Merge pull request #4 from fourhu/master

Fixed the issue where the DCU device plugin failed to run on K100
Project-HAMi · Dec 18, 2024 · 7c0f320 · 7c0f320
2 parents af86d04 + f7408f3
commit 7c0f320
Show file tree

Hide file tree

Showing 3 changed files with 53 additions and 43 deletions.
diff --git a/Dockerfile b/Dockerfile
@@ -20,7 +20,7 @@ RUN cd /device-plugin && go build -o ./k8s-device-plugin cmd/k8s-device-plugin/m
 FROM ubuntu:20.04
 ENV TZ=Asia/Dubai
 RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
-RUN apt-get update && apt-get -y install libhwloc-dev libdrm-dev pciutils
+RUN apt-get update && apt-get -y install libhwloc-dev libdrm-dev pciutils libelf-dev kmod
 ENV LD_LIBRARY_PATH=/opt/hygondriver/hip/lib:/opt/hygondriver/llvm/lib:/opt/hygondriver/lib:/opt/hygondriver/lib64:/opt/hyhal/lib:/opt/hyhal/lib64:/opt/hygondriver/.hyhal/lib:/opt/hygondriver/.hyhal/lib64:
 ENV PATH=/opt/hygondriver/bin:/opt/hygondriver/llvm/bin:/opt/hygondriver/hip/bin:/opt/hygondriver/hip/bin/hipify:/opt/hyhal/bin:/opt/hygondriver/.hyhal/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
 ENV C_INCLUDE_PATH=/opt/hygondriver/include:/opt/hyhal/include:/opt/hygondriver/llvm/include:/opt/hygondriver/.hyhal/include:

diff --git a/internal/pkg/dcu/server.go b/internal/pkg/dcu/server.go
@@ -95,7 +95,7 @@ func (p *Plugin) Start() error {
 	}
 	p.count = 0
 
-	cmd := exec.Command("hy-smi", "--showmeminfo", "vram")
+	cmd := exec.Command("/opt/hyhal/bin/hy-smi", "--showmeminfo", "vram")
 	out, err := cmd.CombinedOutput()
 	if err != nil {
 		log.Fatalf("cmd.Run() failed with %s\n", err)
@@ -124,7 +124,7 @@ func (p *Plugin) Start() error {
 		p.count++
 	}
 
-	cmd = exec.Command("hy-smi", "--showproductname")
+	cmd = exec.Command("/opt/hyhal/bin/hy-smi", "--showproductname")
 	out, err = cmd.CombinedOutput()
 	if err != nil {
 		log.Fatalf("cmd.Run() failed with %s\n", err)
@@ -148,7 +148,7 @@ func (p *Plugin) Start() error {
 		index++
 	}
 
-	cmd = exec.Command("hy-smi", "--showbus")
+	cmd = exec.Command("/opt/hyhal/bin/hy-smi", "--showbus")
 	out, err = cmd.CombinedOutput()
 	if err != nil {
 		log.Fatalf("cmd.Run() failed with %s\n", err)
@@ -167,7 +167,7 @@ func (p *Plugin) Start() error {
 	}
 	fmt.Println("collecting pcibus=", p.pcibusid)
 
-	cmd = exec.Command("hy-virtual", "--show-device-info")
+	cmd = exec.Command("/opt/hyhal/bin/hy-virtual", "--show-device-info")
 	out, err = cmd.CombinedOutput()
 	if err != nil {
 		log.Fatalf("cmd.Run() failed with %s\n", err)

diff --git a/k8s-dcu-plugin.yaml b/k8s-dcu-plugin.yaml
@@ -21,55 +21,65 @@ spec:
       - key: CriticalAddonsOnly
         operator: Exists
       containers:
-      - image: projecthami/dcu-vgpu-device-plugin:v1.0.1
+      - image: projecthami/dcu-vgpu-device-plugin:master
         #command: ["/bin/bash","-c","source /opt/hygondriver/env.sh && sleep infinity"]
-        command: ["/root/k8s-device-plugin"]
+        command: [ "/root/k8s-device-plugin" ]
         name: dcu-dp-cntr
         env:
-          - name: NODE_NAME
-            valueFrom:
-              fieldRef:
-                fieldPath: spec.nodeName
-          - name: HYGONPATH
-            value: /opt/dtk
-          - name: BASH_ENV
-            value: ~/.bashrc
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: HYGONPATH
+          value: /opt/dtk
+        - name: BASH_ENV
+          value: ~/.bashrc
         securityContext:
           privileged: true
           allowPrivilegeEscalation: true
           capabilities:
-            drop: ["ALL"]
-            add: ["SYS_ADMIN"]
+            drop: [ "ALL" ]
+            add: [ "SYS_ADMIN" ]
         volumeMounts:
-          - name: dp
-            mountPath: /var/lib/kubelet/device-plugins
-          - name: sys
-            mountPath: /sys
-          - name: hwpath
-            mountPath: /usr/share/hwdata
-          - name: hygonloc
-            mountPath: /opt/hygondriver/
-          - name: lib
-            mountPath: /usr/local/vgpu
-          - name: hyhal
-            mountPath: /opt/hyhal
-      volumes:
         - name: dp
-          hostPath:
-            path: /var/lib/kubelet/device-plugins
+          mountPath: /var/lib/kubelet/device-plugins
         - name: sys
-          hostPath:
-            path: /sys
+          mountPath: /sys
+        - name: dev
+          mountPath: /dev
+        - name: vdev
+          mountPath: /etc/vdev
         - name: hwpath
-          hostPath:
-            path: /usr/share/hwdata
+          mountPath: /usr/share/hwdata
         - name: hygonloc
-          hostPath:
-            path: /opt/dtk
+          mountPath: /opt/hygondriver/
         - name: lib
-          hostPath:
-            path: /usr/local/vgpu
+          mountPath: /usr/local/vgpu
         - name: hyhal
-          hostPath:
-            path: /opt/hyhal
-
+          mountPath: /opt/hyhal
+      volumes:
+      - name: dp
+        hostPath:
+          path: /var/lib/kubelet/device-plugins
+      - name: sys
+        hostPath:
+          path: /sys
+      - name: dev
+        hostPath:
+          path: /dev
+      - name: vdev
+        hostPath:
+          path: /etc/vdev
+          type: Directory
+      - name: hwpath
+        hostPath:
+          path: /usr/share/hwdata
+      - name: hygonloc
+        hostPath:
+          path: /opt/dtk
+      - name: lib
+        hostPath:
+          path: /usr/local/vgpu
+      - name: hyhal
+        hostPath:
+          path: /opt/hyhal