The use case of ordering a meal is eventually consistent. Publishing the events reliably still ensures that all clients eventually know about the status of their order.

It is somewhat fine if processing the order doesn't happen immediately or if the order will fail for some reason. However, it must not happen that an order gets lost in the system due to unavailable applications. This needs to be ensured when publishing the events.

There are still transactions involved here, but on a much smaller scale and not involving external systems. Doing so enables distributed systems to cover transactional use cases while still enabling horizontal scalability.

The fact that some reliability is required for approaches like event-driven architectures is an important aspect in distributed systems, and should be considered when designing solutions and choosing technology.