Skip to content

add glibc/lib to library search paths - #751

Closed
Hexoplon wants to merge 1 commit into
NVIDIA:mainfrom
Hexoplon:main
Closed

Hexoplon wants to merge 1 commit into
NVIDIA:mainfrom
Hexoplon:main

Conversation

@Hexoplon

Copy link
Copy Markdown

Issue

When running Talos Linux, the Nvidia driver libraries are located under /usr/local/glibc/lib, and the binaries under /usr/local/bin. Using the default search paths, and a driver-root value of /usr/local the validator is able to locate the binaries, but not the libraries.

Proposed solution

Add "glibc/lib" to the library search path. This has been tested to work on a Talos cluster running 1.8.1, production drivers (550.90.07) and H100 GPUs

This PR is part of a set of PRs to better support Talos Linux for the GPU Operator.

Signed-off-by: Zackeus Bengtsson <[email protected]>
@elezar

elezar commented Oct 21, 2024

Copy link
Copy Markdown
Member

@Hexoplon for CDI spec generation we already support specifying --library-search-path for alternative library search paths. Would exposing this through a config option address your use case?

It has recently been pointed out that certain ldcache-based lookups are not working as expected in the latest version. Could it be that this issue is also affecting you? See #735

@elezar

elezar commented Oct 24, 2024

Copy link
Copy Markdown
Member

@Hexoplon please also take a look at the changes included in #755. In oder to address the bug mentioned, this reworks the way the ldcache is handled. Are the libraries you mention in the ldcache, or only present on the filesystem?

@Hexoplon

Copy link
Copy Markdown
Author

@elezar Just noticed that Talos has made changes to the Nvidia extensions to support the GPU operator without requiring modifications, see siderolabs/extensions@c7eb377. This should make this PR irrelevant.

@Hexoplon Hexoplon closed this Oct 28, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants