Feat/custom resource name

.gitignore

@@ -10,3 +10,11 @@ bin
 *.out
 *.log
 ginkgo.json
+docs/custom-resource-name-tasks.md
+docs/custom-resource-name-prefix.md
+docs/custom-resource-name-prefix_zh.md
+CLAUDE.md
+custom-nvidia-device-plugin.yaml
+build-application.sh
+push-application.sh
+docs/nvidia-device-plugin-driver-interaction.md

README.md

@@ -199,6 +199,47 @@ Test PASSED
 Done
 ```
 
+#### Using Custom Resource Name Prefix
+
+If you have configured the device plugin with a custom resource name prefix, you can request GPUs using that prefix:
+
+```shell
+# First, deploy the device plugin with a custom prefix
+cat <<EOF | kubectl apply -f -
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: nvidia-plugin-config
+  namespace: nvidia-device-plugin
+data:
+  config.yaml: |
+    version: v1
+    flags:
+      resourceNamePrefix: "custom.domain"
+EOF
+```
+
+```shell
+cat <<EOF | kubectl apply -f -
+apiVersion: v1
+kind: Pod
+metadata:
+  name: gpu-pod-custom
+spec:
+  restartPolicy: Never
+  containers:
+    - name: cuda-container
+      image: nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0
+      resources:
+        limits:
+          custom.domain/gpu: 1 # requesting 1 GPU with custom prefix
+  tolerations:
+  - key: nvidia.com/gpu
+    operator: Exists
+    effect: NoSchedule
+EOF
+```
+
 > [!WARNING]
 > If you do not request GPUs when you use the device plugin, the plugin exposes all the GPUs on the machine inside your container.
 
@@ -221,6 +262,7 @@ deploying the plugin via `helm`.
 | `--pass-device-specs`    | `$PASS_DEVICE_SPECS`    | `false`         |
 | `--device-list-strategy` | `$DEVICE_LIST_STRATEGY` | `"envvar"`      |
 | `--device-id-strategy`   | `$DEVICE_ID_STRATEGY`   | `"uuid"`        |
+| `--resource-name-prefix` | `$RESOURCE_NAME_PREFIX` | `"nvidia.com"`  |
 | `--config-file`          | `$CONFIG_FILE`          | `""`            |
 
 ### As a configuration file
@@ -231,6 +273,7 @@ flags:
   migStrategy: "none"
   failOnInitError: true
   nvidiaDriverRoot: "/"
+  resourceNamePrefix: "nvidia.com"
   plugin:
     passDeviceSpecs: false
     deviceListStrategy: "envvar"
@@ -352,6 +395,38 @@ options outside of this section are shared.
   launch time. As described below, a `ConfigMap` can be used to point the
   plugin at a desired configuration file when deploying via `helm`.
 
+**`RESOURCE_NAME_PREFIX`**:
+  customize the prefix for GPU resource names and labels
+
+  `(default 'nvidia.com')`
+
+  The `RESOURCE_NAME_PREFIX` option allows you to customize the domain prefix
+  for all GPU-related resources and labels. By default, resources are exposed as
+  `nvidia.com/gpu`, but you can change this to any custom domain (e.g.,
+  `custom.domain/gpu`).
+
+  This affects:
+  - GPU resource names in pod specs: `{prefix}/gpu`
+  - MIG resource names: `{prefix}/mig-*`
+  - All node labels generated by gpu-feature-discovery: `{prefix}/*`
+
+  **Example**:
+  ```yaml
+  # Custom resource name prefix
+  flags:
+    resourceNamePrefix: "custom.domain"
+
+  # Resources will be advertised as: custom.domain/gpu
+  # Labels will be: custom.domain/gpu.count, custom.domain/gpu.product, etc.
+  ```
+
+  **Important Notes**:
+  - The prefix must not contain '/'
+  - The prefix must be 253 characters or less
+  - This prefix must match across both the device plugin and gpu-feature-discovery
+  - Changing this prefix requires updating all pod specs to request resources using the new prefix
+  - This is useful for multi-tenant clusters or environments requiring custom resource naming
+
 ### Shared Access to GPUs
 
 The NVIDIA device plugin allows oversubscription of GPUs through a set of

api/config/v1/config.go

@@ -39,6 +39,16 @@ type Config struct {
 	Imex      Imex      `json:"imex,omitempty"      yaml:"imex,omitempty"`
 }
 
+// GetResourceNamePrefix returns the configured resource name prefix.
+// If not set, it returns the default prefix.
+func (c *Config) GetResourceNamePrefix() string {
+	if c.Flags.ResourceNamePrefix != nil && *c.Flags.ResourceNamePrefix != "" {
+		return *c.Flags.ResourceNamePrefix
+	}
+	return DefaultResourceNamePrefix
+}
+
+
 // NewConfig builds out a Config struct from a config file (or command line flags).
 // The data stored in the config will be populated in order of precedence from
 // (1) command line, (2) environment variable, (3) config file.

api/config/v1/consts.go

@@ -22,6 +22,7 @@ import (
 
 // Constants related to resource names
 const (
+	DefaultResourceNamePrefix       = "nvidia.com"
 	ResourceNamePrefix              = "nvidia.com"
 	DefaultSharedResourceNameSuffix = ".shared"
 	MaxResourceNameLength           = 63

api/config/v1/flags.go

@@ -57,6 +57,7 @@ type Flags struct {
 type CommandLineFlags struct {
 	MigStrategy             *string                 `json:"migStrategy"                yaml:"migStrategy"`
 	FailOnInitError         *bool                   `json:"failOnInitError"            yaml:"failOnInitError"`
+	ResourceNamePrefix      *string                 `json:"resourceNamePrefix,omitempty" yaml:"resourceNamePrefix,omitempty"`
 	MpsRoot                 *string                 `json:"mpsRoot,omitempty"          yaml:"mpsRoot,omitempty"`
 	NvidiaDriverRoot        *string                 `json:"nvidiaDriverRoot,omitempty" yaml:"nvidiaDriverRoot,omitempty"`
 	NvidiaDevRoot           *string                 `json:"nvidiaDevRoot,omitempty"    yaml:"nvidiaDevRoot,omitempty"`
@@ -121,6 +122,8 @@ func (f *Flags) UpdateFromCLIFlags(c *cli.Context, flags []cli.Flag) {
 				updateFromCLIFlag(&f.MigStrategy, c, n)
 			case "fail-on-init-error":
 				updateFromCLIFlag(&f.FailOnInitError, c, n)
+			case "resource-name-prefix":
+				updateFromCLIFlag(&f.ResourceNamePrefix, c, n)
 			case "mps-root":
 				updateFromCLIFlag(&f.MpsRoot, c, n)
 			case "driver-root", "nvidia-driver-root":

api/config/v1/resources.go

@@ -19,6 +19,7 @@ package v1
 import (
 	"encoding/json"
 	"fmt"
+	"os"
 	"regexp"
 	"strings"
 
@@ -46,7 +47,8 @@ type Resources struct {
 // NewResourceName builds a resource name from the standard prefix and a name.
 // An error is returned if the format is incorrect.
 func NewResourceName(n string) (ResourceName, error) {
-	if !strings.HasPrefix(n, ResourceNamePrefix+"/") {
+	// Only add default prefix if the name doesn't already contain a '/' (i.e., no custom prefix)
+	if !strings.Contains(n, "/") {
 		n = ResourceNamePrefix + "/" + n
 	}
 
@@ -73,6 +75,8 @@ func NewResource(pattern, name string) (*Resource, error) {
 		Pattern: ResourcePattern(pattern),
 		Name:    resourceName,
 	}
+	// Log to stderr (visible in k8s logs)
+	fmt.Fprintf(os.Stderr, "[INFO] Created resource: pattern=%s, name=%s\n", pattern, resourceName)
 	return r, nil
 }
 

cmd/gpu-feature-discovery/main.go

@@ -61,6 +61,12 @@ func main() {
 			Usage:   "fail the plugin if an error is encountered during initialization, otherwise block indefinitely",
 			EnvVars: []string{"GFD_FAIL_ON_INIT_ERROR", "FAIL_ON_INIT_ERROR"},
 		},
+		&cli.StringFlag{
+			Name:    "resource-name-prefix",
+			Value:   "nvidia.com",
+			Usage:   "the prefix to use for resource names (e.g., 'nvidia.com' for nvidia.com/gpu)",
+			EnvVars: []string{"GFD_RESOURCE_NAME_PREFIX", "RESOURCE_NAME_PREFIX"},
+		},
 		&cli.BoolFlag{
 			Name:    "oneshot",
 			Value:   false,
@@ -138,6 +144,16 @@ func validateFlags(config *spec.Config) error {
 	default:
 		return fmt.Errorf("invalid --device-discovery-strategy option %v", *config.Flags.DeviceDiscoveryStrategy)
 	}
+
+	// Validate resource name prefix format
+	if config.Flags.ResourceNamePrefix != nil && *config.Flags.ResourceNamePrefix != "" {
+		prefix := *config.Flags.ResourceNamePrefix
+		if prefix != "nvidia.com" {
+			klog.Warningf("Using custom resource name prefix: %s (default is nvidia.com)", prefix)
+			klog.Warning("All pods requesting GPU resources must be updated to use the new resource name format")
+		}
+	}
+
 	return nil
 }
 

cmd/gpu-feature-discovery/mig_test.go

@@ -352,3 +352,74 @@ func TestMigStrategyMixed(t *testing.T) {
 	require.Contains(t, labels, "nvidia.com/mig-3g.20gb.count", "Missing label")
 	require.Contains(t, labels, "nvidia.com/mig-1g.5gb.count", "Missing label")
 }
+
+func TestMigStrategySingleWithCustomPrefix(t *testing.T) {
+	// create VGPU mock library with empty vgpu devices
+	vgpuMock := NewTestVGPUMock()
+	devices := []resource.Device{
+		rt.NewMigEnabledDevice(
+			rt.NewMigDevice(3, 0, 20),
+			rt.NewMigDevice(3, 0, 20),
+		),
+	}
+	nvmlMock := rt.NewManagerMockWithDevices(devices...)
+
+	conf := &spec.Config{
+		Flags: spec.Flags{
+			CommandLineFlags: spec.CommandLineFlags{
+				MigStrategy:       ptr("single"),
+				ResourceNamePrefix: ptr("custom.domain"),
+				FailOnInitError:   ptr(true),
+				GFD: &spec.GFDCommandLineFlags{
+					Oneshot:         ptr(true),
+					OutputFile:      ptr("./gfd-test-mig-single-custom"),
+					SleepInterval:   ptr(spec.Duration(time.Second)),
+					NoTimestamp:     ptr(false),
+					MachineTypeFile: ptr(testMachineTypeFile),
+				},
+			},
+		},
+	}
+
+	setupMachineFile(t)
+	defer removeMachineFile(t)
+
+	labelOutputer, err := lm.NewOutputer(conf, flags.NodeConfig{}, flags.ClientSets{})
+	require.NoError(t, err)
+
+	d := gfd{
+		manager:       nvmlMock,
+		vgpu:          vgpuMock,
+		config:        conf,
+		labelOutputer: labelOutputer,
+	}
+	restart, err := d.run(nil)
+	require.NoError(t, err, "Error from run function")
+	require.False(t, restart)
+
+	outFile, err := os.Open(*conf.Flags.GFD.OutputFile)
+	require.NoError(t, err, "Opening output file")
+
+	defer func() {
+		err = outFile.Close()
+		require.NoError(t, err, "Closing output file")
+		err = os.Remove(*conf.Flags.GFD.OutputFile)
+		require.NoError(t, err, "Removing output file")
+	}()
+
+	output, err := io.ReadAll(outFile)
+	require.NoError(t, err, "Reading output file")
+
+	labels, err := buildLabelMapFromOutput(output)
+	require.NoError(t, err, "Building map of labels from output file")
+
+	// Verify custom prefix is used in labels
+	require.Equal(t, labels["custom.domain/mig.strategy"], "single", "Incorrect label")
+	require.Equal(t, labels["custom.domain/gpu.count"], "2", "Incorrect label")
+	require.Equal(t, labels["custom.domain/gpu.product"], "MOCKMODEL-MIG-3g.20gb", "Incorrect label")
+	require.Equal(t, labels["custom.domain/gpu.memory"], "20", "Incorrect label")
+
+	// Verify default nvidia.com labels are NOT present
+	require.NotContains(t, labels, "nvidia.com/mig.strategy", "Default prefix should not be present")
+	require.NotContains(t, labels, "nvidia.com/gpu.count", "Default prefix should not be present")
+}

cmd/nvidia-device-plugin/main.go

@@ -69,6 +69,12 @@ func main() {
 			Usage:   "fail the plugin if an error is encountered during initialization, otherwise block indefinitely",
 			EnvVars: []string{"FAIL_ON_INIT_ERROR"},
 		},
+		&cli.StringFlag{
+			Name:    "resource-name-prefix",
+			Value:   "nvidia.com",
+			Usage:   "the prefix to use for resource names (e.g., 'nvidia.com' for nvidia.com/gpu)",
+			EnvVars: []string{"RESOURCE_NAME_PREFIX"},
+		},
 		&cli.StringFlag{
 			Name:    "driver-root",
 			Aliases: []string{"nvidia-driver-root"},
@@ -228,6 +234,17 @@ func validateFlags(infolib nvinfo.Interface, config *spec.Config) error {
 		return fmt.Errorf("invalid IMEX channel IDs: %w", err)
 	}
 
+	// Validate resource name prefix format
+	if config.Flags.ResourceNamePrefix != nil && *config.Flags.ResourceNamePrefix != "" {
+		prefix := *config.Flags.ResourceNamePrefix
+		if prefix == "nvidia.com" {
+			// This is the default, no special validation needed
+			return nil
+		}
+		klog.Warningf("Using custom resource name prefix: %s (default is nvidia.com)", prefix)
+		klog.Warning("All pods requesting GPU resources must be updated to use the new resource name format")
+	}
+
 	return nil
 }
 

deployments/container/Dockerfile

@@ -20,7 +20,7 @@ FROM base AS devel
 WORKDIR /work
 COPY * .
 
-ARG GOPROXY="https://proxy.golang.org,direct"
+ARG GOPROXY="https://goproxy.cn,direct"
 ENV GOPROXY=$GOPROXY
 
 RUN make install-tools