add log

Author: sonic

internal/computing/k8s_service.go

@@ -717,15 +717,17 @@ type GpuData struct {
 	UsedIndex []string
 }
 
-func (s *K8sService) GetNodeGpuSummary(ctx context.Context) (map[string]map[string]GpuData, error) {
+func (s *K8sService) GetNodeGpuSummary(ctx context.Context) (map[string]map[string]GpuData, map[string]string, error) {
 	nodeGpuInfoMap, err := s.GetResourceExporterPodLog(ctx)
 	if err != nil {
 		logs.GetLogger().Errorf("Collect cluster gpu info Failed, if have available gpu, please check resource-exporter. error: %+v", err)
-		return nil, err
+		return nil, nil, err
 	}
 
+	var nodeMachineId = make(map[string]string)
 	var nodeGpuSummary = make(map[string]map[string]GpuData)
 	for nodeName, gpu := range nodeGpuInfoMap {
+		nodeMachineId[nodeName] = gpu.MachineId + "&" + gpu.ProductUuid
 		if gpu.Gpu.AttachedGpus > 0 {
 			var nodeGpu = make(map[string]GpuData)
 			for _, g := range gpu.Gpu.Details {
@@ -756,7 +758,7 @@ func (s *K8sService) GetNodeGpuSummary(ctx context.Context) (map[string]map[stri
 			nodeGpuSummary[nodeName] = nodeGpu
 		}
 	}
-	return nodeGpuSummary, nil
+	return nodeGpuSummary, nodeMachineId, nil
 }
 
 func (s *K8sService) GetAllActivePod(ctx context.Context) ([]coreV1.Pod, error) {

internal/computing/space_service.go

@@ -1566,7 +1566,7 @@ func checkResourceAvailableForSpace(jobUuid string, jobType int, resourceConfig
 		return false, "", nil, 0, nil, err
 	}
 
-	nodeGpuSummary, err := k8sService.GetNodeGpuSummary(context.TODO())
+	nodeGpuSummary, nodeNameMachineId, err := k8sService.GetNodeGpuSummary(context.TODO())
 	if err != nil {
 		logs.GetLogger().Errorf("Failed collect k8s gpu, error: %+v", err)
 		return false, "", nil, 0, nil, err
@@ -1577,7 +1577,7 @@ func checkResourceAvailableForSpace(jobUuid string, jobType int, resourceConfig
 	needCpu := hardwareDetail.Cpu.Quantity
 	needMemory := float64(hardwareDetail.Memory.Quantity)
 	needStorage := float64(hardwareDetail.Storage.Quantity)
-	logs.GetLogger().Infof("checkResourceForSpace: needCpu: %d, needMemory: %.2f, needStorage: %.2f, needGpu: %s, gpuNum: %d", needCpu, needMemory, needStorage, gpuName, gpuNum)
+	logs.GetLogger().Infof("job_uuid: %s, checkResourceForSpace: needCpu: %d, needMemory: %.2f, needStorage: %.2f, needGpu: %s, gpuNum: %d", jobUuid, needCpu, needMemory, needStorage, gpuName, gpuNum)
 
 	type gpuData struct {
 		Total     int
@@ -1603,7 +1603,7 @@ func checkResourceAvailableForSpace(jobUuid string, jobType int, resourceConfig
 			}
 		}
 
-		logs.GetLogger().Infof("checkResourceForSpace: nodeName: %s,remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f, remainingGpu: %+v", node.Name, remainderCpu, remainderMemory, remainderStorage, freeGpuMap)
+		logs.GetLogger().Infof("nodeName: %s, machineId&productUuid: %s, remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f, remainingGpu: %+v", node.Name, nodeNameMachineId[node.Name], remainderCpu, remainderMemory, remainderStorage, freeGpuMap)
 
 		if remainderCpu < needCpu {
 			noAvailableStr = append(noAvailableStr, fmt.Sprintf("cpu need: %d, remainder: %d", needCpu, remainderCpu))
@@ -1672,7 +1672,7 @@ func checkResourceAvailableForImage(jobUuid string, hardwareType string, resourc
 		return false, "", nil, nil, nil, err
 	}
 
-	nodeGpuSummary, err := k8sService.GetNodeGpuSummary(context.TODO())
+	nodeGpuSummary, nodeNameMachineId, err := k8sService.GetNodeGpuSummary(context.TODO())
 	if err != nil {
 		logs.GetLogger().Errorf("Failed collect k8s gpu, error: %+v", err)
 		return false, "", nil, nil, nil, err
@@ -1714,7 +1714,7 @@ func checkResourceAvailableForImage(jobUuid string, hardwareType string, resourc
 			}
 		}
 
-		logs.GetLogger().Infof("checkResourceForSpace: nodeName: %s,remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f, remainingGpu: %+v", node.Name, remainderCpu, remainderMemory, remainderStorage, freeGpuMap)
+		logs.GetLogger().Infof("nodeName: %s, machineId&productUuid: %s, remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f, remainingGpu: %+v", node.Name, nodeNameMachineId[node.Name], remainderCpu, remainderMemory, remainderStorage, freeGpuMap)
 
 		if remainderCpu < needCpu {
 			noAvailableStr = append(noAvailableStr, fmt.Sprintf("cpu need: %d, remainder: %d", needCpu, remainderCpu))
@@ -1810,7 +1810,7 @@ func checkResourceAvailableForUbi(taskId, taskType int, gpuName string, resource
 		return "", "", 0, 0, 0, nil, nil, err
 	}
 
-	nodeGpuSummary, err := k8sService.GetNodeGpuSummary(context.TODO())
+	nodeGpuSummary, nodeNameMachineId, err := k8sService.GetNodeGpuSummary(context.TODO())
 	if err != nil {
 		logs.GetLogger().Errorf("Failed collect k8s gpu, error: %+v", err)
 		return "", "", 0, 0, 0, nil, nil, err
@@ -1842,7 +1842,7 @@ func checkResourceAvailableForUbi(taskId, taskType int, gpuName string, resource
 		remainderStorage := float64(remainderResource[ResourceStorage] / 1024 / 1024 / 1024)
 
 		logs.GetLogger().Infof("checkResourceAvailableForUbi: needCpu: %d, needMemory: %.2f, needStorage: %.2f", needCpu, needMemory, needStorage)
-		logs.GetLogger().Infof("checkResourceAvailableForUbi: remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f", remainderCpu, remainderMemory, remainderStorage)
+		logs.GetLogger().Infof("nodeName: %s, machineId&productUuid: %s, remainingCpu: %d, remainingMemory: %.2f, remainingStorage: %.2f", node.Name, nodeNameMachineId[node.Name], remainderCpu, remainderMemory, remainderStorage)
 
 		if remainderCpu < needCpu {
 			noAvailableStr = append(noAvailableStr, fmt.Sprintf("cpu need: %d, remainder: %d", needCpu, remainderCpu))

internal/models/resources.go

@@ -23,8 +23,10 @@ type NodeResource struct {
 }
 
 type CollectNodeInfo struct {
-	Gpu     Gpu    `json:"gpu"`
-	CpuName string `json:"cpu_name"`
+	Gpu         Gpu    `json:"gpu"`
+	CpuName     string `json:"cpu_name"`
+	ProductUuid string `json:"product_uuid"`
+	MachineId   string `json:"machine_id"`
 }
 type Gpu struct {
 	DriverVersion string      `json:"driver_version"`