I built this after a couple of throttling alerts forced me to revisit deployment limits.
First deployed this in ACA (live demo on my site), and I've now extended the same idea to Kubernetes on AKS: collecting workload metrics for a target deployment, evaluating node fit, and generating advice on whether the workload needs more resources, still fits after the increase, or needs more cluster headroom.
What I like most is that the recommendation is not just "CPU is high." It looks at:
- workload requests and limits,
- live usage,
- node headroom,
- cluster saturation trends.
Still improving it, but the end-to-end flow is working now in AKS. Next up: add a webhook for Azure Monitor throttling and memory alerts.