An NVIDIA card with no driver is reported, not onboarded as a CPU node
CI / test (push) Failing after 9m12s

The agent found GPUs through nvidia-smi alone, so a machine whose driver
was never installed looked exactly like one with no card and joined
quietly as a CPU node. Only init looked at the PCI bus.

The host discovery now reads the bus too (vendor 0x10de, class 0x03) and
carries it to the control plane. The doctor fails the pre-install checks
on a card without a driver, with the install and the reboot named, and
tells a driver that is installed but not loaded (Secure Boot, nouveau)
apart from a missing one. A node already in shows the warning in ctl
nodes and the console. init uses the same check.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Ilker ManapandClaude Fable 5.1 committed 2026-11-19 12:29:02 +01:00
1 parent a53b2f4be9
commit c2fedeb326
12 files changed
+138 -26

No files matched your search

+6
View File
@@ -181,6 +181,12 @@ Both default to the usual places. Containerized driver installs routinely put
`nvidia-smi` somewhere else, and an agent that cannot find it registers with no
GPUs — which looks exactly like a node that has none.
A node with no driver at all is told apart from one with no card: the PCI bus
is read as well, and an NVIDIA card there with no driver loaded is a failure in
the checks before the agent is installed (what to install, and that a reboot
follows), and a warning on the node in `llmgw ctl nodes` and the console
afterwards. It serves on CPU until the driver is in and its agent restarted.
## Removing a node
```