Merge pull request #1010 from elezar/add-imex-init-container

Add init container to handle imex nodes config mount

api/config/v1/config.go

@@ -61,7 +61,6 @@ func NewConfig(c *cli.Context, flags []cli.Flag) (*Config, error) {
 	if c.IsSet("imex-required") {
 		config.Imex.Required = c.Bool("imex-required")
 	}
-	updateFromCLIFlag(&config.Imex.NodesConfigFile, c, "imex-nodes-config-file")
 
 	// If nvidiaDevRoot (the path to the device nodes on the host) is not set,
 	// we default to using the driver root on the host.

api/config/v1/imex.go

@@ -39,10 +39,6 @@ type Imex struct {
 	// If it is not required its injection is skipped if the device nodes do not exist or if its
 	// existence cannot be queried.
 	Required bool `json:"required,omitempty" yaml:"required,omitempty"`
-	// NodesConfigFile defines the location to the IMEX nodes config file.
-	// Such a nodes config file contains the IP addresses of nodes that are part of the IMEX domain.
-	// Note that this is the absolute path to the file in the device plugin container.
-	NodesConfigFile *string `json:"nodesConfigFile,omitempty" yaml:"nodesConfigFile,omitempty"`
 }
 
 // AssertChannelIDsIsValid checks whether the specified list of channel IDs is valid.

cmd/gpu-feature-discovery/main.go

@@ -86,12 +86,6 @@ func main() {
 			Value:   "/etc/kubernetes/node-feature-discovery/features.d/gfd",
 			EnvVars: []string{"GFD_OUTPUT_FILE"},
 		},
-		&cli.StringFlag{
-			Name:    "imex-nodes-config-file",
-			Usage:   "Path to the IMEX nodes config file. This file contains a list of IP addresses of the nodes in the IMEX domain.",
-			Value:   "/etc/nvidia-imex/nodes_config.cfg",
-			EnvVars: []string{"GFD_IMEX_NODES_CONFIG_FILE"},
-		},
 		&cli.StringFlag{
 			Name:    "machine-type-file",
 			Value:   "/sys/class/dmi/id/product_name",

deployments/helm/nvidia-device-plugin/templates/daemonset-gfd.yml

@@ -57,7 +57,35 @@ spec:
       {{- end }}
       {{- if $options.hasConfigMap }}
       shareProcessNamespace: true
+      {{- end }}
       initContainers:
+      - image: {{ include "nvidia-device-plugin.fullimage" . }}
+        name: gpu-feature-discovery-imex-init
+        command: ["/bin/bash", "-c"]
+        args:
+        - |
+          IMEX_NODES_CONFIG_FILE=/etc/nvidia-imex/nodes_config.cfg
+          if [[ -f /config/${IMEX_NODES_CONFIG_FILE} ]]; then
+            echo "Removing cached IMEX nodes config"
+            rm -f /config/${IMEX_NODES_CONFIG_FILE}
+          fi
+
+          if [[ ! -f /driver-root/${IMEX_NODES_CONFIG_FILE} ]]; then
+            echo "No IMEX nodes config path detected; Skipping"
+            exit 0
+          fi
+
+          echo "Copying IMEX nodes config"
+          mkdir -p $(dirname /config/${IMEX_NODES_CONFIG_FILE})
+          cp /driver-root/${IMEX_NODES_CONFIG_FILE} /config/${IMEX_NODES_CONFIG_FILE}
+        volumeMounts:
+        - name: config
+          mountPath: /config
+        - name: driver-root
+          mountPath: /driver-root/etc
+          subPath: etc
+          readOnly: true
+      {{- if $options.hasConfigMap }}
       - image: {{ include "nvidia-device-plugin.fullimage" . }}
         name: gpu-feature-discovery-init
         command: ["config-manager"]
@@ -182,14 +210,12 @@ spec:
             mountPath: "/etc/kubernetes/node-feature-discovery/features.d"
           - name: host-sys
             mountPath: "/sys"
-          - name: nvidia-imex-dir
-            mountPath: "/etc/nvidia-imex"
         {{- if $options.hasConfigMap }}
           - name: available-configs
             mountPath: /available-configs
+        {{- end }}
           - name: config
             mountPath: /config
-        {{- end }}
         {{- with .Values.resources }}
         resources:
           {{- toYaml . | nindent 10 }}
@@ -201,17 +227,17 @@ spec:
         - name: host-sys
           hostPath:
             path: "/sys"
-        - name: nvidia-imex-dir
-          type: DirectoryOrCreate
+        - name: driver-root
           hostPath:
-            path: {{ clean ( join "/" ( list "/" .Values.nvidiaDriverRoot "/etc/nvidia-imex" ) ) | quote }}
+            path: {{ clean ( join "/" ( list "/" .Values.nvidiaDriverRoot ) ) | quote }}
+            type: Directory
       {{- if $options.hasConfigMap }}
         - name: available-configs
           configMap:
             name: {{ $configMapName }}
+      {{- end }}
         - name: config
           emptyDir: {}
-      {{- end }}
       {{- with .Values.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}

internal/lm/fabric.go → internal/lm/imex.go

@@ -34,19 +34,16 @@ import (
 	"github.com/NVIDIA/k8s-device-plugin/internal/resource"
 )
 
-func newImexLabeler(config *spec.Config, devices []resource.Device) (Labeler, error) {
-	if config.Imex.NodesConfigFile == nil || *config.Imex.NodesConfigFile == "" {
-		// No imex config file, return empty labels
-		return empty{}, nil
-	}
-
-	nodesConfigFiles := []string{*config.Imex.NodesConfigFile}
-	if root := config.Flags.Plugin.ContainerDriverRoot; root != nil && *root != "" {
-		nodesConfigFiles = append(nodesConfigFiles, filepath.Join(*root, *config.Imex.NodesConfigFile))
-	}
+const (
+	// ImexNodesConfigFilePath is the path to the IMEX nodes config file.
+	// This file contains a list of IP addresses of the nodes in the IMEX domain.
+	ImexNodesConfigFilePath = "/etc/nvidia-imex/nodes_config.cfg"
+)
 
+func newImexLabeler(config *spec.Config, devices []resource.Device) (Labeler, error) {
 	var errs error
-	for _, configFilePath := range nodesConfigFiles {
+	for _, root := range imexNodesConfigFilePathSearchRoots(config) {
+		configFilePath := filepath.Join(root, ImexNodesConfigFilePath)
 		imexLabeler, err := imexLabelerForConfigFile(configFilePath, devices)
 		if err != nil {
 			errs = errors.Join(errs, err)
@@ -64,6 +61,19 @@ func newImexLabeler(config *spec.Config, devices []resource.Device) (Labeler, er
 	return empty{}, nil
 }
 
+// imexNodesConfigFilePathSearchRoots returns a list of roots to search for the IMEX nodes config file.
+func imexNodesConfigFilePathSearchRoots(config *spec.Config) []string {
+	// By default, search / and /config for config files.
+	roots := []string{"/", "/config"}
+
+	if config == nil || config.Flags.Plugin == nil || config.Flags.Plugin.ContainerDriverRoot == nil {
+		return roots
+	}
+
+	// If a driver root is specified, it is also searched.
+	return append(roots, *config.Flags.Plugin.ContainerDriverRoot)
+}
+
 func imexLabelerForConfigFile(configFilePath string, devices []resource.Device) (Labeler, error) {
 	imexConfigFile, err := os.Open(configFilePath)
 	if os.IsNotExist(err) {