Predictive Autoscaler

predictive-scaler reads historical metric data from PostgreSQL, runs Holt-Winters exponential smoothing forecasts, and generates proactive scaling recommendations that operators or AI agents can approve.

Forecasting Algorithm

Double exponential smoothing (Holt-Winters without seasonality) is implemented in pure Go — no Python or ML framework dependencies:

Level:   L(t) = α × y(t) + (1-α) × (L(t-1) + T(t-1))
Trend:   T(t) = β × (L(t) - L(t-1)) + (1-β) × T(t-1)
Forecast at h steps ahead: ŷ(t+h) = L(t) + h × T(t)

Prediction interval: ŷ ± 1.96 × RMSE

Requirements:

• Minimum 24 data points before a forecast is generated
• Alpha (level smoothing) and beta (trend smoothing) are configurable
• Supports metrics: cpu_usage_pct, memory_usage_pct, request_rate

Domain Model

MetricSeries
  └── cluster_id, namespace, workload, metric, value, timestamp

Forecast
  ├── cluster_id, namespace, workload, metric
  ├── algorithm: holt_winters | moving_avg
  ├── points: []ForecastPoint { timestamp, value, lower_bound, upper_bound }
  ├── confidence, horizon_mins
  └── generated_at

ScalingDecision
  ├── forecast_id, cluster_id, namespace, workload
  ├── current_replicas, recommend_replicas
  ├── reason, confidence
  ├── status: pending | approved | applied | rejected
  └── executed_at

REST API

Method	Path	Description
GET	/api/v1/forecasts	List latest forecasts (?cluster_id=&namespace=&workload=)
POST	/api/v1/forecasts/generate	On-demand forecast for a specific workload
GET	/api/v1/scaling-decisions	List recommendations (?status=pending)
POST	/api/v1/scaling-decisions/{id}/approve	Execute scaling action
POST	/api/v1/scaling-decisions/{id}/reject	Reject recommendation

Approval Flow

When a ScalingDecision is approved (via UI, API, or AI agent):

1. predictive-scaler patches the workload's HPA spec.minReplicas to the recommended value
2. If the recommendation requires new nodes (recommended replicas > current node capacity), it calls nodes-manager via HTTP to scale the node group
3. Updates ScalingDecision.status to applied and records executed_at

Kubernetes RBAC Requirements

rules:
  - apiGroups: ["autoscaling"]
    resources: ["horizontalpodautoscalers"]
    verbs: ["list", "get", "patch", "update"]
  - apiGroups: ["apps"]
    resources: ["deployments"]
    verbs: ["list", "get", "patch"]

Environment Variables

Variable	Default	Description
DATABASE_URL	—	PostgreSQL connection (shared with k8s-monitor)
RABBITMQ_URL	—	Optional. For consuming metric snapshots via exchange
HOLT_WINTERS_ALPHA	0.3	Level smoothing factor (0–1)
HOLT_WINTERS_BETA	0.1	Trend smoothing factor (0–1)
FORECAST_HORIZON_MINS	60	How far ahead to forecast
NODES_MANAGER_BASE_URL	—	URL for node group scaling
PORT	8089	HTTP port