Enable NFD rule for GPU resource driver Helm chart (#68)

* add gpu nfd rule
intel · Jan 4, 2025 · 5326261 · 5326261
1 parent 0020d7e
commit 5326261
Show file tree

Hide file tree

Showing 7 changed files with 137 additions and 7 deletions.
diff --git a/charts/intel-gpu-resource-driver/Chart.yaml b/charts/intel-gpu-resource-driver/Chart.yaml
@@ -3,5 +3,13 @@ name: intel-gpu-resource-driver
 description: A Helm chart for a Dynamic Resource Allocation (DRA) Intel GPU Resource Driver
 
 type: application
-version: 0.6.0
-appVersion: "v0.6.0"
+version: 0.7.0
+appVersion: "v0.7.0"
+home: https://github.com/intel/helm-charts
+
+dependencies:
+  - name: node-feature-discovery
+    alias: nfd
+    version: "0.16.6"
+    condition: nfd.enabled
+    repository: https://kubernetes-sigs.github.io/node-feature-discovery/charts
diff --git a/charts/intel-gpu-resource-driver/README.md b/charts/intel-gpu-resource-driver/README.md
@@ -16,7 +16,9 @@ helm repo update
 You can execute `helm search repo intel` command to see pulled charts [optional].
 
 ## Install Helm Chart
+When installing, update the dependencies:
 ```
+helm dependency update
 helm install intel-gpu-resource-driver intel/intel-gpu-resource-driver
 ```
 ## Upgrade Chart
@@ -43,7 +45,7 @@ You may also run `helm show values` on this chart's dependencies for additional
 | image.repository | string | `intel` |
 | image.name | string | `"intel-gpu-resource-driver"` |
 | image.pullPolicy | string | `"IfNotPresent"` |
-| image.tag | string | `"v0.6.0"` |
+| image.tag | string | `"v0.7.0"` |
 
 > [!Note]
 > When upgrading, CRDs from previous version need to be removed manually because Helm supports neither upgrading nor deleting CRDs, see: https://github.com/helm/community/blob/main/hips/hip-0011.md
diff --git a/charts/intel-gpu-resource-driver/templates/device-class.yaml b/charts/intel-gpu-resource-driver/templates/device-class.yaml
@@ -1,4 +1,4 @@
-apiVersion: resource.k8s.io/v1alpha3
+apiVersion: resource.k8s.io/v1beta1
 kind: DeviceClass
 metadata:
   name: gpu.intel.com

diff --git a/charts/intel-gpu-resource-driver/templates/nfd.yaml b/charts/intel-gpu-resource-driver/templates/nfd.yaml
@@ -0,0 +1,96 @@
+{{- if .Values.nfd.enabled }}
+apiVersion: nfd.k8s-sigs.io/v1alpha1
+kind: NodeFeatureRule
+metadata:
+  name: intel-gpu-device-rule
+spec:
+  rules:
+    - name: "intel.gpu"
+      labels:
+        "intel.feature.node.kubernetes.io/gpu": "true"
+      matchFeatures:
+        - feature: pci.device
+          matchExpressions:
+            vendor: {op: In, value: ["8086"]}
+            class: {op: In, value: ["0300", "0380"]}
+      matchAny:
+        - matchFeatures:
+          - feature: kernel.loadedmodule
+            matchExpressions:
+              i915: {op: Exists}
+        - matchFeatures:
+          - feature: kernel.enabledmodule
+            matchExpressions:
+              i915: {op: Exists}
+---
+apiVersion: nfd.k8s-sigs.io/v1alpha1
+kind: NodeFeatureRule
+metadata:
+  name: intel-gpu-platform-labeling
+spec:
+  rules:
+    # A_Series (Alchemist)
+    - labels:
+        gpu.intel.com/family: "A_Series"
+      matchFeatures:
+        - feature: pci.device
+          matchExpressions:
+            class: {op: In, value: ["0300"]}
+            vendor: {op: In, value: ["8086"]}
+            device:
+              op: In
+              value:
+                - "56a6"
+                - "56a5"
+                - "56a1"
+                - "56a0"
+                - "5694"
+                - "5693"
+                - "5692"
+                - "5691"
+                - "5690"
+                - "56b3"
+                - "56b2"
+                - "56a4"
+                - "56a3"
+                - "5697"
+                - "5696"
+                - "5695"
+                - "56b1"
+                - "56b0"
+      name: intel.gpu.a.series
+    # Max_Series
+    - labels:
+        gpu.intel.com/family: "Max_Series"
+      matchFeatures:
+        - feature: pci.device
+          matchExpressions:
+            class: {op: In, value: ["0380"]}
+            vendor: {op: In, value: ["8086"]}
+            device:
+              op: In
+              value:
+                - "0bda"
+                - "0bd5"
+                - "0bd9"
+                - "0bdb"
+                - "0bd7"
+                - "0bd6"
+                - "0bd0"
+      name: intel.gpu.max.series
+    # Flex_Series
+    - labels:
+        gpu.intel.com/family: "Flex_Series"
+      matchFeatures:
+        - feature: pci.device
+          matchExpressions:
+            class: {op: In, value: ["0300", "0380"]}
+            vendor: {op: In, value: ["8086"]}
+            device:
+              op: In
+              value:
+                - "0f00"
+                - "0f01"
+                - "0f02"
+      name: intel.gpu.flex.series
+{{- end }}
diff --git a/charts/intel-gpu-resource-driver/templates/resource-driver.yaml b/charts/intel-gpu-resource-driver/templates/resource-driver.yaml
@@ -73,10 +73,15 @@ spec:
       tolerations:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      {{- if .Values.nfd.enabled }}
+      nodeSelector:
+        intel.feature.node.kubernetes.io/gpu: "true"
+      {{- else }}
       {{- with .Values.kubeletPlugin.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      {{- end }}
       {{- with .Values.kubeletPlugin.affinity }}
       affinity:
         {{- toYaml . | nindent 8 }}

diff --git a/charts/intel-gpu-resource-driver/templates/validating-admission-policy.yaml b/charts/intel-gpu-resource-driver/templates/validating-admission-policy.yaml
@@ -7,7 +7,7 @@ spec:
   matchConstraints:
     resourceRules:
     - apiGroups:   ["resource.k8s.io"]
-      apiVersions: ["v1alpha3"]
+      apiVersions: ["v1beta1"]
       operations:  ["CREATE", "UPDATE", "DELETE"]
       resources:   ["resourceslices"]
   matchConditions:

diff --git a/charts/intel-gpu-resource-driver/values.yaml b/charts/intel-gpu-resource-driver/values.yaml
@@ -9,7 +9,7 @@ image:
   repository: intel
   name: intel-gpu-resource-driver
   pullPolicy: IfNotPresent
-  tag: "v0.6.0"
+  tag: "v0.7.0"
 
 serviceAccount:
   create: true
@@ -19,6 +19,25 @@ serviceAccount:
 
 kubeletPlugin:
   podAnnotations: {}
-  tolerations: []
   nodeSelector: {}
+  # label used when nfd.enabled is true
+    #intel.feature.node.kubernetes.io/gpu: "true"
+  tolerations:
+    - key: node-role.kubernetes.io/master
+      operator: Exists
+      effect: NoSchedule
+    - key: node-role.kubernetes.io/control-plane
+      operator: Exists
+      effect: NoSchedule
+      # Refer to the official documentation for Node Feature Discovery (NFD) 
+      # regarding node tainting: 
+      # https://nfd.sigs.k8s.io/usage/customization-guide#node-tainting
+    - key: "node.kubernetes.io/gpu"
+      operator: "Exists"
+      effect: "NoSchedule"
   affinity: {}
+
+nfd:
+  enabled: false # change to true to install NFD to the cluster
+  nameOverride: intel-gpu-nfd
+  enableNodeFeatureApi: true