How can you address and prevent performance issues in your growing microservices-based application that uses a single large Cloud SQL instance, where CPU utilization appears normal in Cloud Monitoring, while adhering to Google-recommended practices and minimizing major refactoring?