first commit

frameworks/GPU/config/gpu/templates/device-plugin.yaml

@@ -0,0 +1,96 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: probe.orionx-k8s-device-plugin
+  namespace: {{ .Release.Namespace }}
+data:
+  probe.sh: |
+    #!/bin/bash
+    # Author: yuanzhilei@virtaitech.com
+    #
+    health=$(pidof k8s-device-plugin)
+    if [[ -z $health ]]; then
+        echo "k8s-device-plugin not exists."
+        exit 1
+    fi
+---
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: orionx-k8s-device-plugin
+  namespace: {{ .Release.Namespace }}
+  labels:
+    app: orionx-k8s-device-plugin
+spec:
+  selector:
+    matchLabels:
+      app: orionx-k8s-device-plugin
+  updateStrategy:
+    type: RollingUpdate
+  template:
+    metadata:
+      labels:
+        app: orionx-k8s-device-plugin
+        orionx: orionx-k8s-device-plugin
+    spec:
+      priorityClassName: system-cluster-critical
+      tolerations:
+      - key: node-role.kubernetes.io/master
+        effect: NoSchedule
+      hostNetwork: true
+      dnsPolicy: ClusterFirstWithHostNet
+      containers:
+      - image: eball/orionx-k8s-device-plugin:3.1.5-20230707
+        imagePullPolicy: Always
+        name: device-plugin
+        livenessProbe:
+          exec:
+            command:
+            - /tmp/probe.sh
+          periodSeconds: 10
+          initialDelaySeconds: 60
+        securityContext:
+          allowPrivilegeEscalation: false
+          capabilities:
+            drop: ["ALL"]
+        # command: ["bash", "-c", "sleep 1d"]
+        env:
+        - name : ORION_CONTROLLER
+          value : {{ .Values.gpu.server }}
+        - name: ORION_VGPU_RESNAME
+          value: "virtaitech.com/gpu"
+        - name: NVIDIA_GPU_RESNAME
+          value: "nvidia.com/gpu"
+        - name: ORION_GMEM_UNIT
+          value: "1024"
+        - name: ORION_MNT_PATH
+          value: '{"/var/tmp/orion/comm/":"/var/tmp/orion/comm:ro"}'
+        - name: ORION_ENV
+          value: '{"ORION_DRIVER_ONLY":"true","ORION_DEVICE_ENABLE":"1","ORION_EXPORT_CMD":"env | grep ORION; orion-smi -j","ORION_NV_SMI_FORCE_DISPLAY":"1"}'
+        - name: REPORT_ONLY_LOCAL_VGPU
+          value: "false"
+        - name: DP_DISABLE_HEALTHCHECKS
+          value: "all"
+        - name: ORION_HOST_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
+        volumeMounts:
+        - name: device-plugin
+          mountPath: /var/lib/kubelet/device-plugins
+        - name: probe
+          mountPath: /tmp/probe.sh
+          subPath: probe.sh
+        - name: localtime
+          mountPath: /etc/localtime
+      volumes:
+      - name: localtime
+        hostPath:
+          path: /etc/localtime
+      - name: device-plugin
+        hostPath:
+          path: /var/lib/kubelet/device-plugins
+      - name: probe
+        configMap:
+          name: probe.orionx-k8s-device-plugin
+          defaultMode: 0755