The following snippet shows an event handler of the chef system, invoking functionality of a command service:

@Singleton
public class OrderEventHandler {

    @Inject
    MealPreparationService mealService;

    public void handle(@Observes OrderPlaced event) {
        mealService.prepareMeal(event.getOrderInfo());
    }
}

The event handler consumes the event and will invoke the boundary of the subsequent meal preparation use case. The prepareMeal() method itself will result in zero or more events, in this case either MealPreparationStarted or OrderFailedInsufficientIngredients:

public class MealPreparationService {

    @Inject
    EventProducer eventProducer;

    @Inject
    IngredientStore ingredientStore;

    public void prepareMeal(OrderInfo orderInfo) {

        // use ingredientStore to check availability

        if (...)
            eventProducer.publish(new OrderFailedInsufficientIngredients());
        else
            eventProducer.publish(new MealPreparationStarted(orderInfo));
    }
}

The event producer will reliably publish the events to the Kafka cluster. If the publication fails, the whole event processing has to fail, and will be retried later.