`endpoint_run.grape` callback is called without middlewares

Hello. I'm trying to add the monitoring by subscribing to 'endpoint_run.grape' via ActiveSupport::Notifications.subscribe.

It works well until an exception occurs inside API. I have a bunch of rescue_from blocks to return different status codes depending on exceptions, such as

rescue_from ActiveRecord::RecordNotFound do
  error!({ error: :not_found }, 404)
end

The problem is when I call an endpoint which raises ActiveRecord::RecordNotFound, it really returns 404 status code, but in my callback data[:endpoint].status is still 200 (however data[:exception_object] is not null).

A quick look at the code shows, that callback wraps only the main call, without middlewares. I believe, it should wrap middlewares too because:

• middlewares are part of the business logic. It's not good to have 404 returned to a client and 200 reported to the monitoring.
• timing will be more accurate because the middlewares can eat time too.

The issue points to the endpoint_run.grape notification and the callback's wrapping of the main call versus middlewares; start by locating that notification and tracing middleware execution. Reproduce a rescue_from ActiveRecord::RecordNotFound request, then verify the callback observes the final 404 status and includes middleware time.