Operators Guide¶
This guide is for teams who deploy and run DC Suite — the control plane that turns a bare-metal GPU pool into a self-service cloud. It complements the User Guide, which covers using the platform.
What's here¶
| Topic | Page |
|---|---|
| How the pieces fit together | Architecture |
| Installing the control plane | Installation |
| Configuring backends and behavior | Configuration |
| Bringing physical machines into the pool | Bare-Metal Onboarding |
| Wiring up single sign-on | SSO & Identity |
| Shipping new versions | Deploys & Releases |
| Fixing common incidents | Runbooks |
Operating principles¶
- Fail closed on config. In production, DC Suite refuses to start without the endpoints it needs, so a misconfigured deploy never silently degrades.
- Fail open at runtime. If an optional backend (metrics, logs) is unreachable, provisioning keeps working — only the affected read path degrades.
- The fleet manager owns the hardware. DC Suite asks the fleet manager to reserve, power, validate, and release machines; it does not poke hardware behind the fleet manager's back.
- Everything is auditable. Privileged actions are recorded; access is scoped by role.
Public docs, generic values
Examples here use placeholders like your-dc-suite.example.com. Never put
real hostnames, IPs, tenant IDs, or credentials into public documentation.