2nd/10_Wiki/Topics/Architecture/Broker Topology.md

---
id: wiki-2026-0508-broker-topology
title: Broker Topology
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Event Broker Topology, Pub-Sub Topology]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [event-driven, architecture, kafka, messaging]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: typescript
  framework: kafka-nats-rabbitmq
---

# Broker Topology

## 매 한 줄
> **"매 distributed pub-sub mesh — 매 central mediator 없이, broker (queue/topic) 가 routing fabric."**. Broker topology는 Event-Driven Architecture 의 두 종류 중 하나 (vs Mediator topology). 매 single-purpose events 의 fan-out 에 최적. 2026년 Kafka, NATS JetStream, AWS EventBridge, Redpanda, Pulsar 가 dominant brokers.

## 매 핵심

### 매 Broker vs Mediator
| 측면 | Broker | Mediator |
|---|---|---|
| Coordination | None — events fan out | Central mediator orchestrates |
| Coupling | Loosest | Some (mediator knows steps) |
| Use case | Simple fan-out, async notify | Complex multi-step workflow |
| Failure recovery | Each consumer handles | Mediator retries/compensates |
| Examples | Kafka topics, NATS subjects | Apache Camel, Step Functions |

### 매 컴포넌트
- **Producer**: events 의 publish.
- **Broker**: topic/subject/queue store + routing.
- **Consumer**: subscribe + react.
- **Schema Registry** (Confluent, Apicurio): event contract 의 versioning.
- **Dead Letter Queue (DLQ)**: failed messages.

### 매 broker 종류
- **Log-based** (Kafka, Redpanda, Pulsar): replay 가능, partitioned, ordered per partition.
- **Queue-based** (RabbitMQ, SQS): consumed once, no replay.
- **Subject-based** (NATS): lightweight, JetStream 으로 persistence.
- **Cloud-native** (EventBridge, Pub/Sub, EventHubs).

### 매 응용
1. Order events (e-commerce fan-out).
2. CDC (Debezium → Kafka → multiple sinks).
3. IoT telemetry ingestion.
4. Analytics event pipeline.

## 💻 패턴

### Kafka producer (TypeScript / KafkaJS)
```typescript
import { Kafka } from 'kafkajs';

const kafka = new Kafka({
  clientId: 'orders',
  brokers: ['kafka-1:9092', 'kafka-2:9092'],
});
const producer = kafka.producer();

await producer.connect();
await producer.send({
  topic: 'order.created',
  messages: [{
    key: order.id,
    value: JSON.stringify(order),
    headers: { 'content-type': 'application/json' },
  }],
});
```

### Kafka consumer (consumer group fan-out)
```typescript
const consumer = kafka.consumer({ groupId: 'shipping-svc' });
await consumer.connect();
await consumer.subscribe({ topic: 'order.created' });

await consumer.run({
  eachMessage: async ({ message }) => {
    const order = JSON.parse(message.value!.toString());
    await createShipment(order);
  },
});
```

### NATS JetStream
```typescript
import { connect, JSONCodec } from 'nats';

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

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

const sub = await js.subscribe('orders.>', {
  config: { durable_name: 'shipping' },
});
for await (const m of sub) {
  const order = jc.decode(m.data);
  await createShipment(order);
  m.ack();
}
```

### Schema Registry (Avro + Confluent)
```typescript
import { SchemaRegistry } from '@kafkajs/confluent-schema-registry';

const registry = new SchemaRegistry({ host: 'http://schema-registry:8081' });
const schema = `{
  "type": "record",
  "name": "OrderCreated",
  "fields": [
    { "name": "id", "type": "string" },
    { "name": "amount", "type": "double" }
  ]
}`;

const { id } = await registry.register({ type: 'AVRO', schema });
const encoded = await registry.encode(id, order);
await producer.send({ topic: 'order.created', messages: [{ value: encoded }] });
```

### DLQ pattern
```typescript
await consumer.run({
  eachMessage: async ({ topic, message }) => {
    try {
      await handleOrder(JSON.parse(message.value!.toString()));
    } catch (err) {
      await producer.send({
        topic: `${topic}.dlq`,
        messages: [{
          value: message.value,
          headers: {
            ...message.headers,
            error: String(err),
            retryCount: '0',
          },
        }],
      });
    }
  },
});
```

### Partitioning for ordering
```typescript
// Same orderId always goes to same partition → ordered per order
await producer.send({
  topic: 'order.events',
  messages: [{
    key: order.id, // partitioner uses key hash
    value: JSON.stringify(event),
  }],
});
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| High throughput, replay 필요 | Kafka / Redpanda |
| Lightweight, low-latency | NATS JetStream |
| Cloud serverless | EventBridge, Pub/Sub |
| Strong work-queue + ack | RabbitMQ, SQS |
| Multi-tenant, geo | Pulsar |

**기본값**: Kafka (or Redpanda — Kafka API compat) + Schema Registry + Avro/Protobuf for events.

## 🔗 Graph
- 부모: [[Event-Driven Architecture]]
- 변형: [[Mediator Topology]] · [[Choreography]]
- 응용: [[Kafka]] · [[NATS]]
- Adjacent: [[CDC]]

## 🤖 LLM 활용
**언제**: fan-out events, decoupled microservices, CDC pipelines, telemetry ingestion.
**언제 X**: synchronous request-reply (use gRPC/HTTP), small monolith (events 가 overhead).

## ❌ 안티패턴
- **No schema registry**: 모든 producer/consumer 가 schema drift — production 깨짐.
- **Topic per service**: scalability impossible — topic per event type.
- **No DLQ**: poison messages 가 consumer halt — DLQ + alerting.
- **Sync over async**: HTTP request-reply 를 broker 위에 — synchronous 면 broker 의 X.

## 🧪 검증 / 중복
- Verified (Richards "Software Architecture Patterns" / Confluent docs / NATS docs 2026).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — Broker vs Mediator + Kafka/NATS/DLQ 패턴 |