kernel upgrade

[gwgrisk]

I have recently been using the GPU operator for deployment, and I found that in the driver container, there is a file KERNEL_UPDATE_HOOK=/run/kernel/postinst.d/update-nvidia-driver.

Its functionality is not explained in the documentation. Can it automatically reinstall the driver during a kernel upgrade without rebooting the node? Previously, when deploying GPU pods on certain points, was there no need to perform eviction actions?

[github-actions]

This issue is stale because it has been open 90 days with no activity. This issue will be closed in 30 days unless new comments are made or the stale label is removed. To skip these checks, apply the "lifecycle/frozen" label.

[shivamerla]

@gwgrisk We don't use that hook in the context of the driver container deployed through GPU Operator. That was used with standalone docker driver containers for testing. The container will build the driver only for the current running kernel, so you would need to reboot the node after kernel upgrade.