diff --git a/adaptor/PCM-CORE/api/desc/core/pcm-core.api b/adaptor/PCM-CORE/api/desc/core/pcm-core.api index ea60f037..6b116442 100644 --- a/adaptor/PCM-CORE/api/desc/core/pcm-core.api +++ b/adaptor/PCM-CORE/api/desc/core/pcm-core.api @@ -206,6 +206,14 @@ type ( AllCardRunTime float64 `json:"allCardRunTime"` AllJobCount float64 `json:"allJobCount"` AllJobRunTime float64 `json:"allJobRunTime"` + TrainJobs []TrainJob `json:"trainJobs"` + } + TrainJob { + Name string `json:"name"` + Status string `json:"status"` + ServiceName string `json:"ServiceName"` + SynergyStatus string `json:"synergyStatus"` + Strategy int `json:"strategy"` } ) @@ -281,7 +289,7 @@ type ( type ( listClusterReq { - CenterId int32 `path:"centerId"` + CenterId int32 `path:"centerId"` } listClusterResp { Code int32 `json:"code"` @@ -310,8 +318,8 @@ type ( } GiResp { - CpuNum int32 `json:"cpuNum,optional"` - MemoryInGib int32 `json:"memoryInGib,optional"` + CpuNum int32 `json:"cpuNum,optional"` + MemoryInGib int32 `json:"memoryInGib,optional"` StorageInGib int32 `json:"storageInGib,optional"` } ) @@ -322,28 +330,28 @@ type ( DomainResourceList []DomainResource `json:"domainResourceList"` } DomainResource { - Id int64 `json:"id"` // id - DomainId string `json:"domainId"` // 资源域id - DomainName string `json:"domainName"` // 资源域名称 - JobCount int64 `json:"jobCount"` // 资源域任务数量 - DomainSource int64 `json:"domainSource"` // 资源域数据来源：0-nudt，1-鹏城 - Stack string `json:"stack"` // 技术栈 - ResourceType string `json:"resourceType"` // 资源类型 - Cpu float64 `json:"cpu"` // cpu使用率 - Memory float64 `json:"memory"` // 内存使用率 - Disk float64 `json:"disk"` // 存储使用率 - NodeCount float64 `json:"nodeCount"` //节点使用率 -// DeleteFlag int64 `json:"delete_flag"` // 是否删除 0:未删除，1:已经删除 - Description string `json:"description"` //集群描述 - ClusterName string `json:"clusterName"` //集群名称 - CpuTotal float64 `json:"cpuTotal"` //cpu总核数 - MemoryTotal float64 `json:"memoryTotal"` //内存总量Gi - DiskTotal float64 `json:"diskTotal"` //存储总量GB - NodeTotal float64 `json:"nodeTotal"` //容器节点数 - CpuUsage float64 `json:"cpuUsage"` //cpu已使用核数 - MemoryUsage float64 `json:"memoryUsage"` //内存已使用Gi - DiskUsage float64 `json:"diskUsage"` //存储已使用GB - NodeUsage float64 `json:"nodeUsage"` //容器节点已使用 + Id int64 `json:"id"` // id + DomainId string `json:"domainId"` // 资源域id + DomainName string `json:"domainName"` // 资源域名称 + JobCount int64 `json:"jobCount"` // 资源域任务数量 + DomainSource int64 `json:"domainSource"` // 资源域数据来源：0-nudt，1-鹏城 + Stack string `json:"stack"` // 技术栈 + ResourceType string `json:"resourceType"` // 资源类型 + Cpu float64 `json:"cpu"` // cpu使用率 + Memory float64 `json:"memory"` // 内存使用率 + Disk float64 `json:"disk"` // 存储使用率 + NodeCount float64 `json:"nodeCount"` //节点使用率 + // DeleteFlag int64 `json:"delete_flag"` // 是否删除 0:未删除，1:已经删除 + Description string `json:"description"` //集群描述 + ClusterName string `json:"clusterName"` //集群名称 + CpuTotal float64 `json:"cpuTotal"` //cpu总核数 + MemoryTotal float64 `json:"memoryTotal"` //内存总量Gi + DiskTotal float64 `json:"diskTotal"` //存储总量GB + NodeTotal float64 `json:"nodeTotal"` //容器节点数 + CpuUsage float64 `json:"cpuUsage"` //cpu已使用核数 + MemoryUsage float64 `json:"memoryUsage"` //内存已使用Gi + DiskUsage float64 `json:"diskUsage"` //存储已使用GB + NodeUsage float64 `json:"nodeUsage"` //容器节点已使用 } ) @@ -400,17 +408,17 @@ type ClusterMetrics struct { } type MetricResult struct { - MetricName string `json:"metric_name, omitempty"` + MetricName string `json:"metric_name, omitempty"` MetricData MetricData `json:"data, omitempty"` } type MetricData struct { - Result []map[string]interface{} `json:"result"` - ResultType string `json:"resultType"` + Result []map[string]interface{} `json:"result"` + ResultType string `json:"resultType"` } type ResultData struct { - Status string `json:"status"` - Data MetricData `json:"data"` + Status string `json:"status"` + Data MetricData `json:"data"` } -//jccSchedule容器集群资源监控 >end \ No newline at end of file + //jccSchedule容器集群资源监控 >end \ No newline at end of file diff --git a/adaptor/PCM-CORE/api/internal/logic/core/jobtotallogic.go b/adaptor/PCM-CORE/api/internal/logic/core/jobtotallogic.go index 1ef337b4..12055e15 100644 --- a/adaptor/PCM-CORE/api/internal/logic/core/jobtotallogic.go +++ b/adaptor/PCM-CORE/api/internal/logic/core/jobtotallogic.go @@ -1,9 +1,12 @@ package core import ( + "PCM/adaptor/PCM-CORE/model" + "PCM/common/enum" "PCM/common/tool" "context" "k8s.io/apimachinery/pkg/util/json" + "strings" "PCM/adaptor/PCM-CORE/api/internal/svc" "PCM/adaptor/PCM-CORE/api/internal/types" @@ -17,6 +20,21 @@ type JobTotalLogic struct { svcCtx *svc.ServiceContext } +type Job struct { + TotalSize int `json:"totalSize"` + OtJobs []OtJob `json:"otJobs"` +} + +type OtJob struct { + Name string `json:"name"` + Status string `json:"status"` + Tasks []Task `json:"tasks"` +} + +type Task struct { + CenterName []string `json:"centerName"` +} + func NewJobTotalLogic(ctx context.Context, svcCtx *svc.ServiceContext) *JobTotalLogic { return &JobTotalLogic{ Logger: logx.WithContext(ctx), @@ -26,12 +44,64 @@ func NewJobTotalLogic(ctx context.Context, svcCtx *svc.ServiceContext) *JobTotal } func (l *JobTotalLogic) JobTotal() (resp *types.JobTotalResp, err error) { - // todo: add your logic here and delete this line + // 获取任务时间信息 resp = &types.JobTotalResp{} bytes, err := tool.HttpGet("GET", "https://grampus.openi.org.cn/openapi/v1/sharescreen/computepower/alljobinfo") if err != nil { return nil, err } json.Unmarshal(bytes, resp) + + // 获取其他任务信息 + jobs := &Job{} + jobBytes, err := tool.HttpGet("GET", "https://grampus.openi.org.cn/openapi/v1/sharescreen/trainjob?pageIndex=1&pageSize=10") + if err != nil { + return nil, err + } + json.Unmarshal(jobBytes, jobs) + + for _, job := range jobs.OtJobs { + resp.TrainJobs = append(resp.TrainJobs, types.TrainJob{ + Name: job.Name, + Status: job.Status, + ServiceName: job.Tasks[0].CenterName[0], + Strategy: 0, + SynergyStatus: "未协同", + }) + } + + var tasks []model.Task + tx := l.svcCtx.DbEngin.Find(&tasks) + if tx.Error != nil { + logx.Error(err) + return nil, tx.Error + } + if len(tasks) == 0 { + return nil, nil + } + for _, task := range tasks { + tx := l.svcCtx.DbEngin.Raw("SELECT CONCAT_WS(',',GROUP_CONCAT(DISTINCT h.service_name) ,GROUP_CONCAT(DISTINCT a.service_name) ,GROUP_CONCAT(DISTINCT c.service_name))as service_name from task t left join hpc h on t.id = h.task_id left join cloud c on t.id = c.task_id left join ai a on t.id = a.task_id where t.id = ?", task.Id).Scan(&task.ServiceName) + if tx.Error != nil { + logx.Error(err) + return nil, tx.Error + } + // 承接方转义 + if task.ServiceName != "" { + var names []string + servicesName := strings.Split(task.ServiceName, ",") + for _, name := range servicesName { + names = append(names, enum.Partner(name).String()) + } + task.ServiceName = strings.Join(names, ",") + } + resp.TrainJobs = append(resp.TrainJobs, types.TrainJob{ + ServiceName: task.ServiceName, + Name: task.Name, + Strategy: int(task.Strategy), + SynergyStatus: enum.SynergyStatus(task.SynergyStatus).String(), + Status: task.Status, + }) + + } return resp, nil } diff --git a/adaptor/PCM-CORE/api/internal/svc/servicecontext.go b/adaptor/PCM-CORE/api/internal/svc/servicecontext.go index 0e1c8417..434a93fa 100644 --- a/adaptor/PCM-CORE/api/internal/svc/servicecontext.go +++ b/adaptor/PCM-CORE/api/internal/svc/servicecontext.go @@ -60,8 +60,7 @@ func NewServiceContext(c config.Config) *ServiceContext { ModelArtsRpc: modelartsclient.NewModelArts(zrpc.MustNewClient(c.ModelArtsRpcConf)), CephRpc: cephclient.NewCeph(zrpc.MustNewClient(c.CephRpcConf)), ACRpc: hpcacclient.NewHpcAC(zrpc.MustNewClient(c.ACRpcConf)), - //THRpc: hpcthclient.NewHpcTH(zrpc.MustNewClient(c.THRpcConf)), - OctopusRpc: octopusclient.NewOctopus(zrpc.MustNewClient(c.OctopusRpcConf)), - DockerClient: dockerClient, + OctopusRpc: octopusclient.NewOctopus(zrpc.MustNewClient(c.OctopusRpcConf)), + DockerClient: dockerClient, } } diff --git a/adaptor/PCM-CORE/api/internal/types/types.go b/adaptor/PCM-CORE/api/internal/types/types.go index ea4a9ad8..4315e473 100644 --- a/adaptor/PCM-CORE/api/internal/types/types.go +++ b/adaptor/PCM-CORE/api/internal/types/types.go @@ -181,9 +181,18 @@ type TaskInfo struct { } type JobTotalResp struct { - AllCardRunTime float64 `json:"allCardRunTime"` - AllJobCount float64 `json:"allJobCount"` - AllJobRunTime float64 `json:"allJobRunTime"` + AllCardRunTime float64 `json:"allCardRunTime"` + AllJobCount float64 `json:"allJobCount"` + AllJobRunTime float64 `json:"allJobRunTime"` + TrainJobs []TrainJob `json:"trainJobs"` +} + +type TrainJob struct { + Name string `json:"name"` + Status string `json:"status"` + ServiceName string `json:"ServiceName"` + SynergyStatus string `json:"synergyStatus"` + Strategy int `json:"strategy"` } type TaskListResp struct {