Migrate reconfigure to golang #216

elezar · 2025-07-08T14:27:31Z

This change builds on the proposal from NVIDIA/vgpu-device-manager#93 and migrates the reconfigure-mig.sh script to golang.

Ideally I would like to still do the following:

factor this out so that we can import a package from the vgpu-device-manager
replace the nvidia-mig-parted CLI invocations with direct API calls

but these can be done as follow-ups.

elezar · 2025-07-09T14:36:37Z

cmd/nvidia-mig-manager/reconfigure_mig.go

This is adapted from NVIDIA/vgpu-device-manager#93

Copilot

Pull Request Overview

This PR replaces the existing reconfigure_mig.sh script with a Go-based implementation, centralizing MIG reconfiguration logic and Kubernetes client interactions into a single binary.

Updated module dependencies (added validator, container-toolkit, and other indirect modules).
Introduced toolkit.go for device-node and nvidia-smi wrapper functions.
Ported the full MIG reconfiguration workflow into Go (reconfigure_mig.go, main.go, clients.go, cdi.go).

Reviewed Changes

Copilot reviewed 6 out of 411 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
go.mod	Upgraded NVIDIA libraries and added validator & container-toolkit
cmd/nvidia-mig-manager/toolkit.go	Implemented `createControlDeviceNodes` & `runNvidiaSMI` wrappers
cmd/nvidia-mig-manager/reconfigure_mig.go	Ported core MIG reconfiguration logic from shell to Go
cmd/nvidia-mig-manager/main.go	Refactored CLI flags into `options` struct and wired Go flow
cmd/nvidia-mig-manager/clients.go	Added Kubernetes client abstractions for stopping/restarting pods
cmd/nvidia-mig-manager/cdi.go	Added CDI spec regeneration via `nvcdi` library

Comments suppressed due to low confidence (2)

cmd/nvidia-mig-manager/toolkit.go:22

New functionality in createControlDeviceNodes lacks unit tests. Consider adding tests to verify the command arguments and execution behavior.

func createControlDeviceNodes(opts *reconfigureMIGOptions) error {

cmd/nvidia-mig-manager/reconfigure_mig.go:43

[nitpick] Global variable hostGPUClientServicesStopped can lead to shared state across calls. Consider moving it into a struct or using a local variable to improve encapsulation.

	hostGPUClientServicesStopped []string

Copilot · 2025-07-10T09:13:02Z

cmd/nvidia-mig-manager/toolkit.go

+
+func runNvidiaSMI(opts *reconfigureMIGOptions) error {
+	if opts.DriverRootCtrPath == opts.DevRootCtrPath {
+		cmd := exec.Command("chroot", opts.NVIDIASMIPath) //nolint:gosec


The chroot invocation is missing the root directory argument; chroot requires both the new root path and the command to run. Consider using exec.Command("chroot", opts.HostRootMount, opts.NVIDIASMIPath) or invoking the binary directly when no chroot is needed.

Suggested change

cmd := exec.Command("chroot", opts.NVIDIASMIPath) //nolint:gosec

cmd := exec.Command("chroot", opts.HostRootMount, opts.NVIDIASMIPath) //nolint:gosec

Copilot · 2025-07-10T09:13:03Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+	// ConfigStateLabel is the label that is used to indicate the state of the config being applied.
+	ConfigStateLabel string
+
+	// DriverLibrayPath is the path to libnvidia-ml.so.1 in the container.


Typo in comment: 'DriverLibrayPath' should be 'DriverLibraryPath'.

Suggested change

// DriverLibrayPath is the path to libnvidia-ml.so.1 in the container.

// DriverLibraryPath is the path to libnvidia-ml.so.1 in the container.

Copilot · 2025-07-10T09:13:03Z

cmd/nvidia-mig-manager/cdi.go

+
+	spec, err := cdilib.GetSpec()
+	if err != nil {
+		return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)


Typo in error message: 'genereate' should be 'generate'.

Suggested change

return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)

return fmt.Errorf("failed to generate CDI spec for management containers: %v", err)

Copilot · 2025-07-10T09:13:03Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+	}
+
+	cmd = exec.Command("chroot", opts.HostRootMount, "systemctl", "-q", "is-enabled", service) // #nosec G204 -- HostRootMount validated via dirpath, service validated via systemd_service_name.
+	if err := cmd.Run(); err != nil {


[nitpick] The service-enabled check (is-enabled) is called twice in this function. Consider caching the result to avoid redundant system calls.

Suggested change

if err := cmd.Run(); err != nil {

isEnabledErr := cmd.Run()

if isEnabledErr != nil {

ArangoGutierrez

First pass, looks good

ArangoGutierrez · 2025-07-10T09:21:10Z

cmd/nvidia-mig-manager/reconfigure_mig.go

@@ -0,0 +1,452 @@
+/*


Since this file is simply adapted, not much to say besides the Copilot comments

ArangoGutierrez · 2025-07-10T09:27:28Z

cmd/nvidia-mig-manager/reconfigure_mig.go

 func reconfigureMIG(clientset *kubernetes.Clientset, opts *reconfigureMIGOptions) error {
-	validate := validator.New(validator.WithRequiredStructEnabled())
-
 	log.Info("Validating reconfigure MIG options")


Shouldn't we move this log to the Validate function itself? That way, when called from other locations (if) the logging will also be consistent

ArangoGutierrez · 2025-07-10T09:35:09Z

cmd/nvidia-mig-manager/clients.go

+	Restart(*kubernetes.Clientset) error
+}
+
+// An operand is a GPU client that is controlled by a deploy label.


It was my understanding that an Operand is a deployment managed/owned by a controller (operator)

Sure, but the relevant information here is that it is something that uses the GPU. This is why it needs to be shut down.

ArangoGutierrez · 2025-07-10T09:45:32Z

cmd/nvidia-mig-manager/clients.go

+}
+
+func stopK8sClients(clientset *kubernetes.Clientset) ([]k8sClient, error) {
+	// TODO: We need to add this namespace to the options.


ArangoGutierrez · 2025-07-10T09:50:41Z

cmd/nvidia-mig-manager/clients.go

+	var k8sGPUClients []k8sClient
+
+	// We first optionally stop the operands managed by the operator:
+	var operands = []*operand{


For now hard coded is ok, but should we query the API server filtered with label "app=gpu-operator" and grab all the current operands managed by the operator? that way we don't have to manage this list

For the first iteration I wanted to mirror what is being done in reconfigure_mig.sh. Once we have a stable go-based implementation we have much more flexibility.

Also note that the mig manager itself should NOT be stopped in this was as it is busy performing the required updated. The NVIDIA Container Toolkit container is also NOT a GPU client and doesn't need to be stopped.

ArangoGutierrez · 2025-07-10T09:55:40Z

cmd/nvidia-mig-manager/cdi.go

+
+func regenerateCDISpec(opts *reconfigureMIGOptions) error {
+	log.Info("Generating CDI spec for management containers")
+	cdilib, err := nvcdi.New(


the new API 😍

This isn't the new API. This API has existed for as long as the nvcdi package has existed. The "New" API was just cleaning up some internals and REMOVING functions that were largely unused.

ArangoGutierrez · 2025-07-10T09:56:42Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+	// ConfigStateLabel is the label that is used to indicate the state of the config being applied.
+	ConfigStateLabel string
+
+	// DriverLibrayPath is the path to libnvidia-ml.so.1 in the container.


ArangoGutierrez · 2025-07-10T09:57:57Z

cmd/nvidia-mig-manager/main.go

 func main() {
+	o := &options{}
+
 	c := cli.NewApp()


Note to self: this is a candidate for V3 migration

ArangoGutierrez · 2025-07-10T09:59:00Z

cmd/nvidia-mig-manager/clients.go

-	nodename  string
-	app       string
+	app string
+	// manager stores a reference to the mig manager managing these clients.


"A manager stores a reference to the manager handling these clients."

ArangoGutierrez · 2025-07-10T09:59:58Z

cmd/nvidia-mig-manager/clients.go

 	k8sClient
 }

+// A pod represents a kubernetes pod with a specified app= label.


Note that in the type above, the "a" is lower case; we should do all lower or all upper, for consistency

jgehrcke · 2025-07-10T11:34:51Z

migrates the reconfigure_mig.sh script to golang.

Where can I find reconfigure_mig.sh?

ArangoGutierrez · 2025-07-10T12:17:09Z

migrates the reconfigure_mig.sh script to golang.

Where can I find reconfigure_mig.sh?

https://github.com/NVIDIA/mig-parted/blob/main/deployments/container/reconfigure-mig.sh

tariq1890 · 2025-07-09T18:46:55Z

cmd/nvidia-mig-manager/cdi.go

+
+	spec, err := cdilib.GetSpec()
+	if err != nil {
+		return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)


Suggested change

return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)

return fmt.Errorf("failed to genereate CDI spec for management containers: %w", err)

tariq1890 · 2025-07-09T18:48:18Z

cmd/nvidia-mig-manager/cdi.go

+		nvcdi.WithClass("gpu"),
+	)
+	if err != nil {
+		return fmt.Errorf("failed to create CDI library for management containers: %v", err)


Suggested change

return fmt.Errorf("failed to create CDI library for management containers: %v", err)

return fmt.Errorf("failed to create CDI library for management containers: %w", err)

tariq1890 · 2025-07-09T18:48:33Z

cmd/nvidia-mig-manager/cdi.go

+		transformroot.WithTargetDevRoot(opts.DevRoot),
+	)
+	if err := transformer.Transform(spec.Raw()); err != nil {
+		return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)


Suggested change

return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)

return fmt.Errorf("failed to transform driver root in CDI spec: %w", err)

tariq1890 · 2025-07-09T18:48:42Z

cmd/nvidia-mig-manager/cdi.go

+	}
+	err = spec.Save("/var/run/cdi/management.nvidia.com-gpu.yaml")
+	if err != nil {
+		return fmt.Errorf("failed to save CDI spec for management containers: %v", err)


Suggested change

return fmt.Errorf("failed to save CDI spec for management containers: %v", err)

return fmt.Errorf("failed to save CDI spec for management containers: %w", err)

tariq1890 · 2025-07-09T18:50:22Z

cmd/nvidia-mig-manager/clients.go

+
+func (o *pod) delete() error {
+	err := o.manager.clientset.CoreV1().Pods(o.manager.Namespace).DeleteCollection(
+		context.TODO(),


Let's not use context.TODO(). Can we pass the application's primary context and modify this method's signature to include ctx as a param?

tariq1890 · 2025-07-11T01:42:08Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+func (m *migManager) getNodeLabelValue(label string) (string, error) {
+	node, err := m.clientset.CoreV1().Nodes().Get(context.TODO(), m.NodeName, metav1.GetOptions{})
+	if err != nil {
+		return "", fmt.Errorf("unable to get node object: %v", err)


Suggested change

return "", fmt.Errorf("unable to get node object: %v", err)

return "", fmt.Errorf("unable to get node object: %w", err)

tariq1890 · 2025-07-11T01:42:40Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+}
+
+func (m *migManager) getNodeLabelValue(label string) (string, error) {
+	node, err := m.clientset.CoreV1().Nodes().Get(context.TODO(), m.NodeName, metav1.GetOptions{})


Let's propagate ctx down the method call stack

tariq1890 · 2025-07-11T01:42:50Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+func (m *migManager) setNodeLabelValue(label, value string) error {
+	node, err := m.clientset.CoreV1().Nodes().Get(context.TODO(), m.NodeName, metav1.GetOptions{})
+	if err != nil {
+		return fmt.Errorf("unable to get node object: %v", err)


Suggested change

return fmt.Errorf("unable to get node object: %v", err)

return fmt.Errorf("unable to get node object: %w", err)

tariq1890 · 2025-07-11T01:42:58Z

cmd/nvidia-mig-manager/reconfigure_mig.go

+	node.SetLabels(labels)
+	_, err = m.clientset.CoreV1().Nodes().Update(context.TODO(), node, metav1.UpdateOptions{})
+	if err != nil {
+		return fmt.Errorf("unable to update node object: %v", err)


Suggested change

return fmt.Errorf("unable to update node object: %v", err)

return fmt.Errorf("unable to update node object: %w", err)

jgehrcke · 2025-07-11T09:39:04Z

migrates the reconfigure_mig.sh script to golang.

About my: "Where can I find reconfigure_mig.sh?" Thanks @ArangoGutierrez -- I was searching quite wildly for reconfigure_mig.sh, but this was an underscore vs. hyphen problem.

Permanent URL: https://github.com/NVIDIA/mig-parted/blob/15f09c791454833c44a78f5b44a1bcd6a62ec6d0/deployments/container/reconfigure-mig.sh

cdesiniotis

First pass. I have not reviewed everything in detail yet.

cdesiniotis · 2025-08-14T17:06:35Z

cmd/nvidia-mig-manager/main.go

-	}
-	if withRebootFlag {
-		args = append(args, "-r")
+		options = append(options,


Question -- do we actually need the conditional here? Or can we always add these options to the options slice?

cdesiniotis · 2025-08-14T17:07:03Z

cmd/nvidia-mig-manager/main.go

-	if withShutdownHostGPUClientsFlag {
-		args = append(args, "-d")
+
+	options = append(options,


Question -- is there any value in appending these options here? Can we add them to the list of options during initialization of the options slice?

cdesiniotis · 2025-08-14T17:10:17Z

pkg/mig/reconfigure/api.go

nit: this file is missing a license header

cdesiniotis · 2025-08-14T17:17:15Z

pkg/mig/reconfigure/options.go

+	DevRoot string
+
+	CDIEnabled    bool
+	NVIDIASMIPath string


nit: let's add a comment for this field indicating whether the path is relative to the container or host

cdesiniotis · 2025-08-14T17:19:46Z

pkg/mig/reconfigure/options.go

+	// NVIDIACDIHookPath is the path to the nvidia-cdi-hook executable on the HOST.
+	NVIDIACDIHookPath string
+
+	hostNVIDIADir string


nit: let's add a comment documenting this field

cdesiniotis · 2025-08-14T17:24:09Z

pkg/mig/reconfigure/reconfigure.go

+	configStatePending   = "pending"
+	configStateRebooting = "rebooting"


Question -- shouldn't the "failed" and "success" states be defined as constants as well?

cdesiniotis · 2025-08-14T18:10:46Z

pkg/mig/reconfigure/reconfigure.go

+
+	if opts.WithShutdownHostGPUClients {
+		log.Info("Shutting down all GPU clients on the host by stopping their systemd services")
+		if err := opts.hostStopSystemdServices(systemdClients); err != nil {


This doesn't appear right to me. systemdClients is declared at the top of this function (line 177) but it is never populated with values. Shouldn't we be stopping all the services present in the opts.HostGPUClientServices slice?

cdesiniotis · 2025-08-14T18:26:44Z

pkg/mig/reconfigure/reconfigure.go

+
+	if opts.WithShutdownHostGPUClients {
+		log.Info("Restarting all GPU clients previously shutdown on the host by restarting their systemd services")
+		if err := opts.hostStartSystemdServices(systemdClients); err != nil {


Same comment as before. I don't see systemdClients being set to anything.

cdesiniotis · 2025-08-14T21:54:26Z

pkg/mig/reconfigure/reconfigure.go

+			return err
+		}
+		if mustRestart {
+			systemdGPUClients = append(systemdGPUClients, service)


Okay so we are appending to this list here... this is hard to follow. I just read the original bash script and I better understand what this is doing now. Some of my previous comments / questions can be disregarded now.

Question -- should we be passing the systemGPUClients slice by reference instead of by value? I believe the answer is yes. When we append() here, a new slice gets created and the original systemdGPUClients identifier at the call-site will not see the newly created slice.

cdesiniotis · 2025-08-14T21:59:31Z

pkg/mig/reconfigure/reconfigure.go

+func (opts *reconfigureMIGOptions) hostStartSystemdServices(systemdGPUClients gpuClients) error {
+	if len(systemdGPUClients) == 0 {
+		for _, serviceName := range opts.HostGPUClientServices {
+			service := opts.newSystemdService(serviceName)
+
+			if mustRestart, _ := service.shouldRestart(); mustRestart {
+				systemdGPUClients = append(systemdGPUClients, service)
+			}
+		}
+	}
+
+	// TODO: We should allow restarts to continue on failure.
+	if err := systemdGPUClients.Restart(); err != nil {
+		return fmt.Errorf("some services failed to start: %w", err)
+	}
+	return nil
+}


Question -- instead of maintaining a list of systemdGPUClients that need to be restarted BEFORE calling this function, what if we always construct this list in this function? That would eliminate the need to pass any arguments to hostStopSystemdServices() / hostStartSystemdServices() and make this much more readable in my opinion.

This is potentially out-of-scope for this PR which is simply a 1:1 port of the bash script to Go.

tariq1890 · 2025-08-15T06:27:43Z

pkg/mig/reconfigure/api.go

+// nodeLabeller defines an interface for interacting with node labels.
+//
+//go:generate moq -rm -fmt=goimports -out node-labeller_mock.go . nodeLabeller
+type nodeLabeller interface {


Can we approach the interface differently? What if we had an interface that captured all the Kubernetes APIServer actions? The interface would have methods where each one is a wrapper over kube client calls

tariq1890 · 2025-08-15T06:28:08Z

pkg/mig/reconfigure/cdi.go

+
+	spec, err := cdilib.GetSpec()
+	if err != nil {
+		return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)


Suggested change

return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)

return fmt.Errorf("failed to generate CDI spec for management containers: %w", err)

tariq1890 · 2025-08-15T06:28:19Z

pkg/mig/reconfigure/cdi.go

+		nvcdi.WithClass("gpu"),
+	)
+	if err != nil {
+		return fmt.Errorf("failed to create CDI library for management containers: %v", err)


Suggested change

return fmt.Errorf("failed to create CDI library for management containers: %v", err)

return fmt.Errorf("failed to create CDI library for management containers: %w", err)

tariq1890 · 2025-08-15T06:28:42Z

pkg/mig/reconfigure/cdi.go

+		transformroot.WithTargetDevRoot(opts.DevRoot),
+	)
+	if err := transformer.Transform(spec.Raw()); err != nil {
+		return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)


Suggested change

return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)

return fmt.Errorf("failed to transform driver root in CDI spec: %w", err)

tariq1890 · 2025-08-15T06:29:04Z

pkg/mig/reconfigure/cdi.go

+	if err := transformer.Transform(spec.Raw()); err != nil {
+		return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)
+	}
+	err = spec.Save("/var/run/cdi/management.nvidia.com-gpu.yaml")


can we move "/var/run/cdi" to a constant?

tariq1890 · 2025-08-15T06:30:14Z

pkg/mig/reconfigure/cdi.go

+	return opts.Run(cmd)
+}
+
+func (opts *reconfigurer) runNvidiaSMI() error {


Instead of running nvidia-smi, can we use go-nvlib/go-nvml to retrieve the relevant info?

tariq1890 · 2025-08-15T06:30:35Z

pkg/mig/reconfigure/clients.go

+	"k8s.io/apimachinery/pkg/watch"
+)
+
+// A gpuClient represents a client that can be stoped or restarted.


Suggested change

// A gpuClient represents a client that can be stoped or restarted.

// A gpuClient represents a client that can be stopped or restarted.

tariq1890 · 2025-08-15T06:31:28Z

pkg/mig/reconfigure/clients.go

+
+func (o *pod) delete() error {
+	err := o.node.clientset.CoreV1().Pods(o.namespace).DeleteCollection(
+		context.TODO(),


we should propagate the cli context here

tariq1890 · 2025-08-15T06:31:49Z

pkg/mig/reconfigure/clients.go

+
+func (o *pod) waitForDeletion() error {
+	timeout := 5 * time.Minute
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)


propagate CLI context here

tariq1890 · 2025-08-15T06:33:20Z

pkg/mig/reconfigure/clients.go

+}
+
+func (o *systemdService) Restart() error {
+	cmd := exec.Command("chroot", o.hostRootMount, "systemctl", "start", o.name) // #nosec G204 -- HostRootMount validated via dirpath, service validated via systemd_service_name.


Can we use a go library instead of shelling out?

Would this work? https://github.com/coreos/go-systemd

This change adds a mig/reconfigure package that implements the functionality added to the vgpu-device-manager. Signed-off-by: Evan Lezar <elezar@nvidia.com>

Signed-off-by: Evan Lezar <elezar@nvidia.com>

elezar requested review from cdesiniotis and tariq1890 July 8, 2025 14:27

elezar force-pushed the migrate-reconfigure-to-golang branch 2 times, most recently from 0ea344f to c497223 Compare July 8, 2025 15:16

elezar mentioned this pull request Jul 9, 2025

Add MIG config support when MIG-backed vGPU type NVIDIA/vgpu-device-manager#93

Merged

elezar commented Jul 9, 2025

View reviewed changes

elezar marked this pull request as ready for review July 9, 2025 15:02

ArangoGutierrez assigned elezar Jul 9, 2025

ArangoGutierrez requested review from ArangoGutierrez and Copilot July 9, 2025 16:13

Copilot AI reviewed Jul 10, 2025

View reviewed changes

ArangoGutierrez reviewed Jul 10, 2025

View reviewed changes

elezar mentioned this pull request Jul 10, 2025

Add reconfigure api #218

Open

tariq1890 reviewed Jul 11, 2025

View reviewed changes

elezar force-pushed the migrate-reconfigure-to-golang branch from 3d5d277 to bb6a01a Compare August 12, 2025 11:27

elezar marked this pull request as draft August 12, 2025 11:27

cdesiniotis reviewed Aug 14, 2025

View reviewed changes

tariq1890 reviewed Aug 15, 2025

View reviewed changes

elezar force-pushed the migrate-reconfigure-to-golang branch from bb6a01a to b324fe6 Compare September 16, 2025 12:04

elezar added 7 commits September 16, 2025 14:26

Add MIG Reconfigure API for vgpu-device-manager

5ea34a6

This change adds a mig/reconfigure package that implements the functionality added to the vgpu-device-manager. Signed-off-by: Evan Lezar <elezar@nvidia.com>

Return formatted errors instead of logging

89e8bfc

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add commandRunnner interface and a basic test

d4ed26a

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Update vendoring

f793f1a

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add migParted interface for testing

6544a19

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add nodeLabeller interface for testing

36f8eef

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add additional test cases

9cf3052

Signed-off-by: Evan Lezar <elezar@nvidia.com>

elezar added 19 commits September 16, 2025 14:28

TOFIX: Use internal reconfigure API

997f1ab

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: foo

5e326b1

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Fix CDI enabled

14260b7

TOFIX

fbfb5a2

Signed-off-by: Evan Lezar <elezar@nvidia.com>

More fixes

152a49c

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: DriverLibrary path is optional

b687e49

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Log application version

e867eb2

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: Ensure that config state label is set

9f26086

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: Use Mkdirall

a31f59a

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Ensure that clientset is set

6b69ef4

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Fix Dockerfile lint errors

87f983b

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add logs for waiting for deletion

2b43091

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: Use cmd to delete pods

878a2bb

Signed-off-by: Evan Lezar <elezar@nvidia.com>

TOFIX: Remove logging

5285e2c

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Only restart validator if successful

62cb648

Signed-off-by: Evan Lezar <elezar@nvidia.com>

FIX: return from select statement

c5dfa7b

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add additional logging

f4d3c11

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Ignore missing pods on shutdown

df9f766

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Fix delete

675541a

Signed-off-by: Evan Lezar <elezar@nvidia.com>

elezar force-pushed the migrate-reconfigure-to-golang branch from b324fe6 to 675541a Compare September 16, 2025 12:30

	cmd := exec.Command("chroot", opts.NVIDIASMIPath) //nolint:gosec
	cmd := exec.Command("chroot", opts.HostRootMount, opts.NVIDIASMIPath) //nolint:gosec

	// DriverLibrayPath is the path to libnvidia-ml.so.1 in the container.
	// DriverLibraryPath is the path to libnvidia-ml.so.1 in the container.

	return fmt.Errorf("failed to genereate CDI spec for management containers: %v", err)
	return fmt.Errorf("failed to generate CDI spec for management containers: %v", err)

	if err := cmd.Run(); err != nil {
	isEnabledErr := cmd.Run()
	if isEnabledErr != nil {

	return fmt.Errorf("failed to create CDI library for management containers: %v", err)
	return fmt.Errorf("failed to create CDI library for management containers: %w", err)

	return fmt.Errorf("failed to transform driver root in CDI spec: %v", err)
	return fmt.Errorf("failed to transform driver root in CDI spec: %w", err)

	return fmt.Errorf("failed to save CDI spec for management containers: %v", err)
	return fmt.Errorf("failed to save CDI spec for management containers: %w", err)

	return "", fmt.Errorf("unable to get node object: %v", err)
	return "", fmt.Errorf("unable to get node object: %w", err)

	return fmt.Errorf("unable to get node object: %v", err)
	return fmt.Errorf("unable to get node object: %w", err)

	return fmt.Errorf("unable to update node object: %v", err)
	return fmt.Errorf("unable to update node object: %w", err)

		configStatePending = "pending"
		configStateRebooting = "rebooting"

	// A gpuClient represents a client that can be stoped or restarted.
	// A gpuClient represents a client that can be stopped or restarted.

Migrate reconfigure to golang #216

Are you sure you want to change the base?

Migrate reconfigure to golang #216

Uh oh!

Conversation

elezar commented Jul 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

ArangoGutierrez left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

elezar Jul 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jgehrcke commented Jul 10, 2025

Uh oh!

ArangoGutierrez commented Jul 10, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

elezar commented Jul 8, 2025 •

edited

Loading

elezar Jul 11, 2025 •

edited

Loading

cdesiniotis Aug 14, 2025 •

edited

Loading