Attempt nvidia-device-plugin app fixes

Author: nwthomas

helm/nvidia-device-plugin/Chart.yaml

@@ -3,4 +3,4 @@ name: nvidia-device-plugin
 description: NVIDIA Device Plugin for Kubernetes
 type: application
 version: 0.1.0
-appVersion: "v0.17.1"
+appVersion: "v0.18.0"

helm/nvidia-device-plugin/templates/daemonset.yaml

@@ -25,7 +25,7 @@ spec:
       - image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
         imagePullPolicy: {{ .Values.image.pullPolicy }}
         name: nvidia-device-plugin-ctr
-        args: ["--fail-on-init-error=false"]
+        command: ["nvidia-device-plugin"]
         securityContext:
           {{- toYaml .Values.securityContext | nindent 10 }}
         resources:
@@ -35,8 +35,13 @@ spec:
         env:
         - name: PASS_DEVICE_SPECS
           value: "true"
+        {{- if typeIs "bool" .Values.failOnInitError }}
+        - name: FAIL_ON_INIT_ERROR
+          value: {{ .Values.failOnInitError | quote }}
+        {{- else }}
         - name: FAIL_ON_INIT_ERROR
           value: "false"
+        {{- end }}
         {{- if .Values.env }}
         {{- toYaml .Values.env | nindent 8 }}
         {{- else }}
@@ -50,5 +55,4 @@ spec:
           value: "compute,utility"
         {{- end }}
       volumes:
-        {{- toYaml .Values.volumes | nindent 8 }}
-      hostNetwork: true
+        {{- toYaml .Values.volumes | nindent 8 }}

helm/nvidia-device-plugin/values.yaml

@@ -3,6 +3,8 @@ image:
   tag: v0.18.0
   pullPolicy: IfNotPresent
 
+failOnInitError: false
+
 nodeSelector:
   kubernetes.io/arch: amd64
   gpu: "true"
@@ -33,6 +35,15 @@ env:
     value: "all"
   - name: NVIDIA_DRIVER_CAPABILITIES
     value: "compute,utility"
+  # Set NVIDIA driver root to help locate driver libraries
+  # Default is "/" which should work with privileged containers
+  - name: NVIDIA_DRIVER_ROOT
+    value: "/"
+  # Device discovery strategy: "auto" (default) or "nvml"
+  # Use "auto" to let the plugin automatically detect the best strategy
+  # Only use "nvml" if you know NVML is available and auto-detection fails
+  # - name: DEVICE_DISCOVERY_STRATEGY
+  #   value: "auto"
 
 volumeMounts:
   - name: device-plugin

helm/ollama/values.yaml

@@ -1,4 +1,3 @@
-# Custom values for ollama-helm with GPU support
 # Based on official otwld/ollama-helm chart with custom configurations
 
 replicaCount: 1
@@ -172,9 +171,11 @@ persistentVolume:
   enabled: true
   accessModes:
     - ReadWriteOnce
-  annotations: {}
+  annotations:
+    # Set the number of replicas for this Longhorn volume by overriding the storage class default
+    longhorn.io/number-of-replicas: "1"
   existingClaim: ""
-  size: 500Gi
+  size: 200Gi
   storageClass: "longhorn"
   volumeMode: ""
   subPath: ""