You have internal-only microservices on GKE, need to set per-service replica counts, and want a uniform way for services to address each other regardless of replica count. What should you do?

The correct approach is to deploy each microservice as a Deployment and expose it using a Service. A Kubernetes Deployment manages replica sets, allowing you to easily scale the number of instances for each microservice independently. A Service provides a stable DNS name and load balancing across the Pods managed by the Deployment, ensuring that other microservices can consistently address it without needing to know individual Pod IPs or replica counts. Ingress is primarily for external access to services, not for internal cluster communication. Deploying microservices as individual Pods is not ideal because Pods are ephemeral and lack built-in scaling and management capabilities that Deployments provide.