Skip to main content
Version: 3.0 (next)

Edge plus DMZ hub

An edge on each line publishes upward into a highly available hub in the plant DMZ. Business systems talk to the hub only, and nothing opens a connection into the control network. This matches plants that separate IT and OT into zones with defined conduits between them.

Scenario
2 of 5
Shape
Tier 1 per line · Tier 3 in the DMZ
Applies to
3.0

Choose it when IT and OT are separated, lines must keep running while the DMZ is down, and business systems need an API that survives a server failure.

Edge plus DMZ hubTier 1 per line · Tier 3 in the DMZ
  • MaestroHub
  • Reads machines
  • Data
  • Buffered on disk
  • Management
  • Your tooling
L5EnterpriseL4BusinessL3.5Plant DMZL3OperationsL2SupervisoryL1ControlPlant DMZhub onlyStore-and-forward: queued on the sender's disk while the far side is downStore-and-forward: queued on the sender's disk while the far side is downStore-and-forward: queued on the sender's disk while the far side is downCorporate ERP · lakeKafka · SnowflakeMESsiteLIMSqualitySAPsiteAPI gatewayyoursMaestroHub · Tier 3Three-node Kubernetes · one Helm chartOne pod per module · 2 to 3 replicasNATS ×3 · EMQX ×3 · TimescaleDB ×3 · ES ×3RKE2 or K3s · 3 × (16 vCPU · 24 GB)LinuxRegistry · GitOpsHarbor · Flux: yoursimages · Helm valuesMaestroHub · Tier 1Line 1 edgeConnectors · line UNS · queueNative serviceMaestroHub · Tier 1Line 2 edgeConnectors · line UNS · queueNative serviceStandby edgeno failover todaycold spare only!SCADAreads Line 2 UNSS7 PLCLine 1S7 PLCLine 1S7 PLCLine 2DrivesLine 2 · ModbusMQTTMQTT · outboundREST · SQL · IDocKafkaimages · values

What you install

LevelComponentHow it runsRecommended hardware
L3.5MaestroHub site hub, Tier 3Helm chart with production values on three Kubernetes nodes (RKE2, K3s or any conformant distribution)3 nodes, each 16 vCPU, 24 GB RAM, 500 GB SSD
L3MaestroHub edge, Tier 1, one per lineOne binary as a native service8 vCPU, 16 GB RAM, 200 GB SSD each

The edges publish to the hub's EMQX over MQTT. Each edge needs an outbound route to the hub's MQTT listener and nothing else; the hub never connects down to an edge. A registry mirror and a GitOps tool can deliver the hub's images and Helm values; MaestroHub needs neither.

When something fails

What failsWhat happens
One hub nodePods restart on the other nodes, and leaders and connection ownership move to surviving replicas through NATS leases. Spreading the nodes over racks, power feeds and storage is part of your platform design.
The DMZ, or the whole hubEach edge keeps reading its line and serving local subscribers such as SCADA. Its writes to the hub wait in Store & Forward buffers on the edge.
An edge PCThat line is not read until a spare PC with the same configuration takes over.

Limits. Two Tier 1 edges do not fail over to each other: both would poll the same PLCs. Keep a configured spare, or connect a line that cannot wait directly to the Tier 3 hub, where connection ownership already moves between replicas. Tier 3 needs a team that runs Kubernetes, storage and upgrades.

Previous: Line-side edge · Next: One hub, many sites · How to read the drawings: Deployment scenarios