Fleet control plane for 86 on-prem servers

A large ERP runs on its own on-prem server at each of 88 sites (86 servers). Every update, configuration fix and support session used to need a person on site, and versions drifted apart.

• Sites sit behind NAT on consumer-grade links, mostly Windows boxes.
• Updates must never interrupt work during opening hours.
• Remote access must be audited; no shared SSH keys.
• The application is big: 1,001 schema migrations to roll out safely.

• A Go management server (68 releases) that holds desired state per site: modules, configuration and versions.
• Per-site agents: a sidecar that pulls and applies desired state, a migrator for safe schema rollout, and a sentinel for health and alerting.
• A CLI and an audited remote terminal that replace on-site visits and shared credentials.
• An operator dashboard in React + TanStack Router/Query/Table with an xterm.js web terminal (45 releases).
• dockwarden, my open-source Go container auto-updater, running on every box.

• 88 sites live on 86 on-prem servers (70 Windows, 15 Linux), 85 of them on the mesh
• 83 of 88 sites on the current release through automatic pull updates
• 80 sites heartbeating in the last 24 hours (most switch off overnight)
• 600k+ patient records and 590k+ billing records under management (a lower bound, from a 26-server sample)
• 42 fleet-wide automation runs since Aug 2026, every remote command audited

Production figures, October 2026.

Transfers to: Edge and retail fleets, IoT, and any company running software on customer premises.

• Go
• WireGuard
• Docker
• PostgreSQL
• S3
• Laravel
• React
• TanStack