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

Message Broker

매 한 줄

"매 producer 와 consumer 사이의 매 async middleman.". Message broker 는 매 service 간 매 decouple, 매 buffering, 매 fan-out 을 제공하는 매 infrastructure. 매 2026 의 big three 는 매 Kafka (event streaming), 매 RabbitMQ (traditional queue), 매 NATS (lightweight low-latency). 매 cloud-native 는 SQS / Pub/Sub / EventBridge 도 매 흔함.

매 핵심

매 왜 broker

Decoupling: producer 가 consumer 의 위치/존재 모름.
Buffering: spike 흡수 — consumer 가 천천히 처리.
Fan-out: 1 message → N consumer.
Reliability: persistent queue → consumer crash 후 replay.
Async: producer 가 response 안 기다림.

매 model

Queue (point-to-point): 1 message → 1 consumer (load balance).
Pub/Sub (topic): 1 message → N subscriber (broadcast).
Event log: 매 immutable append-only — replay/time-travel (Kafka).

매 보장

At-most-once: 매 fast, 매 loss 가능.
At-least-once: 매 default, 매 duplicate 가능 → idempotent consumer 필요.
Exactly-once: 매 어렵 (Kafka EoS transactions, idempotent producer + transactional consumer).

매 Kafka 핵심

Topic → Partition (parallelism unit) → Offset.
매 Consumer group 으로 partition load balance.
매 retention (time / size) — log 그대로 저장.
매 replay 가능 — offset reset.

매 RabbitMQ 핵심

Exchange (direct/topic/fanout/headers) → Binding → Queue.
매 AMQP 0.9.1 (현재) / AMQP 1.0 / MQTT / STOMP.
매 ack/nack, dead-letter exchange.
매 message TTL, priority queue.

매 NATS 핵심

매 lightweight, sub-ms latency.
매 Core NATS (fire-and-forget) + JetStream (persistence + at-least-once).
매 subject hierarchy: orders.eu.de.

💻 패턴

Kafka producer/consumer (Node.js, kafkajs)

import { Kafka } from 'kafkajs';

const kafka = new Kafka({ brokers: ['kafka:9092'], clientId: 'orders-svc' });

// Producer
const producer = kafka.producer({ idempotent: true });  // exactly-once-ish
await producer.connect();
await producer.send({
  topic: 'orders',
  messages: [{
    key: order.userId,  // partition key → ordering per user
    value: JSON.stringify(order),
  }],
});

// Consumer
const consumer = kafka.consumer({ groupId: 'fulfillment' });
await consumer.subscribe({ topic: 'orders', fromBeginning: false });
await consumer.run({
  eachMessage: async ({ message }) => {
    const order = JSON.parse(message.value.toString());
    await fulfill(order);  // must be idempotent
  },
});

RabbitMQ (Node.js, amqplib)

import amqp from 'amqplib';

const conn = await amqp.connect('amqp://localhost');
const ch = await conn.createChannel();

// Topic exchange + DLX
await ch.assertExchange('orders', 'topic', { durable: true });
await ch.assertQueue('fulfillment', {
  durable: true,
  deadLetterExchange: 'orders.dlx',
});
await ch.bindQueue('fulfillment', 'orders', 'order.created.*');

// Publish
ch.publish('orders', 'order.created.eu', Buffer.from(JSON.stringify(order)),
  { persistent: true });

// Consume with manual ack
ch.consume('fulfillment', async (msg) => {
  try {
    await fulfill(JSON.parse(msg.content.toString()));
    ch.ack(msg);
  } catch (e) {
    ch.nack(msg, false, false);  // → DLX
  }
}, { noAck: false });

NATS JetStream (Node.js)

import { connect, JSONCodec } from 'nats';

const nc = await connect({ servers: 'nats://localhost:4222' });
const js = nc.jetstream();
const jc = JSONCodec();

// Producer
await js.publish('orders.created.eu', jc.encode(order));

// Durable consumer
const sub = await js.subscribe('orders.>', {
  config: { durable_name: 'fulfillment', ack_policy: 'explicit' },
});
for await (const m of sub) {
  await fulfill(jc.decode(m.data));
  m.ack();
}

Idempotent consumer (deduplication)

async function handleMessage(msg) {
  const id = msg.headers['message-id'];
  // Try insert into processed table — unique constraint on id
  const inserted = await db.query(
    'INSERT INTO processed (id, ts) VALUES ($1, NOW()) ON CONFLICT DO NOTHING',
    [id]
  );
  if (inserted.rowCount === 0) return;  // already processed
  await businessLogic(msg);
}

Outbox pattern (transactional publish)

-- Same DB transaction: write business state + outbox entry
BEGIN;
  UPDATE orders SET status='paid' WHERE id=$1;
  INSERT INTO outbox (topic, payload, ts) VALUES ('orders.paid', $2, NOW());
COMMIT;

-- Separate process: poll outbox → publish to broker → mark sent

매 결정 기준

상황	Broker
Event streaming, replay, analytics	Kafka
Complex routing, traditional queue	RabbitMQ
Sub-ms latency, microservices	NATS
AWS-native, simple	SQS + SNS
GCP-native	Pub/Sub
Azure-native	Service Bus / Event Hubs
Single-process / dev	Redis Streams, in-memory

기본값: 매 cloud-native team → managed (SQS/Pub-Sub). 매 self-host + event log → Kafka. 매 traditional queue → RabbitMQ.

🔗 Graph

부모: Distributed Systems · Event-Driven Architecture
변형: Kafka · RabbitMQ · NATS · SQS · Pub-Sub
응용: Microservices · Event Sourcing · CQRS
Adjacent: Message-Queues-and-Event-Streams · Dead Letter Queue

🤖 LLM 활용

언제: service decoupling, async workflow, event-driven design, spike absorption. 언제 X: 매 simple sync RPC 면 충분 — 매 broker = operational overhead.

❌ 안티패턴

At-most-once 로 critical data: 매 loss 허용 안되면 매 ack 필수.
No idempotency: 매 at-least-once 에서 매 duplicate → side effect 2번.
Big payload: 매 broker 에 large blob → 매 S3 reference 패턴 사용.
No DLQ: 매 poison message 가 매 무한 retry → consumer 마비.
Kafka 를 queue 로: 매 ordering 만 필요한 매 short job → RabbitMQ 가 더 적합.

🧪 검증 / 중복

Verified (Apache Kafka docs, RabbitMQ docs, NATS docs, Confluent blog, Microservices.io patterns).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Kafka/RabbitMQ/NATS 비교 + outbox/idempotent 패턴

6.6 KiB Raw Blame History