[PR PATCH] nvidia-container-toolkit: add nvidia-cdi-hook

[PR PATCH] nvidia-container-toolkit: add nvidia-cdi-hook

[JkktBkkt]

[-- Attachment #1: Type: text/plain, Size: 519 bytes --]

There is a new pull request by JkktBkkt against master on the void-packages repository

https://github.com/JkktBkkt/void-packages nvidia-cdi-hook
https://github.com/void-linux/void-packages/pull/55088

nvidia-container-toolkit: add nvidia-cdi-hook
Fixes #55062 
#### Testing the changes
- I tested the changes in this PR: **briefly**

#### Local build testing
- I built this PR locally for my native architecture, x86_64


A patch file from https://github.com/void-linux/void-packages/pull/55088.patch is attached

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: github-pr-nvidia-cdi-hook-55088.patch --]
[-- Type: text/x-diff, Size: 1167 bytes --]

From 9af0dbbc1dcdc4343d577a4213df35eae618bdee Mon Sep 17 00:00:00 2001
From: JkktBkkt <apkabikov@gmail.com>
Date: Thu, 17 Apr 2025 04:33:26 +0300
Subject: [PATCH] nvidia-container-toolkit: add nvidia-cdi-hook

---
 srcpkgs/nvidia-container-toolkit/template | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/srcpkgs/nvidia-container-toolkit/template b/srcpkgs/nvidia-container-toolkit/template
index a5d724c1f3db3b..6748d41699b64c 100644
--- a/srcpkgs/nvidia-container-toolkit/template
+++ b/srcpkgs/nvidia-container-toolkit/template
@@ -1,13 +1,14 @@
 # Template file for 'nvidia-container-toolkit'
 pkgname=nvidia-container-toolkit
 version=1.16.1
-revision=2
+revision=3
 archs="x86_64"
 build_style=go
 go_import_path=github.com/NVIDIA/nvidia-container-toolkit
 go_package="${go_import_path}/cmd/nvidia-container-runtime-hook
  ${go_import_path}/cmd/nvidia-container-runtime
- ${go_import_path}/cmd/nvidia-ctk"
+ ${go_import_path}/cmd/nvidia-ctk
+ ${go_import_path}/cmd/nvidia-cdi-hook"
 depends="libnvidia-container"
 short_desc="Build and run containers leveraging NVIDIA GPUs"
 maintainer="Quentin Freimanis <quentinfreimanis@gmail.com>"

Re: [PR PATCH] [Updated] nvidia-container-toolkit: add nvidia-cdi-hook

[JkktBkkt]

[-- Attachment #1: Type: text/plain, Size: 524 bytes --]

There is an updated pull request by JkktBkkt against master on the void-packages repository

https://github.com/JkktBkkt/void-packages nvidia-cdi-hook
https://github.com/void-linux/void-packages/pull/55088

nvidia-container-toolkit: add nvidia-cdi-hook
Fixes #55062 
#### Testing the changes
- I tested the changes in this PR: **briefly**

#### Local build testing
- I built this PR locally for my native architecture, x86_64


A patch file from https://github.com/void-linux/void-packages/pull/55088.patch is attached

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: github-pr-nvidia-cdi-hook-55088.patch --]
[-- Type: text/x-diff, Size: 1365 bytes --]

From 2f2dc2203fc92407a790ca83f6cc46cdf0b54922 Mon Sep 17 00:00:00 2001
From: JkktBkkt <apkabikov@gmail.com>
Date: Thu, 17 Apr 2025 04:33:26 +0300
Subject: [PATCH] nvidia-container-toolkit: add nvidia-cdi-hook, fix some tests

---
 srcpkgs/nvidia-container-toolkit/template | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/srcpkgs/nvidia-container-toolkit/template b/srcpkgs/nvidia-container-toolkit/template
index a5d724c1f3db3b..9c13a2d66cac27 100644
--- a/srcpkgs/nvidia-container-toolkit/template
+++ b/srcpkgs/nvidia-container-toolkit/template
@@ -1,13 +1,14 @@
 # Template file for 'nvidia-container-toolkit'
 pkgname=nvidia-container-toolkit
 version=1.16.1
-revision=2
+revision=3
 archs="x86_64"
 build_style=go
 go_import_path=github.com/NVIDIA/nvidia-container-toolkit
 go_package="${go_import_path}/cmd/nvidia-container-runtime-hook
  ${go_import_path}/cmd/nvidia-container-runtime
- ${go_import_path}/cmd/nvidia-ctk"
+ ${go_import_path}/cmd/nvidia-ctk
+ ${go_import_path}/cmd/nvidia-cdi-hook"
 depends="libnvidia-container"
 short_desc="Build and run containers leveraging NVIDIA GPUs"
 maintainer="Quentin Freimanis <quentinfreimanis@gmail.com>"
@@ -23,3 +24,6 @@ post_install() {
 	 --config-file=${DESTDIR}/etc/nvidia-container-runtime/config.toml \
 	 --in-place
 }
+do_check() {
+	PATH+=":${GOPATH}/bin" go test ./...
+}

Re: nvidia-container-toolkit: add nvidia-cdi-hook

[JkktBkkt]

[-- Attachment #1: Type: text/plain, Size: 1814 bytes --]

New comment by JkktBkkt on void-packages repository

https://github.com/void-linux/void-packages/pull/55088#issuecomment-2812076065

Comment:
Apparently this is a duplicate of an earlier pull, #52507 (except the test fix via paths)

A different test errors out when building on my local system, on both current shipped and latest from upstream versions.
I've spent the hours following this submission trying to find how to mitigate that test erroring out (3 of them in TestCreateControlDevices gets nil instead of /dev/nvidia-...), to no avail, errors in the same tests. An upstream [commit](https://github.com/NVIDIA/nvidia-container-toolkit/commit/2529aebd6c26c69b21620bac01ffcd5c04115634) claims to have fixed that, but that is included in latest release and still errors out on my machine, or there's something else going on which I've been unable to figure out. 
Since I run the daily system with the nvidia-open-dkms module, I tested this with the proprietary module as well, just in case. No difference.

Considering that the CI run doesn't error out on the same test my local build does, perhaps this test works well when there's no nv gpu detected in the system and fails when there is one.
For what it's worth, the Gentoo release simply restricts tests and ships as-is.


Update to 1.17.5 requires rebuilding the libnvidia-container to the same version as nvidia-container-toolkit, with a patch for compatibility with go-1.24, which I lifted out of arch's gitlab. I saw no mention of errors in building process of this package on arch but there's also nothing really different about their build process. 
I can submit a separate PR / push (or force-push into this PR?) updated versions into this PR with the 1.17.5 version which also seems to work fine, albeit with only brief testing too. 

Re: nvidia-container-toolkit: add nvidia-cdi-hook

[JkktBkkt]

[-- Attachment #1: Type: text/plain, Size: 265 bytes --]

New comment by JkktBkkt on void-packages repository

https://github.com/void-linux/void-packages/pull/55088#issuecomment-2812292864

Comment:
Re-built both this revision and version 1.17.5 in VM with no nvidia device attached, both did so with tests and no errors.

Re: nvidia-container-toolkit: add nvidia-cdi-hook

[JkktBkkt]

[-- Attachment #1: Type: text/plain, Size: 1832 bytes --]

New comment by JkktBkkt on void-packages repository

https://github.com/void-linux/void-packages/pull/55088#issuecomment-2812076065

Comment:
Apparently this is a duplicate of an earlier pull, #52507 (except the test fix via paths)

A different test errors out when building on my local system, on both current shipped and latest from upstream versions.
I've spent the hours following this submission trying to find how to mitigate that test erroring out (3 of them in TestCreateControlDevices gets nil instead of /dev/nvidia-...), to no avail, errors in the same tests. An upstream [commit](https://github.com/NVIDIA/nvidia-container-toolkit/commit/2529aebd6c26c69b21620bac01ffcd5c04115634) claims to have fixed that, but that is included in latest release and still errors out on my machine, or there's something else going on which I've been unable to figure out. 
Since I run the daily system with the nvidia-open-dkms module, I tested this with the proprietary module as well, just in case. No difference.

Considering that the CI run doesn't error out on the same test my local build does, perhaps this test works well when there's no nv gpu detected in the system and fails when there is one.
For what it's worth, the Gentoo release simply restricts tests and ships as-is.


Update to 1.17.5 requires rebuilding the libnvidia-container to the same version as nvidia-container-toolkit, with a patch for compatibility with go-1.24, which I lifted out of arch's gitlab. I saw no mention of errors in building process of this package on arch but there's also nothing really different about their build process. 
I've ~~can~~ submit a separate PR #55093 ~~/ push (or force-push into this PR?) updated versions into this PR~~ with the 1.17.5 version which also seems to work fine, albeit with only brief testing too. 

Re: [PR PATCH] [Closed]: nvidia-container-toolkit: add nvidia-cdi-hook

[classabbyamp]

[-- Attachment #1: Type: text/plain, Size: 358 bytes --]

There's a closed pull request on the void-packages repository

nvidia-container-toolkit: add nvidia-cdi-hook
https://github.com/void-linux/void-packages/pull/55088

Description:
Fixes #55062 
#### Testing the changes
- I tested the changes in this PR: **briefly**

#### Local build testing
- I built this PR locally for my native architecture, x86_64