Distributed event bus works in-process (since default implementation is `LocalDistributedEventBus`) unless you configure an actual provider (e.g. [Kafka](Distributed-Event-Bus-Kafka-Integration.md) or [Redis](Distributed-Event-Bus-RabbitMQ-Integration.md)). In-process event bus always executes event handlers in the same [unit of work](Unit-Of-Work.md) scope that you publishes the events in. That means, if an event handler throws an exception, then the related unit of work (the database transaction) is rolled back. In this way, your application logic and event handling logic becomes transactional (atomic) and consistent. If you want to ignore errors in an event handler, you must use a `try-catch` block in your handler and shouldn't re-throw the exception.
Distributed event bus works in-process (since default implementation is `LocalDistributedEventBus`) unless you configure an actual provider (e.g. [Kafka](Distributed-Event-Bus-Kafka-Integration.md) or [RabbitMQ](Distributed-Event-Bus-RabbitMQ-Integration.md)). In-process event bus always executes event handlers in the same [unit of work](Unit-Of-Work.md) scope that you publishes the events in. That means, if an event handler throws an exception, then the related unit of work (the database transaction) is rolled back. In this way, your application logic and event handling logic becomes transactional (atomic) and consistent. If you want to ignore errors in an event handler, you must use a `try-catch` block in your handler and shouldn't re-throw the exception.
When you switch to an actual distributed event bus provider (e.g. [Kafka](Distributed-Event-Bus-Kafka-Integration.md) or [Redis](Distributed-Event-Bus-RabbitMQ-Integration.md)), then the event handlers will be executed in different processes/applications as their purpose is to create distributed systems. In this case, the only way to implement transactional event publishing is to use the outbox/inbox patterns as explained in the *Outbox / Inbox for Transactional Events* section.
When you switch to an actual distributed event bus provider (e.g. [Kafka](Distributed-Event-Bus-Kafka-Integration.md) or [RabbitMQ](Distributed-Event-Bus-RabbitMQ-Integration.md)), then the event handlers will be executed in different processes/applications as their purpose is to create distributed systems. In this case, the only way to implement transactional event publishing is to use the outbox/inbox patterns as explained in the *Outbox / Inbox for Transactional Events* section.
If you don't configure outbox/inbox pattern or use the `LocalDistributedEventBus`, then events are published at the end of the unit of work by default, just before the unit of work is completed (that means throwing exception in an event handler still rollbacks the unit of work), even if you publish them in the middle of unit of work. If you want to immediately publish the event, you can set `onUnitOfWorkComplete` to `false` while using `IDistributedEventBus.PublishAsync` method.
liangshiwei
3 years ago
committed by