Skip to main content

Hub-and-spoke architecture

One central hub gateway plus N spoke gateways. Spokes are lightweight and deployed close to the data, so they default to the Edge edition; the hub aggregates them. The spoke count drives a RAM advisory so a large demo does not quietly outgrow the host.

What you get

GatewayRoleHTTP portEdition
hubhub9088standard
spoke-1spoke9089Edge by default
spoke-2spoke9090Edge by default
spoke-Nspoke9088 + NEdge by default
Gateways1 hub + N spokes
Default databasePostgres
Network splitoff (hub-and-spoke is typically single-tier)
Editionspokes run Edge by default; the hub is always standard

Each spoke gets its own host port, stepping up from 9089, so you can open any gateway directly from your laptop.

The RAM advisory

A gateway costs roughly 1.5 GB of RAM, so spoke count is the knob most likely to overcommit a demo machine. The advisory has three tiers:

  • green (1-4 spokes): proceed without prompting.
  • yellow (5-8 spokes): the wizard asks for confirmation; the non-interactive path proceeds.
  • red (9 or more spokes): the wizard asks you to acknowledge the cost; the non-interactive path refuses unless you pass --force.

The wizard surfaces the estimate against available memory so the cost is visible at decision time. If you decline a yellow or red prompt, the wizard steps the count back down to 4 (the top of green) rather than stranding you.

Run it

ignition-stack create demo --arch hub-and-spoke --spokes 3
ignition-stack create big-demo --arch hub-and-spoke --spokes 12 --force

The second command crosses the red tier, so --force is required to proceed non-interactively.

Gateway network

Each spoke is joined to the hub over Ignition's Gateway Network, and the link auto-forms on first boot with no UI approval. Every spoke opens a plain (non-SSL, port 8088) outgoing connection to the hub, and each gateway runs an Unrestricted incoming policy so the connection is accepted on sight. The generated POST-SETUP.md carries a verification readout — naming both ends of each link — rather than a manual approval step.

Plain transport is a deliberate demo-only default. For a cross-host or production deployment, switch to SSL on port 8060 with approved certificates, which reintroduces a one-time cert-approval step; the redundancy guide walks through that variant.

Editions

Spokes default to Edge; pass --edge-role none to run them as standard instead. The hub is always standard and cannot be made Edge: the spokes aggregate into it, and Edge is a leaf edition that can't be an aggregation target.

When to use it

Use hub-and-spoke to show a central gateway aggregating many edge sites: remote Edge gateways feeding a hub. Keep the spoke count realistic for the host running the demo, and let the advisory be the guardrail.