feat(dataset): update training status api

packages/global/openapi/admin/core/index.ts

@@ -2,11 +2,13 @@ import { DashboardPath } from './dashboard';
 import { AdminAppPath } from './app';
 import { AdminRoutesPath } from '../routes';
 import { AdminCommonPath } from '../common';
+import { StatusPath } from './status';
 import type { OpenAPIPath } from '../../type';
 
 export const AdminCorePath: OpenAPIPath = {
   ...DashboardPath,
   ...AdminAppPath,
   ...AdminRoutesPath,
-  ...AdminCommonPath
+  ...AdminCommonPath,
+  ...StatusPath
 };

packages/global/openapi/admin/core/status/api.ts

@@ -0,0 +1,94 @@
+import z from 'zod';
+import { ObjectIdSchema } from '../../../../common/type/mongo';
+import { PaginationResponseSchema, PaginationSchema } from '../../../api';
+
+export const TrainingRecordStateEnumSchema = z.enum([
+  'queued',
+  'processing',
+  'retrying',
+  'final_error',
+  'blocked',
+  'exhausted_without_error'
+]);
+export type TrainingRecordState = z.infer<typeof TrainingRecordStateEnumSchema>;
+
+export const TrainingRecordStateFilterEnumSchema = z.enum(['queued', 'processing', 'error']);
+export type TrainingRecordStateFilter = z.infer<typeof TrainingRecordStateFilterEnumSchema>;
+
+export const TrainingRecordActionEnumSchema = z.enum(['retry', 'delete']);
+export type TrainingRecordAction = z.infer<typeof TrainingRecordActionEnumSchema>;
+
+export const TrainingTaskSummarySchema = z.object({
+  mode: z.literal('chunk').meta({ description: '固定为 Chunk 向量训练' }),
+  queued: z.number().meta({ description: '排队任务数' }),
+  processing: z.number().meta({ description: '处理中任务数' }),
+  retrying: z.number().meta({ description: '自动重试任务数' }),
+  finalError: z.number().meta({ description: '最终失败任务数' }),
+  blocked: z.number().meta({ description: '永久阻塞任务数' }),
+  exhaustedWithoutError: z.number().meta({ description: '无错误但重试耗尽任务数' }),
+  total: z.number().meta({ description: '当前训练任务总数' }),
+  completed: z.number().meta({ description: '已完成 Chunk 数量' }),
+  severity: z.enum(['normal', 'warning', 'severe']).meta({ description: '运维告警等级' })
+});
+export type TrainingTaskSummaryType = z.infer<typeof TrainingTaskSummarySchema>;
+
+export const GetTrainingTaskListResponseSchema = TrainingTaskSummarySchema;
+export type GetTrainingTaskListResponseType = z.infer<typeof GetTrainingTaskListResponseSchema>;
+
+export const GetTrainingRecordListBodySchema = PaginationSchema.extend({
+  search: z.string().trim().min(1).optional().meta({ description: '训练 ID、知识库或团队名称' }),
+  state: TrainingRecordStateFilterEnumSchema.optional().meta({ description: '训练状态筛选' })
+});
+export type GetTrainingRecordListBody = z.infer<typeof GetTrainingRecordListBodySchema>;
+
+export const TrainingRecordSchema = z.object({
+  id: ObjectIdSchema.meta({ description: '训练任务 ID' }),
+  datasetId: ObjectIdSchema.meta({ description: '知识库 ID' }),
+  collectionId: ObjectIdSchema.meta({ description: '集合 ID' }),
+  datasetName: z.string().meta({ description: '知识库名称' }),
+  collectionName: z.string().optional().meta({ description: '集合名称' }),
+  teamName: z.string().optional().meta({ description: '团队名称' }),
+  model: z.string().optional().meta({ description: 'Embedding 模型' }),
+  chunkIndex: z.number().optional().meta({ description: '分块序号' }),
+  dataId: ObjectIdSchema.optional().meta({ description: '重建关联的数据 ID' }),
+  q: z.string().optional().meta({ description: '分块主文本' }),
+  a: z.string().optional().meta({ description: '分块补充文本' }),
+  retryCount: z.number().meta({ description: '剩余重试次数' }),
+  state: TrainingRecordStateEnumSchema.meta({ description: '训练状态' }),
+  actions: z
+    .array(TrainingRecordActionEnumSchema)
+    .meta({ description: '当前状态允许的管理员操作' }),
+  error: z.string().optional().meta({ description: '最近错误信息' }),
+  lockTime: z.string().optional().meta({ description: '任务租约时间' })
+});
+export type TrainingRecordType = z.infer<typeof TrainingRecordSchema>;
+
+export const GetTrainingRecordListResponseSchema = PaginationResponseSchema(TrainingRecordSchema);
+export type GetTrainingRecordListResponseType = z.infer<typeof GetTrainingRecordListResponseSchema>;
+
+export const TrainingActionBodySchema = z.object({
+  id: ObjectIdSchema.meta({ description: '训练任务 ID' }),
+  action: z.enum(['retry', 'delete']).meta({ description: '管理员操作' })
+});
+export type TrainingActionBody = z.infer<typeof TrainingActionBodySchema>;
+
+export const TrainingActionResponseSchema = z.object({
+  affectedCount: z.number().meta({ description: '实际影响的任务数量' })
+});
+export type TrainingActionResponseType = z.infer<typeof TrainingActionResponseSchema>;
+
+export const StatusOverviewAlertSchema = z.object({
+  level: z.literal('warning').meta({ description: '告警等级' }),
+  title: z.string().meta({ description: '告警标题' }),
+  description: z.string().meta({ description: '告警描述' }),
+  source: z.string().meta({ description: '告警来源' }),
+  scope: z.string().meta({ description: '告警范围' }),
+  count: z.number().meta({ description: '异常数量' })
+});
+export type StatusOverviewAlertType = z.infer<typeof StatusOverviewAlertSchema>;
+
+export const GetStatusOverviewResponseSchema = z.object({
+  unresolvedCount: z.number().meta({ description: '未恢复告警数' }),
+  alerts: z.array(StatusOverviewAlertSchema).meta({ description: '告警消息列表' })
+});
+export type GetStatusOverviewResponseType = z.infer<typeof GetStatusOverviewResponseSchema>;

packages/global/openapi/admin/core/status/index.ts

@@ -0,0 +1,67 @@
+import type { OpenAPIPath } from '../../../type';
+import {
+  GetStatusOverviewResponseSchema,
+  GetTrainingRecordListBodySchema,
+  GetTrainingRecordListResponseSchema,
+  GetTrainingTaskListResponseSchema,
+  TrainingActionBodySchema,
+  TrainingActionResponseSchema
+} from './api';
+import { DevApiTagsMap } from '../../../tag';
+
+export const StatusPath: OpenAPIPath = {
+  '/admin/status/overview': {
+    get: {
+      summary: '获取 Chunk 训练总览告警',
+      tags: [DevApiTagsMap.adminStatus],
+      responses: {
+        200: {
+          description: '成功获取训练总览告警',
+          content: { 'application/json': { schema: GetStatusOverviewResponseSchema } }
+        }
+      }
+    }
+  },
+  '/admin/status/dataset/tasks': {
+    get: {
+      summary: '获取 Chunk 训练状态汇总',
+      tags: [DevApiTagsMap.adminStatus],
+      responses: {
+        200: {
+          description: '成功获取训练状态汇总',
+          content: { 'application/json': { schema: GetTrainingTaskListResponseSchema } }
+        }
+      }
+    }
+  },
+  '/admin/status/dataset/training': {
+    post: {
+      summary: '获取 Chunk 训练任务列表',
+      tags: [DevApiTagsMap.adminStatus],
+      requestBody: {
+        content: { 'application/json': { schema: GetTrainingRecordListBodySchema } }
+      },
+      responses: {
+        200: {
+          description: '成功获取训练任务列表',
+          content: { 'application/json': { schema: GetTrainingRecordListResponseSchema } }
+        }
+      }
+    }
+  },
+  '/admin/status/dataset/training/action': {
+    post: {
+      summary: '重试或删除 Chunk 训练任务',
+      tags: [DevApiTagsMap.adminStatus],
+      requestBody: {
+        content: { 'application/json': { schema: TrainingActionBodySchema } }
+      },
+      responses: {
+        200: {
+          description: '成功执行训练任务操作',
+          content: { 'application/json': { schema: TrainingActionResponseSchema } }
+        }
+      }
+    }
+  }
+};

packages/global/openapi/path.ts

@@ -115,7 +115,12 @@ export const adminOpenAPIPaths: NonNullable<OpenAPIPath> = {
 export const adminOpenAPITagGroups = [
   {
     name: '管理员-系统概览',
-    tags: [DevApiTagsMap.adminDashboard, DevApiTagsMap.adminLogs, DevApiTagsMap.adminLicense]
+    tags: [
+      DevApiTagsMap.adminDashboard,
+      DevApiTagsMap.adminStatus,
+      DevApiTagsMap.adminLogs,
+      DevApiTagsMap.adminLicense
+    ]
   },
   {
     name: '管理员-资源管理',

packages/global/openapi/tag.ts

@@ -70,6 +70,7 @@ export const DevApiTagsMap = {
 
   /* 管理员-系统管理 */
   adminDashboard: '仪表盘',
+  adminStatus: '运维数据',
   adminInform: '通知管理',
   adminApps: '应用管理',
   adminWalletCoupon: '兑换码管理',

packages/service/core/dataset/training/query.ts

@@ -3,6 +3,7 @@ import {
   TrainingModeEnum
 } from '@fastgpt/global/core/dataset/constants';
 import type { DatasetTrainingSchemaType } from '@fastgpt/global/core/dataset/type';
+import { subMinutes } from 'date-fns';
 
 type TrainingStatusCount = {
   activeCount: number;
@@ -11,6 +12,18 @@ type TrainingStatusCount = {
 
 export const BLOCKED_LOCK_TIME = new Date('2050-01-01');
 
+export const trainingModeLockTimeoutMinutes: Record<TrainingModeEnum, number> = {
+  [TrainingModeEnum.parse]: 10,
+  [TrainingModeEnum.imageParse]: 10,
+  [TrainingModeEnum.qa]: 10,
+  [TrainingModeEnum.image]: 10,
+  [TrainingModeEnum.auto]: 10,
+  [TrainingModeEnum.chunk]: 3
+};
+
+export const getTrainingLockExpireTime = (mode: TrainingModeEnum, now = new Date()) =>
+  subMinutes(now, trainingModeLockTimeoutMinutes[mode]);
+
 export const trainingModeRankMap: Record<TrainingModeEnum, number> = {
   [TrainingModeEnum.parse]: 0,
   [TrainingModeEnum.imageParse]: 1,