event-driven

name	event-driven
description	Kafka topic naming, producer/consumer configuration, outbox pattern, and idempotency strategies. Use when implementing event-driven messaging with Kafka.
triggers	["kafka producer","kafka consumer","kafka topic","outbox pattern","idempotent consumer","event driven architecture","카프카 설정","이벤트 발행"]

1. Kafka Topic Naming

Pattern	Example	Usage
{domain}.{event-type}	order.created	Domain events
{domain}.{entity}.{action}	product.stock.reserved	Fine-grained events
{source}.{target}.dlq	order.payment.dlq	Dead letter queue
{domain}.retry	payment.retry	Retry topic

2. Producer Configuration

Property	Value	Reason
acks	all	Durability guarantee
retries	3	Transient failure handling
enable.idempotence	true	Exactly-once semantics
key.serializer	StringSerializer	Partition by order ID
value.serializer	JsonSerializer	Event payload
max.in.flight.requests	5 (with idempotence)	Ordering guarantee

3. Consumer Configuration

Property	Value	Reason
group.id	{service-name}-group	Consumer group per service
auto.offset.reset	earliest	Don't miss events on new consumer
enable.auto.commit	false	Manual commit after processing
max.poll.records	100	Batch size control
isolation.level	read_committed	Transactional consistency

4. Outbox Pattern

• outbox_event table: id, aggregate_type, aggregate_id, event_type, payload (JSONB), created_at, published_at
• Write to outbox in same transaction as aggregate mutation
• Polling publisher reads unpublished events, publishes to Kafka, marks as published
• Alternative: @TransactionalEventListener + Kafka producer (simpler, less reliable)
• Outbox records retained for audit, purged after N days

5. Idempotency

Strategy	How	When
Idempotency key	Store processed event IDs in processed_events table	All consumers
Natural idempotency	UPDATE SET status = 'PAID' WHERE status = 'PENDING'	State transitions
Deduplication window	Check event ID against recent N minutes	High-throughput consumers