ScalePilot: Predictive, Cost-Aware Kubernetes Autoscaling

Kubernetes HPA reacts , it scales up after the spike hits. By then , users are already seeing errors and your SLOs are already breached.

ScalePilot is an open source Kubernetes operator that fixes this with four custom resources:

• ForecastPolicy: trains ARIMA or Holt-Winters models on your Prometheus metrics and raises HPA minReplicas before the predicted spike arrives not after.

• ScalingBudget: enforces cloud cost limits across AWS, GCP , and Azure.When a namespace breaches its budget, ScalePilot blocks further scale-ups via an admission webhook and fires Slack/PagerDuty alerts.

• FederatedScaledObject: spills workloads to secondary clusters when the primary is under pressure multi-cluster overflow without custom glue code.

• ClusterScaleProfile: apply global scaling constraints across all workloads in a cluster from a single cluster-scoped resource.

In this talk, I'll walk through why reactive autoscaling is not enough, how ScalePilot's forecast engine works under the hood, and a live demo of ForecastPolicy preventing a traffic spike from degrading a workload all driven by plain Kubernetes YAML.

1. Why ARIMA/Holt-winters beat simple threshold rules for predicting traffic spikes, and how to pick between them.

2. How ScalePilot's operator pattern works: CRDs as the API, reconcilers as the control loop, admission webhooks as the enforcement layer.

3. How to get started: install CRDs, write a ForecastPolicy pointing at a Prometheus query, and let ScalePilot manage your HPA minReplicas.