Fixed replicas are always wrong. You're either wasting resources or under-provisioned.
Here's how to build Kubernetes autoscaling that goes from 0→5 replicas based on actual traffic, using tools you probably already have.
youtu.be/K0aaB2L3sXI
youtu.be
Kubernetes Serverless Without the Vendor Lock-In (Here's How)
YouTube video by DevOps & AI Toolkit