Bare-Metal Onboarding¶

Capacity comes from physical machines you register and hand to the fleet manager. This page covers the machine lifecycle and the day-2 states you'll see.

Registering a machine¶

An operator registers a machine with its hardware spec and BMC (out-of-band management) coordinates and credentials. The BMC password is write-only — DC Suite stores it encrypted and never returns it in API responses; responses only report whether a password is set.

Registration flow:

stateDiagram-v2
    [*] --> REGISTERED
    REGISTERED --> ONBOARDING
    ONBOARDING --> VALIDATING
    VALIDATING --> READY
    ONBOARDING --> ERROR
    VALIDATING --> ERROR
    ERROR --> ONBOARDING
    READY --> DECOMMISSIONING
    DECOMMISSIONING --> DECOMMISSIONED

State	Meaning
REGISTERED	Known to DC Suite, not yet handed to the fleet manager.
ONBOARDING	Registration with the fleet manager is in flight.
VALIDATING	The fleet manager is discovering/validating the machine.
READY	Validated and usable; available to the pool.
ERROR	Onboarding or a later report failed; `last_error` explains why.
IN_USE	Display only — an in-cluster machine the fleet reports in error; shown as in-use rather than a scary error.
DECOMMISSIONING / DECOMMISSIONED	Being removed / removed (record kept).

Before onboarding, DC Suite validates the BMC is reachable with the given credentials; onboarding is gated on a fresh passing BMC check.

Power-state contract¶

Idle pool machines are parked powered off to save energy. A responsive BMC on a cold host is the "machine is healthy and available" signal — DC Suite tolerates the powered-off state for parked machines and powers a machine on only when it's being allocated to a cluster.

Fleet sync and pool health¶

A periodic sync reconciles DC Suite's node table with the fleet manager:

A machine the fleet reports available projects as AVAILABLE (bookable).
A machine the fleet reports assigned/sanitizing/unhealthy is QUARANTINE so it's never handed out by mistake.
A quarantined, unowned machine the fleet certifies healthy again is recovered back to AVAILABLE automatically — once its bare-metal registration is itself READY.

Day-2 states you'll see¶

ERROR on a machine that's actually in a running cluster — often a stale background validation the fleet couldn't run while the box was busy. DC Suite displays such machines as IN_USE rather than a red error; the stored state stays ERROR so auto-recovery still applies once the fleet clears it.
A quarantined machine that won't recover — its bare-metal row is stuck in ERROR/VALIDATING, or the fleet manager is stuck mid-terminate. See the node recovery runbook.

Decommissioning¶

Decommissioning removes a machine from service. The record is retained for history. Only machines not currently serving a cluster should be decommissioned.

Next: SSO & Identity.