เหตุการณ์ภายในแอปพลิเคชันและตัวรับฟัง

In a modular monolith, an Order module often needs to trigger work in other modules: send a confirmation email, update inventory, award loyalty points. The naive approach is to have OrderService call EmailService, InventoryService, and LoyaltyService directly.

• This creates tight coupling — Order depends on every downstream module.
• Adding a new reaction means editing OrderService again.
• Modules can no longer evolve or be tested independently.

Spring's internal application events invert this: the Order module simply publishes a domain event, and interested modules listen for it. The publisher never knows who is listening.

An internal event is just a plain object — in Spring Boot 4 (Java 21+) a record is the idiomatic choice. It should be immutable and carry only the data listeners need.

• Name it in the past tense (OrderCompleted) — it describes something that already happened.
• Keep it inside the owning module's package.
• No Spring annotations are required on the event itself.

The owning module publishes the event through Spring's ApplicationEventPublisher. Inject it via the constructor and call publishEvent() after the business operation succeeds.

• The publisher has no compile-time reference to any listener.
• By default publishing is synchronous — listeners run on the same thread, inside the same transaction.

Any Spring bean in another module can react by annotating a method with @EventListener. The single method parameter declares the event type the listener cares about.

• The method can have any name — Spring matches on the parameter type.
• The listener lives in its own module (com.shop.notification) and never references the Order service.

By default publishEvent() invokes every matching @EventListener synchronously, in order, on the calling thread. Because the listeners run inside the publisher's transaction:

• A listener that throws will propagate the exception back to the publisher and roll back the whole transaction.
• A slow listener (e.g. sending an email) blocks the business call until it finishes.

This is rarely what you want for side effects like email. The next scenes show how to fix both problems with transaction-bound and asynchronous listeners.

You usually want side effects to fire only if the transaction actually commits — you should not send a "your order is complete" email if the order save later rolls back. Use @TransactionalEventListener, which by default binds to the AFTER_COMMIT phase.

• The listener runs after the publisher's transaction commits successfully.
• If the transaction rolls back, the listener is skipped.
• You can choose other phases via phase = BEFORE_COMMIT, AFTER_ROLLBACK, or AFTER_COMPLETION.

To stop a slow listener from blocking the caller, add @Async. The listener then runs on a separate thread from a task executor, so publishEvent() returns immediately.

• Enable async support once with @EnableAsync on a configuration class.
• Important: an @Async listener runs in its own thread with no transaction and no access to the publisher's context. Combine it with @TransactionalEventListener(AFTER_COMMIT) when you want "after commit, on a background thread."
• Async listeners that throw are logged but do not affect the publisher — handle failures yourself.

Plain Spring events are in-memory only: if the application crashes between commit and the listener running, the event is lost. Spring Modulith adds durability with its Event Publication Registry.

• Add the spring-modulith-events-jpa (or JDBC/MongoDB) starter.
• Each @TransactionalEventListener invocation is recorded in an event_publication table within the same transaction.
• Once the listener completes, its publication is marked as completed. Incomplete publications can be republished on restart, guaranteeing at-least-once delivery.

With the events starter on the classpath, Modulith auto-creates the registry schema and wires interception. A little YAML controls recovery behaviour.

• republish-outstanding-events-on-restart resubmits any publications that never completed before the last shutdown.
• Modulith also exposes the registry so you can monitor or manually resubmit stuck events.

Sometimes a listener should only react to some events of a type. The condition attribute takes a SpEL expression evaluated against the event; the listener runs only when it returns true.

• Reference the event payload with #root.event or simply the property via # root.
• This keeps filtering declarative instead of an if at the top of the method.

Strip away Spring and the pattern is just a registry of listeners keyed by event type. The snippet below is a plain Java program that mimics the synchronous publish/subscribe behaviour — useful to internalize what ApplicationEventPublisher does under the hood.

You publish an OrderCompleted event inside a @Transactional order-save method. The email listener must run only if the order is actually persisted, and it must not delay or break the order-save call. Which listener configuration is correct?

You learned how to decouple modules with in-application domain events:

• Model events as immutable past-tense records owned by the publishing module.
• Publish via ApplicationEventPublisher.publishEvent() — the publisher never references listeners.
• React with @EventListener (synchronous, in-transaction) or @TransactionalEventListener (runs in a chosen transaction phase, default AFTER_COMMIT).
• Add @Async + @EnableAsync to run side effects off the caller's thread.
• Filter declaratively with the condition SpEL attribute.
• Use Spring Modulith's Event Publication Registry to persist publications and republish incomplete ones on restart for reliable, at-least-once delivery.

The result: modules that communicate through events stay independently testable, replaceable, and free of direct dependencies.