Files

T

koriweb d8a80f6272 chore(wiki): dangling 링크 canonical 정규화 (768파일/1200건)

이름만 다른(표기 변형) [[위키링크]]를 대상 문서의 canonical 제목으로 치환해
끊겼던 1,200개 링크를 연결. 제목/파일명 정규화 일치만 적용하고 별칭 매칭은
과병합 위험으로 제외(애매성 가드). 원본은 _link_reconcile_backup/ 에 백업.
도구: Datacollect/scripts/link_reconcile_apply.mjs

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-08 12:24:15 +09:00

5.8 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

Netflix 마이크로서비스 전환

매 한 줄

"매 monolith → 매 1000+ microservice, 매 7년 의 journey". 2008 DB 손상 outage → 2009 AWS migration 시작 → 2016 완전 전환. 매 Chaos Engineering, Spinnaker, Hystrix, Eureka 의 birthplace. 매 modern microservice playbook 의 reference.

매 핵심

매 timeline

2008.08: DVD shipping DB corruption — 3-day outage. Monolith fragility 의 trigger.
2009-2012: Cassandra adoption, Edge service split, AWS migration 시작.
2013-2015: Hystrix, Eureka, Ribbon, Zuul OSS 공개.
2016.01: 마지막 datacenter 종료. 100% AWS.
2020+: Service mesh (Envoy), gRPC migration, GraphQL federation.

매 driver

Scale: 매 30M (2008) → 200M+ (2020) subscribers.
Velocity: 매 daily deploy 의 thousands.
Resilience: 매 region failure 의 graceful degrade.
Polyglot: 매 service 의 own stack 의 freedom.

매 응용

Stream startup: 매 Netflix OSS 의 reuse (Eureka, Spinnaker).
Enterprise migration: strangler pattern + edge first.
Resilience: Chaos Monkey 의 culture adoption.

💻 패턴

Strangler Fig migration

// API Gateway routes — gradual cutover
@Component
public class FeatureRouter {
    @Value("${migration.user-service.percent}") int migrationPercent;

    public Route route(Request req) {
        if (req.path().startsWith("/users")
            && ThreadLocalRandom.current().nextInt(100) < migrationPercent) {
            return Route.NEW_MICROSERVICE;
        }
        return Route.LEGACY_MONOLITH;
    }
}

Circuit breaker (Resilience4j, Hystrix successor)

import io.github.resilience4j.circuitbreaker.*;

CircuitBreakerConfig cfg = CircuitBreakerConfig.custom()
    .failureRateThreshold(50)
    .waitDurationInOpenState(Duration.ofSeconds(30))
    .slidingWindowSize(20)
    .build();

CircuitBreaker cb = CircuitBreaker.of("recommendations", cfg);

Supplier<Recommendations> call = CircuitBreaker.decorateSupplier(
    cb, () -> recoClient.fetch(userId));

Recommendations result = Try.ofSupplier(call)
    .recover(t -> Recommendations.fallback()) // degraded UX
    .get();

Eureka service discovery (client)

@SpringBootApplication
@EnableDiscoveryClient
public class RecommendationServiceApp { }

// caller
@Autowired DiscoveryClient discovery;
List<ServiceInstance> instances = discovery.getInstances("user-service");

Chaos Monkey schedule

# spinnaker chaos-monkey config
chaos:
  enabled: true
  schedule: "MON-FRI 09:00-15:00 PT"
  meanTimeBetweenKillsInWorkDays: 2
  exceptions:
    - { account: prod, region: us-east-1, stack: critical-payment }

Spinnaker pipeline (deploy)

{
  "stages": [
    { "type": "bake", "package": "user-service", "baseOs": "bionic" },
    { "type": "deploy",
      "clusters": [{ "account": "prod", "region": "us-east-1",
                     "strategy": "redblack", "capacity": { "min": 6, "max": 60 } }] },
    { "type": "wait", "waitTime": 600 },
    { "type": "checkPreconditions",
      "preconditions": [{ "type": "expression", "context": "kayentaPass" }] }
  ]
}

Bulkhead isolation

ThreadPoolBulkheadConfig bulkhead = ThreadPoolBulkheadConfig.custom()
    .maxThreadPoolSize(10)
    .coreThreadPoolSize(5)
    .queueCapacity(20)
    .build();

Backpressure with reactive

// Spring WebFlux
public Flux<Movie> stream(String userId) {
    return userClient.getProfile(userId)
        .timeout(Duration.ofMillis(200))
        .flatMapMany(p -> recoClient.recommend(p))
        .onBackpressureBuffer(1000, BufferOverflowStrategy.DROP_OLDEST);
}

Canary analysis (Kayenta)

canaryConfig:
  metrics:
    - name: error-rate
      query: "avg(error_count) / avg(request_count)"
      criticality: critical
      direction: increase
    - name: latency-p99
      criticality: high

매 결정 기준

상황	Approach
Startup pre-PMF	Monolith. Don't copy Netflix yet.
Growing (50+ eng)	Extract edge services first
Scale (Netflix-class)	Full microservices + service mesh
Resilience-critical	Chaos engineering + canary mandatory

기본값: Modular monolith → strangler extraction when team/load demands.

🔗 Graph

부모: Microservices
변형: Service-Oriented-Architecture · Service Mesh
응용: Chaos-Engineering · Spinnaker
Adjacent: Strangler-Pattern · Circuit-Breaker

🤖 LLM 활용

언제: 매 case study 의 summarize, OSS Netflix tool 의 explain, migration sequence 의 propose. 언제 X: 매 own org 의 readiness 판단 — team maturity, ops capacity 의 honest assessment.

❌ 안티패턴

Cargo cult: 매 5-person startup 의 microservices = 매 distributed monolith hell.
No observability first: 매 100 services + no tracing = debug 의 impossible.
Big bang migration: 매 monolith 의 1 day 의 split = outage.
Skip chaos: 매 production failure mode 의 unknown until customer hits it.

🧪 검증 / 중복

Verified (Netflix Tech Blog 2009-2024, "Building Microservices" by Newman, AWS re:Invent Netflix talks).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Netflix microservices migration case study + patterns

5.8 KiB Raw Blame History