3.1. Planning and Implementing Compute Resources

💡 First Principle: Every Google Cloud compute option sits somewhere on a spectrum between "full control, full responsibility" and "no control, no responsibility" — and picking the right one is really answering "how much operational burden is this specific workload willing to carry in exchange for how much control?"

Choosing wrong in either direction has a real cost: forcing a stateless, bursty API onto Compute Engine means paying for idle capacity and building autoscaling logic Google already provides elsewhere for free; forcing a workload that needs privileged kernel access or specific licensing onto a fully serverless platform means hitting a wall that no amount of configuration can work around, because that platform was never designed to expose that level of control.

⚠️ Common Misconception: GKE Autopilot removes all capacity-planning responsibility from the operator. In reality, Autopilot manages the underlying nodes, but you still define Pod resource requests and limits, and those settings directly drive what you're billed for.