Skip to content

docs: clarify GPU CDI hooks and nvidia-container-runtime modes#13383

Open
a7i wants to merge 1 commit into
google:masterfrom
a7i:docs/gpu-cdi-hooks
Open

docs: clarify GPU CDI hooks and nvidia-container-runtime modes#13383
a7i wants to merge 1 commit into
google:masterfrom
a7i:docs/gpu-cdi-hooks

Conversation

@a7i

@a7i a7i commented Jun 6, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Document both nvidia-cdi-hook and legacy nvidia-ctk hook CDI hook forms.
  • Document that disable-device-node-modification is a no-op under nvproxy and is skipped when the hook path is nvidia-cdi-hook; recommend NVIDIA_CDI_HOOK_PATH=/usr/bin/nvidia-cdi-hook for k8s-device-plugin defaults that emit the nvidia-ctk form.
  • Scope the legacy-vs-csv mode note to the nvidia-container-runtime shim only; CDI bypasses the shim and is unaffected.

Test plan

  • Rendered markdown review only; no code changes.

@a7i
a7i marked this pull request as draft June 7, 2026 23:28
Document nvidia-cdi-hook vs nvidia-ctk hook CDI forms, static vs
device-plugin CDI spec sources, and host CDI enablement without
nvidia-ctk cdi generate. Document disable-device-node-modification
skip under nvproxy, NVIDIA_CDI_HOOK_PATH, CDI_FEATURE_FLAGS for
gVisor (disable-ipc-discoverer), and scope the legacy-vs-csv mode note
to the nvidia-container-runtime shim only.
@a7i
a7i force-pushed the docs/gpu-cdi-hooks branch from 475a2c3 to a740617 Compare June 10, 2026 15:27
@a7i
a7i marked this pull request as ready for review June 10, 2026 15:28
Comment thread g3doc/user_guide/gpu.md
When running GPU containers under gVisor, also set
`CDI_FEATURE_FLAGS=disable-ipc-discoverer` on the device plugin. This omits the
`nvidia-persistenced` Unix socket bind mount from the CDI spec; gVisor's gofer
cannot bind-mount that socket and container creation fails if it is present.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Is it true that it fails if it's present? I don't think that's accurate (at least when I tested it). I introduced this flag mainly because I realized it shouldn't be passed through in most cases.

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