Support topology-awareness for Kunlunxin device (#1121)

Support topology-awareness for Kunlunxin device

Signed-off-by: limengxuan <[email protected]>

Author: archlitchi

charts/hami/templates/scheduler/configmap.yaml

@@ -49,6 +49,14 @@ data:
                     },
                     {{- end }}
                     {{- end }}
+                    {{- if .Values.devices.kunlun.enabled }}
+                    {{- range .Values.devices.kunlun.customresources }}
+                    {
+                      "name": "{{ . }}",
+                      "ignoredByScheduler": true
+                    },
+                    {{- end }}
+                    {{- end }}
                     {
                         "name": "{{ .Values.resourceName }}",
                         "ignoredByScheduler": true

charts/hami/templates/scheduler/configmapnew.yaml

@@ -76,4 +76,10 @@ data:
         ignoredByScheduler: true
       {{- end }}
       {{- end }}
+      {{- if .Values.devices.kunlun.enabled }}
+      {{- range .Values.devices.kunlun.customresources }}
+      - name: {{ . }}
+        ignoredByScheduler: true
+      {{- end }}
+      {{- end }}
 {{- end }}

charts/hami/values.yaml

@@ -204,13 +204,17 @@ devicePlugin:
 #        memory: 100Mi
 
 devices:
+  kunlun:
+    enabled: true
+    customresources:
+      - kunlunxin.com/xpu
   enflame:
-    enabled: false
+    enabled: true
     customresources:
       - enflame.com/vgcu
       - enflame.com/vgcu-percentage
   mthreads:
-    enabled: false
+    enabled: true
     customresources:
       - mthreads.com/vgpu
   nvidia:

pkg/device/cambricon/device.go

@@ -184,7 +184,10 @@ func (dev *CambriconDevices) CheckHealth(devType string, n *corev1.Node) (bool,
 func (dev *CambriconDevices) GetNodeDevices(n corev1.Node) ([]*util.DeviceInfo, error) {
 	nodedevices := []*util.DeviceInfo{}
 	i := 0
-	cards, _ := n.Status.Capacity.Name(corev1.ResourceName(MLUResourceCores), resource.DecimalSI).AsInt64()
+	cards, ok := n.Status.Capacity.Name(corev1.ResourceName(MLUResourceCores), resource.DecimalSI).AsInt64()
+	if !ok || cards == 0 {
+		return []*util.DeviceInfo{}, fmt.Errorf("device not found %s", MLUResourceCores)
+	}
 	memoryTotal, _ := n.Status.Capacity.Name(corev1.ResourceName(MLUResourceMemory), resource.DecimalSI).AsInt64()
 	for int64(i)*100 < cards {
 		nodedevices = append(nodedevices, &util.DeviceInfo{

pkg/device/enflame/device.go

@@ -118,7 +118,7 @@ func (dev *EnflameDevices) GetNodeDevices(n corev1.Node) ([]*util.DeviceInfo, er
 	i := 0
 	cards, ok := n.Status.Capacity.Name(corev1.ResourceName(CountNoSharedName), resource.DecimalSI).AsInt64()
 	if !ok || cards == 0 {
-		return nodedevices, nil
+		return []*util.DeviceInfo{}, fmt.Errorf("device not found %s", CountNoSharedName)
 	}
 	shared, _ := n.Status.Capacity.Name(corev1.ResourceName(SharedResourceName), resource.DecimalSI).AsInt64()
 	dev.factor = int(shared / cards)

pkg/device/iluvatar/device.go

@@ -89,7 +89,10 @@ func (dev *IluvatarDevices) MutateAdmission(ctr *corev1.Container, p *corev1.Pod
 func (dev *IluvatarDevices) GetNodeDevices(n corev1.Node) ([]*util.DeviceInfo, error) {
 	nodedevices := []*util.DeviceInfo{}
 	i := 0
-	cards, _ := n.Status.Capacity.Name(corev1.ResourceName(IluvatarResourceCores), resource.DecimalSI).AsInt64()
+	cards, ok := n.Status.Capacity.Name(corev1.ResourceName(IluvatarResourceCores), resource.DecimalSI).AsInt64()
+	if !ok || cards == 0 {
+		return []*util.DeviceInfo{}, fmt.Errorf("device not found %s", IluvatarResourceCores)
+	}
 	memoryTotal, _ := n.Status.Capacity.Name(corev1.ResourceName(IluvatarResourceMemory), resource.DecimalSI).AsInt64()
 	for int64(i)*100 < cards {
 		nodedevices = append(nodedevices, &util.DeviceInfo{