admin-portal/src/views/dataManager/components/versionList.vue

@@ -208,4 +208,10 @@
       }
     }
   }
-</script>
+</script>
+<style lang="scss" scoped>
+  .block {
+    float: right;
+    margin: 20px;
+  }
+</style>

admin-portal/src/views/devManager/components/algorithm/versionList.vue

@@ -284,4 +284,10 @@
     }
 
   }
-</script>
+</script>
+<style lang="scss" scoped>
+  .block {
+    float: right;
+    margin: 20px;
+  }
+</style>

admin-portal/src/views/modelManager/components/versionList.vue

@@ -274,4 +274,8 @@
     }
 </script>
 <style lang="scss" scoped>
+  .block {
+    float: right;
+    margin: 20px;
+  }
 </style>

admin-portal/src/views/platformManager/platformTrainingTaskList.vue

@@ -11,6 +11,11 @@
           <span>{{ scope.row.name }}</span>
         </template>
       </el-table-column>
+      <el-table-column label="平台名称" align="center">
+        <template slot-scope="scope">
+          <span>{{ scope.row.platformName }}</span>
+        </template>
+      </el-table-column>
       <el-table-column label="镜像" align="center">
         <template slot-scope="scope">
           <span>{{ scope.row.image.name+":"+scope.row.image.version }}</span>

admin-portal/src/views/traningManager/components/taskInfo.vue

@@ -9,7 +9,7 @@
             </el-col>
         </el-row>
         <el-row>
-            <el-col v-if="data.isDistributed" :span="12">
+            <el-col v-if="showInfo" :span="12">
                 <el-form ref="ruleForm" :model="ruleForm">
                     <el-form-item prop="subTaskItem">
                         <div style="font-size: 15px">子任务名:
@@ -45,7 +45,6 @@
 
         <div class="block">
             <el-pagination
-                v-if="showInfo"
                 :current-page="pageIndex"
                 :page-sizes="[10, 20, 50, 80]"
                 :page-size="pageSize"
@@ -94,10 +93,7 @@
                     })
                 }
             }
-            if (!this.data.isDistributed) {
-                this.isDistributed = !this.data.isDistributed
-                this.selectedSubTaskOption()
-            }
+            this.selectedSubTaskOption()
         },
         methods: {
             selectedSubTaskOption() {
@@ -110,7 +106,7 @@
                 }
                 getTempalteInfo(param).then(response => {
                     if (response.success) {
-                        this.showInfo = response.payload.jobEvents.length
+                        this.showInfo = !this.data.isDistributed ? this.data.isDistributed : response.payload.jobEvents.length
                         this.total = response.payload.totalSize
                         let infoMessage = ""
                         response.payload.jobEvents.forEach(function(element) {
@@ -119,12 +115,31 @@
                             infoMessage += "\n" + "[" + title + "]"
                             infoMessage += "\n" + "[" + message + "]" + "\n"
                         })
+                        this.ruleForm.subTaskItem = this.row.config[0].replicaStates[0].key
                         this.subTaskInfo = infoMessage
                     } else {
-                        this.$message({
-                            message: "暂无相关运行信息",
-                            type: 'warning'
-                        });
+                        const data = {
+                            id: this.row.id,
+                            pageIndex: this.pageIndex,
+                            pageSize: this.pageSize,
+                            taskIndex: 0,
+                            replicaIndex: 0
+                        }
+                        getTempalteInfo(data).then(response => {
+                            if (response.success) {
+                                this.total = response.payload.totalSize
+                                let infoMessage = ""
+                                response.payload.jobEvents.forEach(function(element) {
+                                    const title = element.reason
+                                    const message = element.message
+                                    infoMessage += "\n" + "[" + title + "]"
+                                    infoMessage += "\n" + "[" + message + "]" + "\n"
+                                })
+                                this.subTaskInfo = infoMessage
+                            } else {
+                              this.subTaskInfo = "暂无相关运行信息"
+                            }
+                        })
                     }
                 }).catch(err => {
                     console.log("err:", err)

openai-portal/src/views/dataManager/components/versionList.vue

@@ -300,4 +300,8 @@
 </script>
 
 <style lang="scss" scoped>
+  .block {
+    float: right;
+    margin: 20px;
+  }
 </style>

openai-portal/src/views/modelDev/components/algorithm/versionList.vue

@@ -393,4 +393,10 @@
       }
     }
   }
-</script>
+</script>
+<style lang="scss" scoped>
+  .block {
+    float: right;
+    margin: 20px;
+  }
+</style>

openai-portal/src/views/trainingManager/components/detailDialog/taskInfo.vue

@@ -9,7 +9,7 @@
             </el-col>
         </el-row>
         <el-row>
-            <el-col v-if="data.isDistributed" :span="12">
+            <el-col v-if="showInfo" :span="12">
                 <el-form ref="ruleForm" :model="ruleForm">
                     <el-form-item prop="subTaskItem">
                         <div style="font-size: 15px">子任务名:
@@ -45,7 +45,6 @@
 
         <div class="block">
             <el-pagination
-                v-if="showInfo"
                 :current-page="pageIndex"
                 :page-sizes="[10, 20, 50, 80]"
                 :page-size="pageSize"
@@ -94,10 +93,7 @@
                     })
                 }
             }
-            if (!this.data.isDistributed) {
-                this.isDistributed = !this.data.isDistributed
-                this.selectedSubTaskOption()
-            }
+            this.selectedSubTaskOption()
         },
         methods: {
             selectedSubTaskOption() {
@@ -110,7 +106,7 @@
                 }
                 getTempalteInfo(param).then(response => {
                     if (response.success) {
-                        this.showInfo = response.payload.jobEvents.length
+                        this.showInfo = !this.data.isDistributed ? this.data.isDistributed : response.payload.jobEvents.length
                         this.total = response.payload.totalSize
                         let infoMessage = ""
                         response.payload.jobEvents.forEach(function(element) {
@@ -119,12 +115,31 @@
                             infoMessage += "\n" + "[" + title + "]"
                             infoMessage += "\n" + "[" + message + "]" + "\n"
                         })
+                        this.ruleForm.subTaskItem = this.row.config[0].replicaStates[0].key
                         this.subTaskInfo = infoMessage
                     } else {
-                        this.$message({
-                            message: "暂无相关运行信息",
-                            type: 'warning'
-                        });
+                        const data = {
+                            id: this.row.id,
+                            pageIndex: this.pageIndex,
+                            pageSize: this.pageSize,
+                            taskIndex: 0,
+                            replicaIndex: 0
+                        }
+                        getTempalteInfo(data).then(response => {
+                            if (response.success) {
+                                this.total = response.payload.totalSize
+                                let infoMessage = ""
+                                response.payload.jobEvents.forEach(function(element) {
+                                    const title = element.reason
+                                    const message = element.message
+                                    infoMessage += "\n" + "[" + title + "]"
+                                    infoMessage += "\n" + "[" + message + "]" + "\n"
+                                })
+                                this.subTaskInfo = infoMessage
+                            } else {
+                              this.subTaskInfo = "暂无相关运行信息"
+                            }
+                        })
                     }
                 }).catch(err => {
                     console.log("err:", err)

server/admin-server/api/v1/develop.proto

@@ -88,10 +88,10 @@ message NotebookEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message NotebookEventListReply {

server/admin-server/api/v1/platform.proto

@@ -304,25 +304,27 @@ message PlatformTrainJob{
   string name = 2;
   //platformId
   string platformId = 3;
+  //platformName
+  string platformName = 4;
   //job描述
-  string desc = 4;
+  string desc = 5;
   //数据集
-  repeated PlatformDataset datasets = 5;
+  repeated PlatformDataset datasets = 6;
   //镜像
-  PlatformImage image = 6;
+  PlatformImage image = 7;
   //子任务配置信息
-  repeated PlatformTask tasks = 7;
+  repeated PlatformTask tasks = 8;
   //创建时间
-  int64 createdAt = 8;
+  int64 createdAt = 9;
   //更新时间
-  int64 updatedAt = 9;
+  int64 updatedAt = 10;
   //任务状态
-  string status   = 10;
+  string status   = 11;
   //job完成时间
-  int64  completedAt = 11;
+  int64  completedAt = 12;
   //运行时
-  int64 runSec = 12;
+  int64 runSec = 13;
   //启动时间
-  int64  startedAt = 13;
+  int64  startedAt = 14;
 }
 

server/admin-server/api/v1/trainJob.proto

@@ -189,10 +189,10 @@ message JobEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message JobEventListReply {

server/base-server/api/v1/develop.proto

@@ -124,10 +124,10 @@ message NotebookEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message NotebookEventListReply {

server/base-server/api/v1/platformtrainJob.proto

@@ -147,26 +147,28 @@ message PlatformTrainJob{
   string name = 2;
   //platformId
   string platformId = 3;
+  //platformName
+  string platformName = 4;
   //job描述
-  string desc = 4;
+  string desc = 5;
   //数据集
-  repeated PlatformDataset datasets = 5;
+  repeated PlatformDataset datasets = 6;
   //镜像
-  PlatformImage image = 6;
+  PlatformImage image = 7;
   //子任务配置信息
-  repeated PlatformTask tasks = 7;
+  repeated PlatformTask tasks = 8;
   //创建时间
-  int64 createdAt = 8;
+  int64 createdAt = 9;
   //更新时间
-  int64 updatedAt = 9;
+  int64 updatedAt = 10;
   //任务状态
-  string status   = 10;
+  string status   = 11;
   //job完成时间
-  int64  completedAt = 11;
+  int64  completedAt = 12;
   //运行时
-  int64 runSec = 12;
+  int64 runSec = 13;
   //启动时间
-  int64  startedAt = 13;
+  int64  startedAt = 14;
 }
 
 

server/base-server/api/v1/trainJob.proto

@@ -245,10 +245,10 @@ message JobEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message JobEventListReply {

server/base-server/internal/data/dao/platform/platform_train_job.go

@@ -128,7 +128,7 @@ func (d *platformTrainJobDao) GetTrainJobList(ctx context.Context, query *model.
 	}
 
 	if query.OrderBy != "" {
-		sortBy = query.OrderBy
+		orderBy = query.OrderBy
 	}
 
 	db = db.Order(fmt.Sprintf("%s %s", sortBy, orderBy))

server/base-server/internal/data/dao/train_job.go

@@ -366,7 +366,12 @@ func (d *trainJobDao) GetTrainJobEvents(jobEventQuery *model.JobEventQuery) ([]*
 	ReplicaIndex := jobEventQuery.ReplicaIndex
 	events := make([]*model.TrainJobEvent, 0)
 
-	objectName := fmt.Sprintf("%s-task%d-%d", jobEventQuery.Id, TaskIndex-1, ReplicaIndex-1)
+	objectName := ""
+	if TaskIndex > 0 && ReplicaIndex > 0 {
+		objectName = fmt.Sprintf("%s-task%d-%d", jobEventQuery.Id, TaskIndex-1, ReplicaIndex-1)
+	} else {
+		objectName = jobEventQuery.Id
+	}
 
 	countQuery := fmt.Sprintf("SELECT COUNT(%s) FROM octopus..events where object_name = '%s'", keyMessage, objectName)
 	res, err := d.influxdb.Query(countQuery)
@@ -384,8 +389,9 @@ func (d *trainJobDao) GetTrainJobEvents(jobEventQuery *model.JobEventQuery) ([]*
 		return events, 0, errors.Errorf(err, errors.ErroInfluxdbFindFailed)
 	}
 
-	query := fmt.Sprintf("select %s, %s, %s from octopus..events where object_name = '%s' and kind = 'Pod' LIMIT %d OFFSET %d",
+	query := fmt.Sprintf("select %s, %s, %s from octopus..events where object_name = '%s' LIMIT %d OFFSET %d",
 		keyName, keyReason, keyMessage, objectName, PageSize, (PageIndex-1)*PageSize)
+
 	res, err = d.influxdb.Query(query)
 
 	if err != nil {

server/base-server/internal/service/lable/lable.go

@@ -17,12 +17,12 @@ import (
 
 var LABLE_DEFAULT_DESC = map[api.Relegation]map[api.Type][]string{
 	api.Relegation_LABLE_RELEGATION_DATASET: {
-		api.Type_LABLE_TYPE_DATASET_TYPE:  {"图像", "视频", "音频", "文本"},
-		api.Type_LABLE_TYPE_DATASET_APPLY: {"图像分类", "目标检测", "目标跟踪", "语义分割", "文本分类", "中文分词", "音频分类", "数据增强"},
+		api.Type_LABLE_TYPE_DATASET_TYPE:  {"图像", "视频", "音频", "文本", "其他"},
+		api.Type_LABLE_TYPE_DATASET_APPLY: {"图像分类", "目标检测", "目标跟踪", "语义分割", "文本分类", "中文分词", "音频分类", "数据增强", "其他"},
 	},
 	api.Relegation_LABLE_RELEGATION_ALGORITHM: {
-		api.Type_LABLE_TYPE_ALGORITHM_APPLY:     {"图像分类", "目标检测", "目标跟踪", "语义分割", "文本分类", "中文分词", "音频分类", "模型优化"},
-		api.Type_LABLE_TYPE_ALGORITHM_FRAMEWORK: {"TensorFlow", "Pytorch", "MindSpore", "Keras"},
+		api.Type_LABLE_TYPE_ALGORITHM_APPLY:     {"图像分类", "目标检测", "目标跟踪", "语义分割", "文本分类", "中文分词", "音频分类", "模型优化", "其他"},
+		api.Type_LABLE_TYPE_ALGORITHM_FRAMEWORK: {"TensorFlow", "Pytorch", "MindSpore", "Keras", "其他"},
 	},
 }
 

server/base-server/internal/service/platform/train_job.go

@@ -588,7 +588,11 @@ func (s *platformTrainJobService) TrainJobList(ctx context.Context, req *api.Pla
 		if err != nil {
 			return nil, err
 		}
-
+		platform, err := s.getPlatformInfo(ctx, job.PlatformId)
+		if err != nil {
+			return nil, err
+		}
+		trainJob.PlatformName = platform.Name
 		trainJobs = append(trainJobs, trainJob)
 	}
 
@@ -1032,15 +1036,10 @@ func (s *platformTrainJobService) updatePlatfromJobStatus(ctx context.Context, p
 		return err
 	}
 	if url, ok := reply.Config[common.PlatformKeyJobStatusCallbackAddr]; ok {
-		platformReply, err := s.platformService.BatchGetPlatform(ctx, &api.BatchGetPlatformRequest{Ids: []string{platformId}})
+		platform, err := s.getPlatformInfo(ctx, platformId)
 		if err != nil {
 			return err
 		}
-		if len(platformReply.Platforms) <= 0 {
-			s.log.Info(ctx, "updatePlatfromJobStatus failed, cannot find platform ClientSecret:"+info.JobId)
-			return errors.Errorf(err, errors.ErrorDBFindEmpty)
-		}
-		platform := platformReply.Platforms[0]
 		err = s.data.Platform.UpdateJobStatus(ctx, url, platform.ClientSecret, info)
 		if err != nil {
 			return err
@@ -1050,3 +1049,17 @@ func (s *platformTrainJobService) updatePlatfromJobStatus(ctx context.Context, p
 	}
 	return nil
 }
+
+func (s *platformTrainJobService) getPlatformInfo(ctx context.Context, platformId string) (*api.Platform, error) {
+
+	platformReply, err := s.platformService.BatchGetPlatform(ctx, &api.BatchGetPlatformRequest{Ids: []string{platformId}})
+	if err != nil {
+		return nil, err
+	}
+	if len(platformReply.Platforms) <= 0 {
+		s.log.Info(ctx, "updatePlatfromJobStatus failed, cannot find platform ClientSecret")
+		return nil, errors.Errorf(err, errors.ErrorDBFindEmpty)
+	}
+	platform := platformReply.Platforms[0]
+	return platform, nil
+}

server/common/utils/protoc.go

@@ -8,6 +8,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"regexp"
 	"strings"
 
 	jsonpatch "github.com/evanphx/json-patch"
@@ -162,6 +163,8 @@ func GenSwagger() error {
 				"./",
 				"--openapiv2_opt",
 				"logtostderr=true",
+				"--openapiv2_opt",
+				"enums_as_ints=true",
 				name,
 			}
 
@@ -210,7 +213,12 @@ func GenSwagger() error {
 		}
 
 		swaggerStr := strings.ReplaceAll(string(swaggerBytes), `,"default":{"description":"An unexpected error response.","schema":{"$ref":"#/definitions/rpcStatus"}}`, "")
-
+		reg := regexp.MustCompile(`{[^{]*"format":"(int64|uint64)"[\s\S]*?"type":"string"[^}]*}`)
+		swaggerStr = reg.ReplaceAllStringFunc(swaggerStr, func(s string) string { // proto json序列化64位序列化为字符串，octopus用的是标准库json序列化，这个修改为整型
+			s = strings.ReplaceAll(s, `"type":"string"`, `"type":"number"`)
+			r := regexp.MustCompile(`"format":"(int64|uint64)",`)
+			return r.ReplaceAllString(s, "")
+		})
 		err = ioutil.WriteFile(filepath.Join(dir, swaggerFileName), []byte(swaggerStr), 0755)
 		if err != nil {
 			return err

server/openai-server/api/v1/develop.proto

@@ -149,10 +149,10 @@ message NotebookEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message NotebookEventListReply {

server/openai-server/api/v1/trainJob.proto

@@ -385,10 +385,10 @@ message JobEventListRequest {
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   //任务ID
   string id = 3[(validate.rules).string = {min_len: 1}];
-  //子任务序号，从1开始
-  int64 taskIndex = 4[(validate.rules).int64 = {gte:1,lt:100}];
-  //副本序号，从1开始
-  int64 replicaIndex = 5[(validate.rules).int64 = {gte:1,lt:100}];
+  //子任务序号，从0开始
+  int64 taskIndex = 4[(validate.rules).int64 = {gte:0,lt:100}];
+  //副本序号，从0开始
+  int64 replicaIndex = 5[(validate.rules).int64 = {gte:0,lt:100}];
 }
 
 message JobEventListReply {

server/platform-server/api/v1/trainJob.proto

@@ -140,9 +140,9 @@ message StopJobReply {
 }
 
 message TrainJobListRequest{
-  // 页码，从1开始，非必填
+  // 页码，从1开始，必填
   int64 pageIndex = 1[(validate.rules).int64 = {gte:1}];
-  // 页大小，最小1条，最大100条，非必填
+  // 页大小，最小1条，最大100条，必填
   int64 pageSize = 2[(validate.rules).int64 = {gte:1,lt:100}];
   // 分组依据，非必填
   string sortBy = 3;