K8s Monitor

k8s-monitor is the primary data collection service. It polls the Kubernetes API and cloud provider pricing APIs on a configurable interval and exposes aggregated health, cost, and metrics data via a REST API.

What It Does

On every report cycle (executeReportCycle):

1. Queries node and pod states from the Kubernetes API
2. Computes a health score (0–100)
3. Fetches cloud pricing data and calculates per-namespace costs
4. Generates optimisation recommendations
5. Appends a MetricPoint to the time-series store
6. Runs the statistical anomaly detector
7. If anomalies are found and RABBITMQ_URL is set, publishes to k8s.anomalies exchange

REST API

Method	Path	Description
GET	/api/health	Cluster health score, node status, control plane
GET	/api/cost	Cost breakdown by namespace, workload, resource type
GET	/api/optimizer	Over-provisioned and idle workload recommendations
GET	/api/metrics/pods	Per-pod CPU and memory usage
GET	/api/metrics/nodes	Per-node CPU, memory, and disk metrics
GET	/api/history	Historical metric time-series (?start=&end=&cluster_id=)
GET	/api/anomalies	Latest anomaly events from the statistical detector

Health Score Formula

health_score = (
  node_readiness_pct   × 0.40 +
  pod_success_rate     × 0.30 +
  control_plane_health × 0.20 +
  api_latency_score    × 0.10
) × 100

Values are clamped to 0–100. The score is returned as an integer in the health response alongside component breakdowns.

Time-Series Storage

k8s-monitor supports two storage backends, selected at startup:

• In-memory ring buffer (default) — goroutine-safe, holds the last N points (default 1000). Zero config. Data is lost on restart.
• PostgreSQL (opt-in via DATABASE_URL) — persists metric_snapshots to the k8s_monitor schema. Enables the /api/history endpoint and feeds the predictive-scaler.

Anomaly Detection

The StatisticalDetector runs a Z-score calculation over a rolling window after each report cycle:

z = (current_value - rolling_mean) / rolling_std_dev

Default thresholds:

• Z-score ≥ 2.5 → high severity
• Z-score ≥ 3.5 → critical severity

Window size and thresholds are configurable via environment variables.

Environment Variables

Variable	Default	Description
DATABASE_URL	—	Optional. Enables PostgreSQL time-series storage
RABBITMQ_URL	—	Optional. Enables anomaly event publishing
ANOMALY_WINDOW_SIZE	60	Number of data points in rolling window
ANOMALY_Z_THRESHOLD	2.5	Z-score threshold for high severity
CLUSTER_ID	default	Identifier attached to all metrics
REPORT_INTERVAL	30s	How often to run a health cycle