Bare-metal provisioning
Install registered hardware through PXE or iPXE and enroll the resulting host.
Bare-metal provisioning installs an operating system on registered hardware and then enrolls the host as an Infra node. It requires an owned region with a provisioning network and an approved installation image.
Prepare the provisioning network
An installation administrator configures the boot service, DHCP integration, image repository, and management-controller access. Keep boot traffic on a network restricted to the intended hardware. Existing DHCP infrastructure can provide the boot configuration; Infra must not conflict with another DHCP server on the same segment.
The host must support PXE or iPXE and the selected image's firmware mode. Record its management address, boot-interface MAC address, serial number, region, zone, rack, and intended role in Fleet → Provisioning → Hardware.
Create an installation profile
Open Fleet → Provisioning → Installation profiles → Create profile. Pin an approved OS image and checksum, then choose the firmware mode supported by that artifact. The disk layout identifies actual target devices and where host state will live. A default first disk is not sufficient identification for hardware with replaceable device ordering.
Add management addressing, resolver configuration, and the packages needed for agent registration. The enrollment target pins organization, region, zone, pool, and roles. Use short-lived enrollment authority rather than a long-lived API key embedded in the installation image.
Profile edits create new revisions. An in-flight installation keeps the revision it selected and does not adopt newly edited disk layout halfway through writing.
Pin the image revision and verify its checksum before approval. Changes to a profile create a new revision; a running installation retains the revision it started with. Do not embed long-lived organization API keys in an image.
Provision a host
- Open Fleet → Provisioning → Hardware, select the registered host, and choose Provision.
- Select the installation profile and review device discovery.
- Confirm the target disks and acknowledge that installation overwrites them.
- Enable network boot through the management controller or local firmware.
- Follow boot, download, disk installation, reboot, and enrollment events.
- Complete the node admission checks.
The resulting node remains cordoned until validation succeeds and an operator uncordons it. A completed OS install does not prove that storage replication or workload networking is ready.
Retry a failed installation
A boot failure can leave the existing OS unchanged. A failure after disk writes can leave a partially installed host. Read the last completed stage before retrying; the console reports whether destructive work began.
An image download error requires checking repository access, checksum, and available disk space. An enrollment error requires checking the generated token and control-plane trust configuration. Retrying an enrollment stage should not repeat disk installation unless the host is explicitly reprovisioned.
Reprovision existing nodes
Cordon and drain the node, evacuate storage replicas, and retire its previous identity before selecting a destructive installation. Provisioning an occupied host is blocked while Infra records active workload dependencies. For unreachable hardware, establish that the old workload is stopped before removing those dependencies through the recovery procedure.