Possibility to use host `/etc/apt/apt.conf.d/*` during driver installation?

[skirsten]

The existing driver.repoConfig.configMapName makes it possible to use a private mirror or cache.
But its not very portable and I would much rather use the host's existing /etc/apt/apt.conf.d/01proxy.

Acquire::HTTP::Proxy "http://10.0.0.5:8080";
Acquire::HTTPS::Proxy "false";

Would it be possible to either

1. supply custom volumes to the driver values to mount this file into the container
2. supply a custom HTTP_PROXY env that is only valid for the apt requests
3. ... any other ideas?

[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.

[rahulait]

We don't expose mounts for the driver because the driver pod is tightly controlled by the operator, and arbitrary mounts could break the Driver Toolkit compatibility layer, which must remain stable for kernel module builds.

APT honors http_proxy and https_proxy env vars, so you can set the env vars for nvidia-driver daemonset and it will get applied. The drawback is that these vars are not limited to APT, so it can interfere with other operations and can break things unknowingly.

If you really want custom volumes, you can have a mutating webhook which can update the applied nvidia-driver-daemonset config to have custom volume which mounts host's /etc/apt/apt.conf.d/01proxy into the nvidia-driver container. However, we don't provide that solution and its upto the user how they want to set the mutating webhook to achieve this purpose.

Another option is to build custom gpu-operator image with updates to rendering logic and adding additional mount volumes.

[skirsten]

Thanks for the detailed response.