Merge pull request volcano-sh#4779 from kiritoxkiriko/fix/vnpu-hami-bug

Fix: vNPU HAMi mode may not work with multi-type NPUs cluster

Author: volcano-sh-bot

pkg/scheduler/api/node_info.go

@@ -374,7 +374,11 @@ func (ni *NodeInfo) setNodeOthersResource(node *v1.Node) {
 			ignored_list = append(ignored_list, devices.GetIgnoredDevices()...)
 		}
 	}
-	IgnoredDevicesList.Set(ignored_list)
+	klog.V(5).Infof("ignored_list is %v", ignored_list)
+	IgnoredDevicesList.AppendList(
+		ignored_list,
+	)
+	klog.V(5).Infof("ignoredDevicesList is %v", IgnoredDevicesList.List())
 }
 
 // setNode sets kubernetes node object to nodeInfo object without assertion

pkg/scheduler/api/shared_device_pool.go

@@ -17,6 +17,7 @@
 package api
 
 import (
+	"slices"
 	"sync"
 
 	v1 "k8s.io/api/core/v1"
@@ -104,7 +105,45 @@ func (l *ignoredDevicesList) Set(deviceLists ...[]string) {
 	defer l.Unlock()
 	l.ignoredDevices = l.ignoredDevices[:0]
 	for _, devices := range deviceLists {
-		l.ignoredDevices = append(l.ignoredDevices, devices...)
+		for _, device := range devices {
+			if device == "" {
+				continue
+			}
+			if exists := slices.Contains(l.ignoredDevices, device); exists {
+				continue
+			}
+			l.ignoredDevices = append(l.ignoredDevices, device)
+		}
+	}
+}
+
+func (l *ignoredDevicesList) Append(devices ...string) {
+	l.Lock()
+	defer l.Unlock()
+	for _, device := range devices {
+		if device == "" {
+			continue
+		}
+		if exists := slices.Contains(l.ignoredDevices, device); exists {
+			continue
+		}
+		l.ignoredDevices = append(l.ignoredDevices, device)
+	}
+}
+
+func (l *ignoredDevicesList) AppendList(devicesLists ...[]string) {
+	l.Lock()
+	defer l.Unlock()
+	for _, devices := range devicesLists {
+		for _, device := range devices {
+			if device == "" {
+				continue
+			}
+			if exists := slices.Contains(l.ignoredDevices, device); exists {
+				continue
+			}
+			l.ignoredDevices = append(l.ignoredDevices, device)
+		}
 	}
 }
 
@@ -117,3 +156,11 @@ func (l *ignoredDevicesList) Range(f func(i int, device string) bool) {
 		}
 	}
 }
+
+func (l *ignoredDevicesList) List() []string {
+	l.RLock()
+	defer l.RUnlock()
+	devices := make([]string, len(l.ignoredDevices))
+	copy(devices, l.ignoredDevices)
+	return devices
+}

pkg/scheduler/api/shared_device_pool_test.go

@@ -56,15 +56,16 @@ func Test_ignoredDevicesList_Set_BasicUsage(t *testing.T) {
 
 func Test_ignoredDevicesList_Range_BasicUsage(t *testing.T) {
 	lst := ignoredDevicesList{}
-	lst.Set([]string{"volcano.sh/vgpu-memory", "volcano.sh/vgpu-memory-percentage", "volcano.sh/vgpu-cores"})
+	expected := []string{"volcano.sh/vgpu-memory", "volcano.sh/vgpu-memory-percentage", "volcano.sh/vgpu-cores"}
+	lst.Set(expected)
 
 	t.Run("read and copy values from the ignoredDevicesList", func(t *testing.T) {
-		ignoredDevices := make([]string, 0, len(lst.ignoredDevices))
+		var ignoredDevices []string
 		lst.Range(func(_ int, device string) bool {
 			ignoredDevices = append(ignoredDevices, device)
 			return true
 		})
-		assert.Equal(t, lst.ignoredDevices, ignoredDevices)
+		assert.Equal(t, expected, ignoredDevices)
 	})
 
 	t.Run("break iteration through the ignoredDevicesList", func(t *testing.T) {
@@ -140,3 +141,31 @@ func Test_ignoredDevicesList_NoRace(t *testing.T) {
 	}
 	wg.Wait()
 }
+
+func Test_ignoredDevicesList_Set_DeduplicateAndOrder(t *testing.T) {
+	lst := ignoredDevicesList{}
+	lst.Set(
+		[]string{"volcano.sh/vgpu-memory", "volcano.sh/vgpu-memory"},
+		[]string{"", "volcano.sh/vgpu-cores", "volcano.sh/vgpu-memory"},
+	)
+
+	assert.Equal(t, []string{"volcano.sh/vgpu-memory", "volcano.sh/vgpu-cores"}, lst.List())
+}
+
+func Test_ignoredDevicesList_Append_PreservesOrder(t *testing.T) {
+	lst := ignoredDevicesList{}
+	lst.Set([]string{"volcano.sh/vgpu-memory"})
+	lst.Append("volcano.sh/vgpu-memory", "volcano.sh/vgpu-cores", "")
+
+	assert.Equal(t, []string{"volcano.sh/vgpu-memory", "volcano.sh/vgpu-cores"}, lst.List())
+}
+
+func Test_ignoredDevicesList_List_ReturnsCopy(t *testing.T) {
+	lst := ignoredDevicesList{}
+	lst.Set([]string{"volcano.sh/vgpu-memory"})
+
+	copyList := lst.List()
+	copyList[0] = "changed"
+
+	assert.Equal(t, []string{"volcano.sh/vgpu-memory"}, lst.List())
+}