Operations (Day 2)Playbooks

Rebooting a machine

View as Markdown

This page describes how to reboot a machine managed by NVIDIA Infra Controller (NICo) (i.e. a managed host or DPU) in any potential state of its lifecycle.

Important note

This is not a site-provider or tenant facing workflow. Rebooting a machine while it is in-use for a tenant can have unexpected side effects. If a tenant requires a reboot, they should use the InvokeInstancePower request - which is properly integrated into the instance lifecycle.*

Reboot Steps

1

Obtain access to nicocli

Configure nicocli for the target REST API. The caller’s organization must have an Infrastructure Provider that owns the Site containing the Machine, and the caller must have the PROVIDER_ADMIN role.

2

Execute the Machine power control operation

Use GracefulRestart when the operating system can shut down cleanly. Use ForceRestart only when a graceful restart is not possible.

$MACHINE_ID='machine-id'
$nicocli machine power-control-machine \
>  --action GracefulRestart \
>  "$MACHINE_ID"

If a graceful restart is not possible, use the forced action explicitly:

$nicocli machine power-control-machine \
>  --action ForceRestart \
>  "$MACHINE_ID"

If the Machine has an attached Instance, acknowledge the workload disruption explicitly:

$nicocli machine power-control-machine \
>  --action GracefulRestart \
>  --acknowledge-attached-instance true \
>  "$MACHINE_ID"

A successful request returns HTTP 202 after NICo accepts the power-control request. The API does not expose a reboot task or terminal reboot status, and nicocli machine get "$MACHINE_ID" can return the same REST lifecycle state before, during, and after the reboot. Confirm that the machine becomes unavailable and returns through the Site’s host or BMC monitoring instead of treating a single Machine status, or a poll of that status, as completion.