Architecture

System Flow

1. A business write and its outbox record are committed in the same Mongo transaction.
2. OutBoxPublisherService polls records in PENDING / RETRY_PENDING status on a fixed delay.
3. Records are published to Kafka via the OutboxEventProducer; success marks them COMPLETED.

---

Components

Component	Responsibility
OutboxAutoConfiguration	Registers KafkaTemplate<String,OutboxKafkaEvent>, OutboxEventProducer, and @EnableScheduling. Conditional on arya.outbox.enabled=true. Registered via META-INF/spring/org.springframework.boot.autoconfigure.AutoConfiguration.imports.
OutboxProperties	@ConfigurationProperties("arya.outbox") — interval, retries, enable flag.
OutBoxPublisherService<T>	Generic relay. @Scheduled(fixedDelayString = "${arya.outbox.publish-interval-ms:5000}"). publishPendingAndRetry() fetches findByOutboxStatusIn([PENDING, RETRY_PENDING]), sends via the producer keyed by aggregateId. Success → COMPLETED; exception → RETRY_PENDING. handleEventUpdate increments retryCount and marks FAILED when retries reach maxRetries.
OutboxEventProducer	Thin wrapper around KafkaTemplate<String,OutboxKafkaEvent>.
OutboxEventRepository<T>	@NoRepositoryBean MongoRepository contract exposing findByOutboxStatusIn(...).

---

Status Lifecycle

• The producer keys messages by aggregateId, so all events for one aggregate land in the same Kafka partition — preserving per-aggregate ordering.
• Retries are bounded: a record loops PENDING ↔ RETRY_PENDING until retryCount hits maxRetries, then becomes FAILED.