If you were tasked with scaling a microservices architecture, what steps would you take to ensure high availability?