	timeout 5-10s, circuit breaker
3rd-party API	bulkhead + circuit breaker + idempotency
Streaming / WebSocket	heartbeat + auto-reconnect
Critical write	idempotency key 의 mandatory

기본값: 매 every remote call 의 timeout + retry + circuit breaker 의 wrap.

🔗 Graph

부모: Distributed Systems · Software Architecture
변형: CAP Theorem · PACELC
응용: Microservices Architecture · Service Mesh
Adjacent: Resilience Patterns · Chaos Engineering

🤖 LLM 활용

언제: 매 distributed system design review, 매 incident postmortem, 매 SLO 산정. 언제 X: 매 single-process monolith, 매 batch job 의 isolated.

❌ 안티패턴

Infinite retry: 매 retry storm — 매 backoff + max attempts 의 mandatory.
Timeout 의 unset: 매 default infinite — 매 thread pool exhaustion.
Synchronous fan-out: 매 N services 의 sequential await — 매 N×latency.
Trust LAN security: 매 zero-trust / mTLS 의 default.
Ignore tail latency: 매 p50 의 보고 의 — 매 p99 / p99.9 의 user experience.

🧪 검증 / 중복

Verified (Deutsch 1994 / Gosling 1997 original list, AWS Builders' Library, Google SRE book).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — 8 fallacies + resilience patterns

6.6 KiB Raw Blame History Unescape Escape