How would you handle a situation where a critical microservice goes down in production?