move nvidia-ctk hook command into own binary #474

deitch · 2024-04-24T09:25:47Z

Fixes #435

This does two things.

First: creates a new binary called nvidia-cdi-hook, as suggested by @klueska in #435. For backwards compatibility and ease of migration, I kept the functionality in nvidia-ctk as well as in nvidia-cdi-hook, so the following are equal:

$ nvidia-ctk hook create-symlinks
$ nvidia-cdi-hook create-symlinks

To avoid duplication, all of the command code that was under cmd/nvidia-ctk/hook/{create-symlinks,update-ldcache,chmod}/ was moved to cmd/nvidia-cdi-hook/, and cmd/nvidia-ctk/hook/hook.go references those. This keeps the same functionality in place for nvidia-ctk, while avoiding duplication of code.

Second, I changed references in nvidia-ctk cdi generate from putting hooks that call nvidia-ctk hook {subcommand} with nvidia-cdi-generate {subcommand}. I am not convinced I got it correctly, and kept properties that reference nvidiaCTKPath in addition to nvidiaCDIHookPath, just in case.

There are 16 places left where nvidiaCTKPath remains in the code, almost all of them properties or setting options, which means they probably could be removed.

As far as I could tell, and I am not positive, references to nvidia-ctk inside the code are used only for creating hooks. But I am not 100% sure, hence leaving them around. Looking forward to comments.

deitch · 2024-04-24T09:29:48Z

I also am concerned that I missed some of the tests, or that there is a config property that needs to be added.

elezar · 2024-04-24T11:42:39Z

@deitch thanks for the contribution. I'll have a look at it early next week.

lahwaacz · 2024-04-25T06:10:46Z

cmd/nvidia-cdi-hook/README.md

+* `create-symlinks` - Create symlinks inside the directory path to be mounted into a container.
+* `update-ldcache` - Update the dynamic linker cache inside the directory path to be mounted into a container.
+
+The CDI itself is created separately, normally using the `nvidia-ctk cdi` command.


Can you document how to actually use the hook? Is it triggered automatically or users need to configure something?

I don't quite understand. It doesn't change from current behaviour. It's inserted into the cdi yaml when you run nvidia-ctk cdi generate

So, can you document the current behavior?

For me it does not work by default though, nvidia-ctk cdi generate creates a CDI spec where nvidia-ctk is used only in hooks: sections, but it is missing in the mounts: section. Then e.g. when I run podman run --rm -it --gpus all nvidia/cuda:12.4.1-devel-ubuntu22.04, /usr/bin/nvidia-ctk is not available in the container.

So, can you document the current behavior?

Sure, on its way.

For me it does not work by default though, nvidia-ctk cdi generate creates a CDI spec where nvidia-ctk is used only in hooks: sections, but it is missing in the mounts: section. Then e.g. when I run podman run --rm -it --gpus all nvidia/cuda:12.4.1-devel-ubuntu22.04, /usr/bin/nvidia-ctk is not available in the container.

Do you mind posting your CDI here? I have 2 Jetsons, but unfortunately no access to them for the next week, so I have nothing ti generate CDIs on, in order to tests it. I also don't have remote access to anything with proper GPUs, so I am stuck working this through here without proper testing.

FWIW, I see the same behavior on the main branch, but not on the release-1.15 branch. I've compiled the same way from all 3 branches so the issue does not seem related to compiler flags etc.

Oh interesting. So I didn't introduce the bug.

If you've compiled it from main, do you mind generating the CDI yaml from main, so we can compare? It doesn't get us all of the way, but it helps.

Hopefully @klueska or one of the other maintainers will know what we need to do to complete the devices.

@deitch CDI from main: https://gist.github.com/lahwaacz/ca4868a70ca106cb4a90580b6735595e

Thanks @lahwaacz . Those are identical, except for the executable. That is a good sign.

It means that whatever introduced the issue with the busId, is in main, and was not introduced by this PR.

I did a quick check of main vs v1.15.0, there is quite a bit in there that I do not understand. I think we are stuck until someone who gets how this works explains how to fix it in main (and we can apply here).

deitch · 2024-05-01T08:10:42Z

Following up on this. Can someone help us understand what is different in main vs 1.15.0, such that the config it missing devices? How can we fix this?

elezar · 2024-05-03T11:18:18Z

Following up on this. Can someone help us understand what is different in main vs 1.15.0, such that the config it missing devices? How can we fix this?

I was able to reproduce the issue regarding the bus information for the device being empty for main. It seems as if this was related to our changes to the go-nvml wrappers. After updating the go-nvml dependency to v0.12.0-5 this issue no longer presents.

deitch · 2024-05-03T11:26:18Z

Excellent, thank you @elezar . It looks like that already is in, so I should be able to rebase on main and it will work?

deitch · 2024-05-03T11:28:01Z

@lahwaacz I rebased on main, with the fix that @elezar put in. Do you mind running your generation again? Off of v1.15.0, main and this branch, and then posting the 3 in a linked gist, so we can compare?

elezar

@deitch I think I'm ok with the mechanics of this. The one question I have is how to support the transition from the use of nvidia-ctk hook to nvidia-cdi-hook.

Note that we would also need to include the new executable in the deb and rpm package definitions.

elezar · 2024-05-03T12:03:36Z

cmd/nvidia-ctk/cdi/generate/generate.go

-			Name:        "nvidia-ctk-path",
-			Usage:       "Specify the path to use for the nvidia-ctk in the generated CDI specification. If this is left empty, the path will be searched.",
-			Destination: &opts.nvidiaCTKPath,
+			Name:        "nvidia-cdi-hook-path",


Note that this is a breaking change. Does it make sense to add an Alias here instead and then add logic somewhere(TM) for determining whether the nvidia-cdi-hook or nvidia-ctk is used to know how we will generate the CDI spec?

breaking change

Do you mean that someone who might have some command process (human or scripted) that called this with --nvidia-ctk-path will now break, because that option doesn't exist?

I would think (cautiously, and probably wrongly) that it isn't that heavily used an option, and if you install a new version of the toolkit, it is ok for this to break.

If we do not want to allow that, then, sure, I can add an Alias. Your call.

Yes, I know that this may not be the most used option, but I would prefer to either add an alias or add a new option -- possibly marking the old one as deprecated. Given the discussion around a hook constructor below, maybe a separate option is better -- caling out in the usage text that this is intended to replace the nvidia-ctk command.

Sure. Will add it.

What kind of behaviour do we want for that? The deprecation is easy to add to the comment:

&cli.StringFlag{ Name: "nvidia-ctk-path", Usage: "Specify the path to use for the nvidia-ctk in the generated CDI specification. If this is left empty, the path will be searched. DEPRECATED. Use flag --nvidia-cdi-hook-path instead.", },

but when someone uses it, should we:

treat it like they called --nvidia-cdi-hook-path and set that? I don't much like it, as they are distinct binaries.

do nothing with it? That doesn't seem to add value.

save it to somewhere and use it? But we don't use that anywhere.

🤷‍♂️

Note that this ASSUMES that users have not renamed their nvidia-ctk executable for the purpose of injecting the hook

If a user does that, they only have themselves to blame. You always can break something if you try hard enough. If they replace /usr/bin/containerd with a symlink to /usr/sbin/halt, I don't think we can help them 😆

We would include logic to handle both the nvidia-cdi-hook and nvidia-ctk case from the perspective of CDI spec generation. This means that we can add an aliases --nvidia-cdi-hook-path and --nvidia-ctk-path and have them treated consistently

So

if the user uses --nvidia-ctk-path, then it would inject nvidia-ctk hook (2 words) everywhere in CDI

if the user uses --nvidia-cdi-hook-path, then it would inject nvidia-cdi-hook (1 word) everywhere in CDI

if the user uses both, it is an error

if the user uses neither, it uses nvidia-cdi-hook as the default

If I understood you correctly, sure, that works.

Adopted your suggestions right in. Once I saw it inside, totally made sense.

Anything else we need for this?

Does:

diff --git a/cmd/nvidia-ctk/cdi/generate/generate.go b/cmd/nvidia-ctk/cdi/generate/generate.go index fba22301..9f9e994b 100644 --- a/cmd/nvidia-ctk/cdi/generate/generate.go +++ b/cmd/nvidia-ctk/cdi/generate/generate.go @@ -132,13 +132,11 @@ func (m command) build() *cli.Command { Destination: &opts.librarySearchPaths, }, &cli.StringFlag{ - Name: "nvidia-cdi-hook-path", - Usage: "Specify the path to use for the nvidia-cdi-hook in the generated CDI specification. If this is left empty, the path will be searched.", - Destination: &opts.nvidiaCDIHookPath, - }, - &cli.StringFlag{ - Name: "nvidia-ctk-path", - Usage: "Specify the path to use for the nvidia-ctk in the generated CDI specification. If this is left empty, the path will be searched. DEPRECATED. Use flag --nvidia-cdi-hook-path instead.", + Name: "nvidia-cdi-hook-path", + Aliases: []string{"nvidia-ctk-path"}, + Usage: "Specify the path to use for the nvidia-cdi-hook in the generated CDI specification. " + + "If not specified, the PATH will be searched for `nvidia-cdi-hook`. " + + "NOTE: That if this is specified as `nvidia-ctk`, the PATH will be searched for `nvidia-ctk` instead.", Destination: &opts.nvidiaCDIHookPath, }, &cli.StringFlag{

make sense?

internal/discover/graphics.go

deitch · 2024-05-03T12:27:47Z

The one question I have is how to support the transition from the use of nvidia-ctk hook to nvidia-cdi-hook.

Is there what to do for this? Since they are packaged together, if you install and run a newer version of a package with ctk, it installs nvidia-cdi-hook, and nvidia-ctk cdi generate will generate CDIs with nvidia-cdi-hook rather than nvidia-ctk hook. If you update your package, but have a CDI generated by an older version of nvidia-ctk cdi generate, it still will work, because the newer version of nvidia-ctk continues to support nvidia-ctk hook.

Note that we would also need to include the new executable in the deb and rpm package definitions

Is that in here?

elezar · 2024-05-06T09:51:25Z

Is that in here?

Yes. That's the location where the executable should be added (to the nvidia-container-toolkit-base packages).

deitch · 2024-05-06T10:23:23Z

Yes. That's the location where the executable should be added (to the nvidia-container-toolkit-base packages).

OK, I think I got those @elezar ; please take a look.

The only remaining thing is this.

We keep --nvidia-ctk-path <path> around as an option, but what do we do with it?

elezar

Thanks @deitch. It's looking much better already.

There are some minor changes required.

cmd/nvidia-cdi-hook/README.md

elezar · 2024-05-07T13:24:58Z

cmd/nvidia-cdi-hook/README.md

+[`nvidia-ctk cdi`](../nvidia-ctk/) command.
+
+When `nvidia-ctk cdi generate` is run, the CDI specification is generated as a yaml file.
+The CDI file provides instructions for a container runtime to set up devices, files and


Suggested change

The CDI file provides instructions for a container runtime to set up devices, files and

The CDI specification provides instructions for a container runtime to set up devices, files and

cmd/nvidia-cdi-hook/README.md

cmd/nvidia-cdi-hook/main.go

elezar · 2024-05-07T13:38:04Z

cmd/nvidia-ctk/cdi/generate/generate.go

-			Name:        "nvidia-ctk-path",
-			Usage:       "Specify the path to use for the nvidia-ctk in the generated CDI specification. If this is left empty, the path will be searched.",
-			Destination: &opts.nvidiaCTKPath,
+			Name:        "nvidia-cdi-hook-path",


Does:

diff --git a/cmd/nvidia-ctk/cdi/generate/generate.go b/cmd/nvidia-ctk/cdi/generate/generate.go index fba22301..9f9e994b 100644 --- a/cmd/nvidia-ctk/cdi/generate/generate.go +++ b/cmd/nvidia-ctk/cdi/generate/generate.go @@ -132,13 +132,11 @@ func (m command) build() *cli.Command { Destination: &opts.librarySearchPaths, }, &cli.StringFlag{ - Name: "nvidia-cdi-hook-path", - Usage: "Specify the path to use for the nvidia-cdi-hook in the generated CDI specification. If this is left empty, the path will be searched.", - Destination: &opts.nvidiaCDIHookPath, - }, - &cli.StringFlag{ - Name: "nvidia-ctk-path", - Usage: "Specify the path to use for the nvidia-ctk in the generated CDI specification. If this is left empty, the path will be searched. DEPRECATED. Use flag --nvidia-cdi-hook-path instead.", + Name: "nvidia-cdi-hook-path", + Aliases: []string{"nvidia-ctk-path"}, + Usage: "Specify the path to use for the nvidia-cdi-hook in the generated CDI specification. " + + "If not specified, the PATH will be searched for `nvidia-cdi-hook`. " + + "NOTE: That if this is specified as `nvidia-ctk`, the PATH will be searched for `nvidia-ctk` instead.", Destination: &opts.nvidiaCDIHookPath, }, &cli.StringFlag{

make sense?

elezar · 2024-05-07T13:48:00Z

pkg/nvcdi/lib.go

@@ -49,6 +49,7 @@ type nvcdilib struct {
 	driverRoot         string
 	devRoot            string
 	nvidiaCTKPath      string
+	nvidiaCDIHookPath  string


Since we're effectively replacing nvidiaCTKPath with this member, I would recommend applying the following diff:

diff --git a/internal/modifier/cdi.go b/internal/modifier/cdi.go index c53cb996..c5af4f88 100644 --- a/internal/modifier/cdi.go +++ b/internal/modifier/cdi.go @@ -185,7 +185,7 @@ func newAutomaticCDISpecModifier(logger logger.Interface, cfg *config.Config, de func generateAutomaticCDISpec(logger logger.Interface, cfg *config.Config, devices []string) (spec.Interface, error) { cdilib, err := nvcdi.New( nvcdi.WithLogger(logger), - nvcdi.WithNVIDIACTKPath(cfg.NVIDIACTKConfig.Path), + nvcdi.WithNVIDIACDIHookPath(cfg.NVIDIACTKConfig.Path), nvcdi.WithDriverRoot(cfg.NVIDIAContainerCLIConfig.Root), nvcdi.WithVendor("runtime.nvidia.com"), nvcdi.WithClass("gpu"), diff --git a/internal/modifier/csv.go b/internal/modifier/csv.go index 54304428..0905d5da 100644 --- a/internal/modifier/csv.go +++ b/internal/modifier/csv.go @@ -62,7 +62,7 @@ func NewCSVModifier(logger logger.Interface, cfg *config.Config, image image.CUD cdilib, err := nvcdi.New( nvcdi.WithLogger(logger), nvcdi.WithDriverRoot(cfg.NVIDIAContainerCLIConfig.Root), - nvcdi.WithNVIDIACTKPath(cfg.NVIDIACTKConfig.Path), + nvcdi.WithNVIDIACDIHookPath(cfg.NVIDIACTKConfig.Path), nvcdi.WithMode(nvcdi.ModeCSV), nvcdi.WithCSVFiles(csvFiles), ) diff --git a/internal/platform-support/tegra/csv_test.go b/internal/platform-support/tegra/csv_test.go index 2e4aca34..2e8e42fe 100644 --- a/internal/platform-support/tegra/csv_test.go +++ b/internal/platform-support/tegra/csv_test.go @@ -187,7 +187,6 @@ func TestDiscovererFromCSVFiles(t *testing.T) { o := tegraOptions{ logger: logger, - nvidiaCTKPath: "/usr/bin/nvidia-ctk", nvidiaCDIHookPath: "/usr/bin/nvidia-cdi-hook", csvFiles: []string{"dummy"}, ignorePatterns: tc.ignorePatterns, diff --git a/internal/platform-support/tegra/tegra.go b/internal/platform-support/tegra/tegra.go index 597d8b16..1031fc72 100644 --- a/internal/platform-support/tegra/tegra.go +++ b/internal/platform-support/tegra/tegra.go @@ -30,7 +30,6 @@ type tegraOptions struct { csvFiles []string driverRoot string devRoot string - nvidiaCTKPath string nvidiaCDIHookPath string ldconfigPath string librarySearchPaths []string @@ -134,13 +133,6 @@ func WithCSVFiles(csvFiles []string) Option { } } -// WithNVIDIACTKPath sets the path to the nvidia-container-toolkit binary. -func WithNVIDIACTKPath(nvidiaCTKPath string) Option { - return func(o *tegraOptions) { - o.nvidiaCTKPath = nvidiaCTKPath - } -} - // WithNVIDIACDIHookPath sets the path to the nvidia-cdi-hook binary. func WithNVIDIACDIHookPath(nvidiaCDIHookPath string) Option { return func(o *tegraOptions) { diff --git a/pkg/nvcdi/lib-csv.go b/pkg/nvcdi/lib-csv.go index f6f652ba..649b801a 100644 --- a/pkg/nvcdi/lib-csv.go +++ b/pkg/nvcdi/lib-csv.go @@ -44,7 +44,6 @@ func (l *csvlib) GetAllDeviceSpecs() ([]specs.Device, error) { tegra.WithLogger(l.logger), tegra.WithDriverRoot(l.driverRoot), tegra.WithDevRoot(l.devRoot), - tegra.WithNVIDIACTKPath(l.nvidiaCTKPath), tegra.WithNVIDIACDIHookPath(l.nvidiaCDIHookPath), tegra.WithLdconfigPath(l.ldconfigPath), tegra.WithCSVFiles(l.csvFiles), diff --git a/pkg/nvcdi/lib.go b/pkg/nvcdi/lib.go index 31e08e82..c56c18e1 100644 --- a/pkg/nvcdi/lib.go +++ b/pkg/nvcdi/lib.go @@ -48,7 +48,6 @@ type nvcdilib struct { deviceNamers DeviceNamers driverRoot string devRoot string - nvidiaCTKPath string nvidiaCDIHookPath string ldconfigPath string configSearchPaths []string @@ -88,9 +87,6 @@ func New(opts ...Option) (Interface, error) { if l.devRoot == "" { l.devRoot = l.driverRoot } - if l.nvidiaCTKPath == "" { - l.nvidiaCTKPath = "/usr/bin/nvidia-ctk" - } if l.nvidiaCDIHookPath == "" { l.nvidiaCDIHookPath = "/usr/bin/nvidia-cdi-hook" } diff --git a/pkg/nvcdi/options.go b/pkg/nvcdi/options.go index a90450f6..5a490619 100644 --- a/pkg/nvcdi/options.go +++ b/pkg/nvcdi/options.go @@ -63,10 +63,10 @@ func WithLogger(logger logger.Interface) Option { } // WithNVIDIACTKPath sets the path to the NVIDIA Container Toolkit CLI path for the library +// +// Deprecated: Use WithNVIDIACDIHookPath instead. func WithNVIDIACTKPath(path string) Option { - return func(l *nvcdilib) { - l.nvidiaCTKPath = path - } + return WithNVIDIACDIHookPath(path) } // WithNVIDIACDIHookPath sets the path to the NVIDIA Container Toolkit CLI path for the library

Which switches to using nvidiaCDIHookPath everywhere -- except where we read nvidia-ctk.path config optionand pass that to the constructor. We also mark thenvcdi.WithNVIDIACTKPath` option as deprecated (although we could also not do so and just set the member).

The internal usage in the tegra package has also been changed to directly refer to nvidiaCDIHookPath instead. I don't think we need to deprecat that option since we don't consume this in any of our othe components.

make sense?

Sure, a clean alias, and an explanation. Added.

packaging/rpm/SPECS/nvidia-container-toolkit.spec

deitch · 2024-05-07T15:44:10Z

All comments and suggestions added; thanks for them @elezar . Back to you.

deitch · 2024-05-14T16:24:23Z

Is anything else needed for this @elezar ?

deitch · 2024-05-14T17:37:57Z

Was a golang unit test error. Fixed.

elezar

In addition to the comment below, we also need:

diff --git a/internal/discover/ldconfig_test.go b/internal/discover/ldconfig_test.go
index 67149845..0b214c77 100644
--- a/internal/discover/ldconfig_test.go
+++ b/internal/discover/ldconfig_test.go
@@ -25,8 +25,8 @@ import (
 )
 
 const (
-	testNvidiaCDIPath = "/foo/bar/nvidia-cdi-path"
-	testLdconfigPath  = "/bar/baz/ldconfig"
+	testNvidiaCDIHookPath = "/foo/bar/nvidia-cdi-hook"
+	testLdconfigPath      = "/bar/baz/ldconfig"
 )
 
 func TestLDCacheUpdateHook(t *testing.T) {
@@ -92,12 +92,12 @@ func TestLDCacheUpdateHook(t *testing.T) {
 				},
 			}
 			expectedHook := Hook{
-				Path:      testNvidiaCDIPath,
+				Path:      testNvidiaCDIHookPath,
 				Args:      tc.expectedArgs,
 				Lifecycle: "createContainer",
 			}
 
-			d, err := NewLDCacheUpdateHook(logger, mountMock, testNvidiaCDIPath, tc.ldconfigPath)
+			d, err := NewLDCacheUpdateHook(logger, mountMock, testNvidiaCDIHookPath, tc.ldconfigPath)
 			require.NoError(t, err)
 
 			hooks, err := d.Hooks()

elezar · 2024-05-14T17:46:25Z

internal/config/config.go

@@ -186,6 +187,22 @@ func ResolveNVIDIACTKPath(logger logger.Interface, nvidiaCTKPath string) string
 	)
 }

+// ResolveNVIDIACDIHookPath resolves the path to the nvidia-cdi-hook binary.


Looking at this and ResolveNVIDIACTKPath above again, what do you think of rewriting this as:

// ResolveNVIDIACTKPath resolves the path to the nvidia-ctk binary. // This executable is used in hooks and needs to be an absolute path. // If the path is specified as an absolute path, it is used directly // without checking for existence of an executable at that path. // // Deprecated: Use ResolveNVIDIACDIHookPath directly instead. func ResolveNVIDIACTKPath(logger logger.Interface, nvidiaCTKPath string) string { return resolveWithDefault( logger, "NVIDIA Container Toolkit CLI", nvidiaCTKPath, nvidiaCTKDefaultFilePath, ) } // ResolveNVIDIACDIHookPath resolves the path to the nvidia-cdi-hook binary. // This executable is used in hooks and needs to be an absolute path. // If the path is specified as an absolute path, it is used directly // without checking for existence of an executable at that path. func ResolveNVIDIACDIHookPath(logger logger.Interface, nvidiaCDIHookPath string) string { if filepath.Base(nvidiaCDIHookPath) == "nvidia-ctk" { return resolveWithDefault( logger, "NVIDIA Container Toolkit CLI", nvidiaCDIHookPath, nvidiaCTKDefaultFilePath, ) } return resolveWithDefault( logger, "NVIDIA CDI Hook CLI", nvidiaCDIHookPath, nvidiaCDIHookDefaultFilePath, ) }

and then adding:

//nolint:staticcheck // TODO(elezar): We should swith the nvidia-container-runtime from using nvidia-ctk to using nvidia-cdi-hook. cfg.NVIDIACTKConfig.Path = config.ResolveNVIDIACTKPath(r.logger, cfg.NVIDIACTKConfig.Path)

where we call ResolveNVIDIACTKPath?

We can work on properly handling the config changes for the runtime in a follow-up.

I like it. Change made and pushed.

deitch · 2024-05-14T17:56:22Z

In addition to the comment below, we also need:

Yup already working on that one. Since it causes CI to fail, I get to see it. :-)

elezar · 2024-05-17T13:41:14Z

internal/config/config.go

+// This executable is used in hooks and needs to be an absolute path.
+// If the path is specified as an absolute path, it is used directly
+// without checking for existence of an executable at that path.
+func ResolveNVIDIACDIHookPath(logger logger.Interface, nvidiaCDIHookPath string) string {


In responding to cncf-tags/container-device-interface#203 (comment) I had the thought that the concept of a CDI Hook executable could be standardized so that the nvidia-cdi-hook is a vendor-specific implementation. Does that mean that we should use cdiHookPath instead of using nvidiaCDIHookPath for our variables and config values?

I don't think it matters all that much. Until such time as that actually happens, might as well get this in and iterate on it?

Sure. It can definitely be done as a follow-up.

internal/config/config.go

elezar · 2024-05-17T14:01:51Z

One final pass @deitch. I have created deitch#1 with a proposal for a changelog addition and also adding a constructor for the hook commands to ensure that these are consistent across the two binaries.

I think with those in we can wrap this up.

deitch · 2024-05-17T14:07:40Z

@elezar do you mind changing deitch#1 to be a PR against the branch I used for this PR separate-hook-binary rather than against deitch:main? That way I can see just the changed files there and merge it into my branch?

elezar · 2024-05-17T14:52:56Z

@deitch updated.

deitch · 2024-05-17T15:01:01Z

Merged them in. Should probably kick off CI again here.

This change creates an nvidia-cdi-hook binary for implementing CDI hooks. This allows for these hooks to be separated from the nvidia-ctk command which may, for example, require libnvidia-ml to support other functionality. The nvidia-ctk hook subcommand is maintained as an alias for the time being to allow for existing CDI specifications referring to this path to work as expected. Signed-off-by: Avi Deitcher <avi@deitcher.net>

elezar

Thanks for the work on this @deitch.

I think this is cleaner than having the hooks as a subcommand of the nvidia-ctk -- especially given the discussion around cncf-tags/container-device-interface#203

deitch · 2024-05-21T10:57:56Z

Thanks for the work on this @deitch

My pleasure. I am glad it worked out.

deitch force-pushed the separate-hook-binary branch from ab99f33 to 41cc661 Compare April 24, 2024 09:32

elezar self-assigned this Apr 24, 2024

lahwaacz reviewed Apr 25, 2024

View reviewed changes

deitch force-pushed the separate-hook-binary branch from 41cc661 to bd2f3cc Compare April 25, 2024 08:08

deitch mentioned this pull request May 3, 2024

Consider separating nvidia-ctk hook from other functions? #435

Closed

deitch force-pushed the separate-hook-binary branch from 61554a6 to af030d0 Compare May 3, 2024 11:27

elezar requested changes May 3, 2024

View reviewed changes

deitch force-pushed the separate-hook-binary branch from af030d0 to 17a05d2 Compare May 6, 2024 10:21

deitch force-pushed the separate-hook-binary branch from 17a05d2 to 7f2cf83 Compare May 6, 2024 13:35

elezar requested changes May 7, 2024

View reviewed changes

deitch force-pushed the separate-hook-binary branch 2 times, most recently from 9fc3845 to cdc878b Compare May 7, 2024 15:43

deitch force-pushed the separate-hook-binary branch from cdc878b to bad12a5 Compare May 14, 2024 17:37

deitch force-pushed the separate-hook-binary branch from bad12a5 to a647926 Compare May 14, 2024 17:49

elezar requested changes May 14, 2024

View reviewed changes

deitch force-pushed the separate-hook-binary branch from a647926 to 940579d Compare May 14, 2024 17:56

deitch force-pushed the separate-hook-binary branch from 940579d to 764209a Compare May 14, 2024 17:58

elezar mentioned this pull request May 17, 2024

Add support for predefined hooks cncf-tags/container-device-interface#203

Open

elezar reviewed May 17, 2024

View reviewed changes

internal/config/config.go Outdated Show resolved Hide resolved

deitch force-pushed the separate-hook-binary branch from 764209a to 0b65105 Compare May 17, 2024 13:44

elezar force-pushed the separate-hook-binary branch from 667881d to 179d865 Compare May 21, 2024 10:19

elezar approved these changes May 21, 2024

View reviewed changes

elezar merged commit 7b988f1 into NVIDIA:main May 21, 2024
8 checks passed

deitch deleted the separate-hook-binary branch May 21, 2024 10:57

	The CDI file provides instructions for a container runtime to set up devices, files and
	The CDI specification provides instructions for a container runtime to set up devices, files and

move nvidia-ctk hook command into own binary #474

move nvidia-ctk hook command into own binary #474

Conversation

deitch commented Apr 24, 2024

deitch commented Apr 24, 2024

elezar commented Apr 24, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lahwaacz Apr 25, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

deitch commented May 1, 2024

elezar commented May 3, 2024

deitch commented May 3, 2024

deitch commented May 3, 2024

elezar left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

deitch commented May 3, 2024

elezar commented May 6, 2024

deitch commented May 6, 2024

elezar left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

deitch commented May 7, 2024

deitch commented May 14, 2024

deitch commented May 14, 2024

elezar left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

deitch commented May 14, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

elezar commented May 17, 2024

deitch commented May 17, 2024

elezar commented May 17, 2024

deitch commented May 17, 2024

elezar left a comment

Choose a reason for hiding this comment

deitch commented May 21, 2024

lahwaacz Apr 25, 2024 •

edited