Skip to content

Add dependency on the GRID license for kubelet #294

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: develop
Choose a base branch
from
Open

Add dependency on the GRID license for kubelet #294

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d42096d
kmod-6.1-nvidia-r570: Add grid-license-check
yeazelm Oct 11, 2025
d621a5e
kmod-6.12-nvidia-r570: Add grid-license-check
yeazelm Oct 11, 2025
0e100c1
kmod-6.12-nvidia-r580: Add grid-license-check
yeazelm Oct 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions packages/kmod-6.1-nvidia-r570/.gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
NVidiaEULAforAWS.pdf
COPYING
*.rpm
NvidiaGridAWSUserLicenseAgreement.DOCX
26 changes: 26 additions & 0 deletions packages/kmod-6.1-nvidia-r570/grid-license-check.service
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
[Unit]
Description=GRID License Check
RefuseManualStart=true
RefuseManualStop=true
DefaultDependencies=no
Before=kubelet.service
After=nvidia-gridd.service
Requires=nvidia-gridd.service

[Service]
Type=oneshot
ExecCondition=/usr/bin/ghostdog match-nvidia-driver grid
# Otherwise, attempt to load the module.
Copy link
Contributor

@arnaldo2792 arnaldo2792 Oct 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm confused, will this line actually load the module or are you just generating output that will be greped later?

Copy link
Contributor Author

@yeazelm yeazelm Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This comment might make more sense above the ExecCondition

Suggested change
# Otherwise, attempt to load the module.
# Confirm GRID is required, then attempt to check the license

ExecStart=/usr/bin/nvidia-smi -q
# Ensure that the stderr file exists. Otherwise, grep fails on an empty file.
ExecStart=-/usr/bin/touch /tmp/.nvidia-gridd-license
Comment on lines +15 to +16
Copy link
Contributor

@bcressey bcressey Oct 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

truncate may be better than touch to reset the file between iterations

(would have to move this before the nvidia-smi call)

Copy link
Contributor Author

@yeazelm yeazelm Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I like that, I'll give it a shot to see if it still gives me the behavior I want with things rearranged to use truncate

# Succeed unless there was a fatal error.
ExecStart=/usr/bin/grep -Fqvzw Unlicensed /tmp/.nvidia-gridd-license
RemainAfterExit=true
StandardOutput=append:/tmp/.nvidia-gridd-license
Restart=on-failure
RestartSec=1
StartLimitBurst=120

[Install]
RequiredBy=kubelet.service
Comment on lines +25 to +26
Copy link
Contributor

@bcressey bcressey Oct 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Below it's required by nvidia-k8s-device-plugin.service which seems more correct.

Copy link
Contributor Author

@yeazelm yeazelm Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, this was originally nvidia-k8s-device-plugin.service and I missed the refactor. In my mind we have several options for what experience we want:

  • Don't prevent anything but log it. Sort of the behavior we have today just without logs (not ideal)
  • Prevent kubelet from starting if the license isn't valid. This means the node never becomes ready and cannot accept any orchestrated work. (This is what I was intending to enforce in the PR)
  • Prevent the device plugin from running, the node will become ready but not advertise GPU resources. It could take some work but not GPU work. (This might be confusing to users so I leaned away from this but was my original approach)
  • Attempt to block boot entirely so the node doesn't even reach multi-user. (Seems harsh but would be fine too).

4 changes: 3 additions & 1 deletion packages/kmod-6.1-nvidia-r570/kmod-6.1-nvidia-r570.spec
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,7 @@ Source206: nvidia-persistenced.service
Source207: fabricmanager.env
Source208: gridd.conf
Source209: nvidia-gridd.service
Source210: grid-license-check.service

# NVIDIA tesla conf files from 300 to 399
Source300: nvidia-tesla-tmpfiles.conf
Expand Down Expand Up @@ -394,7 +395,7 @@ install kernel-open/nvidia-drm.ko %{buildroot}%{_cross_datadir}/nvidia/grid/driv
# Install nvidia-gridd and related files
install -m 755 nvidia-gridd %{buildroot}%{_cross_bindir}/nvidia-gridd
install -m 644 %{S:208} %{buildroot}%{_cross_factorydir}%{_cross_sysconfdir}/nvidia/gridd.conf
install -p -m 0644 %{S:209} %{buildroot}%{_cross_unitdir}
install -p -m 0644 %{S:209} %{S:210} %{buildroot}%{_cross_unitdir}
popd
# End GRID driver
%endif
Expand Down Expand Up @@ -722,6 +723,7 @@ popd
%{_cross_bindir}/nvidia-gridd
%{_cross_factorydir}%{_cross_sysconfdir}/nvidia/gridd.conf
%{_cross_unitdir}/nvidia-gridd.service
%{_cross_unitdir}/grid-license-check.service

%{_cross_datadir}/nvidia/grid/drivers/nvidia.ko
%{_cross_datadir}/nvidia/grid/drivers/nvidia-uvm.ko
Expand Down