Skip to main content

T4 — Scale: Multi-Site Fleet Delivery (Advanced)

[!NOTE] Advanced tier. This is the legitimate top of the necessary ladder. Reach it only when robots span multiple sites you cannot directly reach. Single-site teams stay at T3 — Production.

T4 is the fleet-delivery control plane: getting validated policies onto robots across sites you cannot directly reach, safely, with a gate before a policy swaps on a physical arm. The defining difference from T3 is multiple sites — which is exactly what makes Azure Arc necessary, as the cross-site reachability and identity broker that single-site k3s did not need. Here, "fleet" means a fleet of robots, not Kubernetes clusters.

[!IMPORTANT] T4 delivers and gates policies. It excludes drift detection, automated retraining, and aggregate telemetry analytics. Those are fleet intelligence at T5 — Operate.

🧱 Minimum Infrastructure

ConcernWhat you need
HardwareRobots across multiple sites you cannot directly reach.
Edge infraAzure Arc + AKS or Arc-enabled Kubernetes + FluxCD + a deployment gating service.
Cloud infraT2 cloud + cross-site connectivity and identity, plus the model registry.
DeliveryFluxCD GitOps; per-site desired state recorded in Git; gating before a policy swaps.

🚀 Where to Go

This is a stub. The multi-site fleet-delivery mechanics are documented in the existing deployment docs. This recipe deliberately does not duplicate them:

🎓 Graduate When

  • The operator explicitly wants production signals to drive retraining and fleet-wide health analytics. This is a deliberate decision, not an automatic consequence of scale: T5 — Operate (roadmap).