Centurion — Kubernetes Cluster Agent
Centurion is a single-replica Deployment that monitors your Kubernetes cluster. It watches all resource events, tracks topology and ownership, detects anomalies, and streams structured data to the Saviour backend.
What it monitors
| Feature | Description |
|---|---|
| Events | All Kubernetes events with severity classification (critical / error / warning / info) |
| Topology | Pod → Deployment → Service ownership graph, updated continuously |
| Service Discovery | Ingress, Gateway API, Istio, OpenShift service exposure |
| Anomaly Detection | Restart loops, OOM kills, scheduling failures, event floods, pending pods |
Architecture
Centurion runs as a single pod (replicaCount: 1) using an in-cluster ServiceAccount with read-only RBAC. It connects to the Kubernetes API via informers (watch + cache) and batches events before sending to the Saviour backend.
The agent survives temporary backend outages via an offline SQLite store (offlineStore) — events are buffered locally and replayed when connectivity resumes.
Modes
| Mode | Behaviour |
|---|---|
server | Events sent to Saviour backend (default) |
log_only | Events logged locally only — no backend required |