add nfd rule for gaudi resource driver

Signed-off-by: Oksana Baranova <[email protected]>
intel · Dec 17, 2024 · 3a06f82 · 3a06f82
1 parent 1f3edad
commit 3a06f82
Show file tree

Hide file tree

Showing 4 changed files with 41 additions and 3 deletions.
diff --git a/charts/intel-gaudi-resource-driver/Chart.yaml b/charts/intel-gaudi-resource-driver/Chart.yaml
@@ -5,3 +5,11 @@ description: A Helm chart for a Dynamic Resource Allocation (DRA) Intel Gaudi Re
 type: application
 version: 0.2.0
 appVersion: "v0.2.0"
+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-gaudi-resource-driver/README.md b/charts/intel-gaudi-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-gaudi-resource-driver intel/intel-gaudi-resource-driver
 ```
 ## Upgrade Chart
@@ -46,4 +48,4 @@ You may also run `helm show values` on this chart's dependencies for additional
 | image.tag | string | `"v0.2.0"` |
 
 > [!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
+> If you change the image tag to be used in Helm chart deployment, ensure that the version of the container image is consistent with deployment YAMLs - they might change between releases.
diff --git a/charts/intel-gaudi-resource-driver/templates/nfd.yaml b/charts/intel-gaudi-resource-driver/templates/nfd.yaml
@@ -0,0 +1,15 @@
+apiVersion: nfd.k8s-sigs.io/v1alpha1
+kind: NodeFeatureRule
+metadata:
+  name: intel-gaudi-device-rule
+spec:
+  rules:
+    - name: "intel.gaudi"
+      labels:
+        "intel.feature.node.kubernetes.io/gaudi": "true"
+      matchFeatures:
+        - feature: pci.device
+          matchExpressions:
+            vendor: {op: In, value: ["1da3"]}
+            device: {op: In, value: ["1020", "1030"]}
+
diff --git a/charts/intel-gaudi-resource-driver/values.yaml b/charts/intel-gaudi-resource-driver/values.yaml
@@ -19,13 +19,26 @@ serviceAccount:
 
 kubeletPlugin:
   podAnnotations: {}
+  nodeSelector:
+    intel.feature.node.kubernetes.io/gaudi: "true"
   tolerations:
     - key: node-role.kubernetes.io/master
       operator: Exists
       effect: NoSchedule
     - key: node-role.kubernetes.io/control-plane
       operator: Exists
       effect: NoSchedule
-  nodeSelector: {}
-  #node-role.kubernetes.io/control-plane: ""
+      # Refer to the official documentation for Node Feature Discovery (NFD) 
+      # regarding node tainting: 
+      # https://nfd.sigs.k8s.io/usage/customization-guide#node-tainting
+    - key: "intel.feature.node.kubernetes.io/gaudi"
+      operator: "Exists"
+      effect: "NoSchedule"  
   affinity: {}
+
+nfd:
+  enabled: false # change to true to install NFD to the cluster
+  nameOverride: intel-gaudi-nfd
+  # TODO: this deprecated NFD option will be replaced in NFD v0.17 with "featureGates.NodeFeatureAPI" (added in v0.16):
+  # https://kubernetes-sigs.github.io/node-feature-discovery/v0.16/deployment/helm.html#general-parameters
+  enableNodeFeatureApi: true