If the response is always 200 even when an error occurs I think we would need to find a way to expose the error into the nginx logs or somewhere else we can sense the error.
In my experience, this is typically where application metrics are used. Instead of having nginx tell you the result, you add custom code to the RPC server to report the error using statsd or prometheus formats.
