Support iluvatar devices

charts/hami-webui/values.yaml

@@ -10,6 +10,7 @@ vendorNodeSelectors:
   DCU: dcu=on
   MLU: mlu=on
   Metax: metax-tech.com/gpu.installed=true
+  Iluvatar: iluvatar.com/gpu.present=true
 
 image:
   frontend:

packages/web/projects/vgpu/views/task/admin/Detail.vue

@@ -42,11 +42,11 @@
   </block-box>
 
   <block-box v-for="{ title, data } in lineConfig" :key="title" :title="title">
-    <template #extra v-if="detail.type && (detail.type.startsWith('NVIDIA') || detail.type.startsWith('MXC'))">
+    <template #extra v-if="detail.type && (detail.type.startsWith('NVIDIA') || detail.type.startsWith('MXC') || detail.type.startsWith('BI-') || detail.type.startsWith('MR-'))">
       <time-picker v-model="times" type="datetimerange" size="small" />
     </template>
     <div style="height: 200px">
-      <template v-if="detail.type && !detail.type.startsWith('NVIDIA') && !detail.type.startsWith('MXC')">
+      <template v-if="detail.type && !detail.type.startsWith('NVIDIA') && !detail.type.startsWith('MXC') && !detail.type.startsWith('BI-') && !detail.type.startsWith('MR-')">
         <el-empty :description="$t('task.noMonitorSupport')" :image-size="60" />
       </template>
       <template v-else>

server/config/config.yaml

@@ -13,4 +13,5 @@ node_selectors:
   Ascend: ascend=on
   DCU: dcu=on
   MLU: mlu=on
-  Metax: metax-tech.com/gpu.installed=true
+  Metax: metax-tech.com/gpu.installed=true
+  Iluvatar: iluvatar.com/gpu.present=true

server/internal/biz/biz.go

@@ -12,10 +12,11 @@ var ProviderSet = wire.NewSet(
 )
 
 const (
-	NvidiaGPUDevice = "NVIDIA"
-	HygonGPUDevice  = "DCU"
-	AscendGPUDevice = "Ascend"
-	MetaxGPUDevice  = "Metax"
+	NvidiaGPUDevice   = "NVIDIA"
+	HygonGPUDevice    = "DCU"
+	AscendGPUDevice   = "Ascend"
+	MetaxGPUDevice    = "Metax"
+	IluvatarGPUDevice = "Iluvatar"
 
 	CambriconGPUDevice = "MLU"
 

server/internal/data/node.go

@@ -4,22 +4,24 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"github.com/go-kratos/kratos/v2/log"
-	corev1 "k8s.io/api/core/v1"
-	k8stypes "k8s.io/apimachinery/pkg/types"
-	"k8s.io/client-go/informers"
-	listerscorev1 "k8s.io/client-go/listers/core/v1"
-	"k8s.io/client-go/tools/cache"
 	"strings"
 	"sync"
 	"time"
 	"vgpu/internal/biz"
 	"vgpu/internal/provider"
 	"vgpu/internal/provider/ascend"
 	"vgpu/internal/provider/hygon"
+	"vgpu/internal/provider/iluvatar"
 	"vgpu/internal/provider/metax"
 	"vgpu/internal/provider/mlu"
 	"vgpu/internal/provider/nvidia"
+
+	"github.com/go-kratos/kratos/v2/log"
+	corev1 "k8s.io/api/core/v1"
+	k8stypes "k8s.io/apimachinery/pkg/types"
+	"k8s.io/client-go/informers"
+	listerscorev1 "k8s.io/client-go/listers/core/v1"
+	"k8s.io/client-go/tools/cache"
 )
 
 type nodeRepo struct {
@@ -45,6 +47,7 @@ func NewNodeRepo(data *Data, nodeSelectors map[string]string, logger log.Logger)
 			ascend.NewAscend(data.promCl, log.NewHelper(logger), nodeSelectors[biz.AscendGPUDevice]),
 			hygon.NewHygon(data.promCl, log.NewHelper(logger), nodeSelectors[biz.HygonGPUDevice]),
 			metax.NewMetax(data.promCl, log.NewHelper(logger), nodeSelectors[biz.MetaxGPUDevice]),
+			iluvatar.NewIluvatar(data.promCl, log.NewHelper(logger), nodeSelectors[biz.IluvatarGPUDevice]),
 		},
 	}
 	nodeRepo.init()
@@ -79,7 +82,7 @@ func (r *nodeRepo) updateLocalNodes() {
 					continue
 				}
 				for _, device := range devices {
-					n[node.UID].Devices = append(bizNode.Devices, &biz.DeviceInfo{
+					n[node.UID].Devices = append(n[node.UID].Devices, &biz.DeviceInfo{
 						Index:    int(device.Index),
 						Id:       device.ID,
 						AliasId:  device.AliasId,

server/internal/exporter/exporter.go

@@ -10,6 +10,7 @@ import (
 	pb "vgpu/api/v1"
 	"vgpu/internal/biz"
 	"vgpu/internal/data/prom"
+	"vgpu/internal/provider/iluvatar"
 	"vgpu/internal/provider/metax"
 	"vgpu/internal/provider/mlu"
 	"vgpu/internal/service"
@@ -234,6 +235,9 @@ func (s *MetricsGenerator) GenerateContainerMetrics(ctx context.Context) error {
 			if provider == "" || provider == metax.MetaxGPUDevice {
 				continue
 			}
+			if mapped, ok := iluvatar.IluvatarProviders[provider]; ok {
+				provider = mapped
+			}
 			HamiContainerVgpuAllocated.WithLabelValues(device.NodeName, provider, device.Type, device.Id, c.PodName, c.Name, c.Namespace, fmt.Sprintf("%s:%s", c.Name, c.PodUID)).Set(float64(vGPU))
 			HamiContainerVmemoryAllocated.WithLabelValues(device.NodeName, provider, device.Type, device.Id, c.PodName, c.Name, c.Namespace, fmt.Sprintf("%s:%s", c.Name, c.PodUID)).Set(float64(memory))
 			HamiContainerVcoreAllocated.WithLabelValues(device.NodeName, provider, device.Type, device.Id, c.PodName, c.Name, c.Namespace, fmt.Sprintf("%s:%s", c.Name, c.PodUID)).Set(float64(core))
@@ -255,6 +259,9 @@ func (s *MetricsGenerator) GenerateContainerMetrics(ctx context.Context) error {
 				case metax.MetaxSGPUDevice:
 					used = float64(taskCoreUsed)
 					util = roundToOneDecimal(100 * float64(taskCoreUsed) / float64(core))
+				case biz.IluvatarGPUDevice:
+					used = float64(taskCoreUsed)
+					util = roundToOneDecimal(100 * float64(taskCoreUsed) / float64(core))
 				default:
 				}
 				cardCoreUtil, err := s.deviceCoreUtil(ctx, provider, device.Id)
@@ -272,6 +279,8 @@ func (s *MetricsGenerator) GenerateContainerMetrics(ctx context.Context) error {
 					taskMemoryUsed = float32((taskMemoryUsed/100)*float32(memory)) * 1024 * 1024
 				case metax.MetaxSGPUDevice:
 					taskMemoryUsed = float32(taskMemoryUsed) * 1024 // KB->Byte
+				case biz.IluvatarGPUDevice:
+					taskMemoryUsed = float32((taskMemoryUsed/100)*float32(memory)) * 1024 * 1024
 				default:
 				}
 				HamiContainerMemoryUsed.WithLabelValues(device.NodeName, provider, device.Type, device.Id, c.PodName, c.Name, c.Namespace).Set(float64(taskMemoryUsed / 1024 / 1024))
@@ -309,6 +318,8 @@ func (s *MetricsGenerator) deviceMemUsed(ctx context.Context, provider, deviceUU
 		query = fmt.Sprintf("avg(dcu_usedmemory_bytes{device_id=\"%s\"})", deviceUUID)
 	case biz.MetaxGPUDevice:
 		query = fmt.Sprintf("avg(mx_memory_used{uuid=\"%s\", type=\"vram\"})", deviceUUID)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(ix_mem_used{uuid=\"%s\"})", deviceUUID)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -340,6 +351,8 @@ func (s *MetricsGenerator) deviceMemTotal(ctx context.Context, provider, deviceU
 		query = fmt.Sprintf("avg(dcu_memorycap_bytes{device_id=\"%s\"})", deviceUUID)
 	case biz.MetaxGPUDevice:
 		query = fmt.Sprintf("avg(mx_memory_total{uuid=\"%s\", type=\"vram\"})", deviceUUID)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(ix_mem_total{uuid=\"%s\"})", deviceUUID)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -371,6 +384,8 @@ func (s *MetricsGenerator) deviceCoreUtil(ctx context.Context, provider, deviceU
 		query = fmt.Sprintf("avg(dcu_utilizationrate{device_id=\"%s\"})", deviceUUID)
 	case biz.MetaxGPUDevice, metax.MetaxGPUDevice, metax.MetaxSGPUDevice:
 		query = fmt.Sprintf("avg(mx_gpu_usage{uuid=\"%s\"})", deviceUUID)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(ix_gpu_utilization{uuid=\"%s\"})", deviceUUID)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -401,6 +416,8 @@ func (s *MetricsGenerator) taskCoreUsed(ctx context.Context, provider, namespace
 	case metax.MetaxSGPUDevice:
 		query = fmt.Sprintf("avg(mx_sgpu_usage{Hostname=\"%s\", deviceId=\"%d\", exported_namespace=\"%s\", exported_pod=\"%s\", exported_container=\"%s\"})",
 			hostname, deviceIndex, namespace, pod, container)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(container_gpu_utilization{pod_name=\"%s\", container_name=\"%s\"})", pod, container)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -424,6 +441,9 @@ func (s *MetricsGenerator) taskMemoryUsed(ctx context.Context, provider, namespa
 	case metax.MetaxSGPUDevice:
 		query = fmt.Sprintf("avg(mx_sgpu_used_memory{Hostname=\"%s\", deviceId=\"%d\", exported_namespace=\"%s\", exported_pod=\"%s\", exported_container=\"%s\"})",
 			hostname, deviceIndex, namespace, pod, container)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(container_gpu_memory_total{container_name=\"%s\", pod_name=\"%s\"})",
+			container, pod)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -444,6 +464,8 @@ func (s *MetricsGenerator) gpuTemperature(ctx context.Context, provider, deviceU
 		query = fmt.Sprintf("avg(dcu_temp{device_id=\"%s\"})", deviceUUID)
 	case biz.MetaxGPUDevice:
 		query = fmt.Sprintf("avg(mx_chip_hotspot_temp{uuid=\"%s\"})", deviceUUID)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(ix_temperature{uuid=\"%s\"})", deviceUUID)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -480,6 +502,8 @@ func (s *MetricsGenerator) gpuPower(ctx context.Context, provider, deviceUUID st
 		query = fmt.Sprintf("avg(dcu_power_usage{device_id=\"%s\"})", deviceUUID)
 	case biz.MetaxGPUDevice:
 		query = fmt.Sprintf("avg(mx_board_power{uuid=\"%s\"})", deviceUUID) // mW
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("avg(ix_power_usage{uuid=\"%s\"})", deviceUUID)
 	default:
 		return 0, errors.New("provider not exists")
 	}
@@ -538,6 +562,9 @@ func (s *MetricsGenerator) queryDeviceAdditional(ctx context.Context, provider,
 		query = fmt.Sprintf("dcu_power_usage{device_id=\"%s\"}", deviceUUID)
 	case biz.MetaxGPUDevice:
 		query = fmt.Sprintf("mx_board_power{uuid=\"%s\"}", deviceUUID)
+	case biz.IluvatarGPUDevice:
+		query = fmt.Sprintf("ix_power_usage{uuid=\"%s\"}", deviceUUID)
+
 	default:
 		return nil, errors.New("provider not exists")
 	}
@@ -567,6 +594,10 @@ func (s *MetricsGenerator) queryDeviceAdditional(ctx context.Context, provider,
 		case biz.MetaxGPUDevice:
 			info.DriverVersion = metric["driver_version"]
 			info.DeviceNo = metric["deviceId"]
+		case biz.IluvatarGPUDevice:
+			info.DriverVersion = metric["driver"]
+			info.DeviceNo = metric["serial"]
+
 		}
 		return info, nil
 	}

server/internal/provider/iluvatar/device.go

@@ -0,0 +1,36 @@
+package iluvatar
+
+import "vgpu/internal/provider/util"
+
+const (
+	IluvatarBI150Device = "Iluvatar-BI-V150"
+	IluvatarBI100Device = "Iluvatar-BI-V100"
+	IluvatarMR100Device = "Iluvatar-MR-V100"
+
+	BI150NodeRegisterAnno = "hami.io/node-BI-V150-register"
+	MR100NodeRegisterAnno = "hami.io/node-MR-V100-register"
+	BI100NodeRegisterAnno = "hami.io/node-BI-V100-register"
+)
+
+var (
+	IluvatarResourceCount     string
+	IluvatarResourceMemory    string
+	IluvatarResourceCores     string
+	IluvatarNodeRegisterAnnos []string
+	IluvatarProviders         map[string]string
+)
+
+func init() {
+	IluvatarProviders = map[string]string{
+		"BI-V150": "Iluvatar",
+		"BI-V100": "Iluvatar",
+		"MR-V100": "Iluvatar",
+	}
+	IluvatarNodeRegisterAnnos = []string{BI150NodeRegisterAnno, MR100NodeRegisterAnno, BI100NodeRegisterAnno}
+	util.InRequestDevices[IluvatarBI150Device] = "hami.io/BI-V150-devices-to-allocate"
+	util.SupportDevices[IluvatarBI150Device] = "hami.io/BI-V150-devices-allocated"
+	util.InRequestDevices[IluvatarBI100Device] = "hami.io/BI-V100-devices-to-allocate"
+	util.SupportDevices[IluvatarBI100Device] = "hami.io/BI-V100-devices-allocated"
+	util.InRequestDevices[IluvatarMR100Device] = "hami.io/MR-V100-devices-to-allocate"
+	util.SupportDevices[IluvatarMR100Device] = "hami.io/MR-V100-devices-allocated"
+}

server/internal/provider/iluvatar/provider.go

@@ -0,0 +1,54 @@
+package iluvatar
+
+import (
+	"fmt"
+	"vgpu/internal/biz"
+	"vgpu/internal/data/prom"
+	"vgpu/internal/provider/util"
+
+	"github.com/go-kratos/kratos/v2/log"
+	corev1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/labels"
+)
+
+type Iluvatar struct {
+	prom *prom.Client
+	log  *log.Helper
+
+	nodeSelectors string
+}
+
+func NewIluvatar(prom *prom.Client, log *log.Helper, nodeSelectors string) *Iluvatar {
+	return &Iluvatar{
+		prom:          prom,
+		log:           log,
+		nodeSelectors: nodeSelectors,
+	}
+}
+
+func (i *Iluvatar) GetNodeDevicePluginLabels() (labels.Selector, error) {
+	return labels.Parse(i.nodeSelectors)
+}
+
+func (i *Iluvatar) GetProvider() string {
+	return biz.IluvatarGPUDevice
+}
+
+type DeviceMeta struct {
+	UUID   string
+	Type   string
+	Driver string
+}
+
+func (i *Iluvatar) FetchDevices(node *corev1.Node) ([]*util.DeviceInfo, error) {
+	for _, anno := range IluvatarNodeRegisterAnnos {
+		anno, ok := node.Annotations[anno]
+		if !ok {
+			log.Infof("anno %s not found", anno)
+			continue
+		}
+		deviceInfos, err := util.DecodeNodeDevices(anno, i.log)
+		return deviceInfos, err
+	}
+	return []*util.DeviceInfo{}, fmt.Errorf("")
+}

server/internal/provider/util/util.go

@@ -25,6 +25,9 @@ const (
 	CambriconGPUDevice  = "MLU"
 	MetaxGPUDevice      = "Metax-GPU"
 	MetaxSGPUDevice     = "Metax-SGPU"
+	IluvatarBI150Device = "Iluvatar-BI-V150"
+	IluvatarBI100Device = "Iluvatar-BI-V100"
+	IluvatarMR100Device = "Iluvatar-MR-V100"
 
 	DsmluProfileAndInstance = "CAMBRICON_DSMLU_PROFILE_INSTANCE"
 
@@ -391,6 +394,24 @@ func DecodePodDevices(pod *corev1.Pod, log *log.Helper) (PodDevices, error) {
 				}
 				pd[devType] = append(pd[devType], cd)
 			}
+		case IluvatarBI100Device, IluvatarBI150Device, IluvatarMR100Device:
+			for i, s := range strings.Split(str, OnePodMultiContainerSplitSymbol) {
+				if i >= len(pod.Spec.Containers) {
+					break
+				}
+				if s == "" {
+					pd[devType] = append(pd[devType], ContainerDevices{})
+					continue
+				}
+				cd, err := DecodeContainerDevices(s, priorities[i])
+				if err != nil {
+					return PodDevices{}, nil
+				}
+				if len(cd) == 0 {
+					continue
+				}
+				pd[devType] = append(pd[devType], cd)
+			}
 		}
 	}
 	log.Infof("Decoded pod annos: poddevices %v", pd)