增加推理任务重启接口

2026-02-28 10:59:48 +08:00 · 2026-02-28 10:59:48 +08:00 · 97b21dfd8b
parent eb71615a91
commit 97b21dfd8b
6 changed files with 103 additions and 7 deletions
--- a/pkg/handler/handler.go
+++ b/pkg/handler/handler.go
@ -262,6 +262,49 @@ func (h *Handler) StopInferenceTask(c *gin.Context) {
 	c.JSON(http.StatusOK, response)
 }

+// ReSubmitTask 处理重启推理任务请求
+// @Summary 重启推理任务
+// @Description 将入参原样转发给第三方 /jsm/v2/jobs/submit，返回新的 jobSetID 与入参中的 localJobID
+// @Tags 任务控制
+// @Accept json
+// @Produce json
+// @Param request body models.SubtaskRequest true "重启任务参数（userID、jobSetInfo.jobs）"
+// @Success 200 {object} models.ReSubmitTaskResponse
+// @Failure 400 {object} Response
+// @Failure 500 {object} Response
+// @Router /api/v1/reSubmitTask [post]
+func (h *Handler) ReSubmitTask(c *gin.Context) {
+	var request models.SubtaskRequest
+	if err := c.ShouldBindJSON(&request); err != nil {
+		c.JSON(http.StatusOK, BadRequestResponse("请求参数错误: "+err.Error()))
+		return
+	}
+	fmt.Printf("[ReSubmitTask] 请求参数: %+v\n", request)
+
+	if len(request.JobSetInfo.Jobs) == 0 {
+		c.JSON(http.StatusOK, BadRequestResponse("jobSetInfo.jobs 不能为空"))
+		return
+	}
+
+	// 获取认证信息
+	authData, err := h.authService.GetToken(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusOK, InternalServerErrorResponse("获取认证信息失败: "+err.Error()))
+		return
+	}
+
+	response, err := h.inferenceService.ReSubmitTask(c.Request.Context(), authData, &request)
+	if err != nil {
+		errResponse := InternalServerErrorResponse("重启任务失败: " + err.Error())
+		fmt.Printf("[ReSubmitTask] 返回错误: %+v\n", errResponse)
+		c.JSON(http.StatusOK, errResponse)
+		return
+	}
+
+	fmt.Printf("[ReSubmitTask] 返回结果: %+v\n", response)
+	c.JSON(http.StatusOK, response)
+}
+
 // SubmitInferenceTask 处理推理任务提交请求
 // @Summary 提交推理任务（同步）
 // @Description 同步提交推理任务，包含代码、模型、增量模型的准备步骤，等待完成后返回结果
@ -378,7 +421,7 @@ func (h *Handler) SubmitInferenceTaskAsync(c *gin.Context) {
 	}

 	response := models.InferenceAsyncSubmitResponse{
-		Code: 200,
+		Code: http.StatusOK,
 		Msg:  "",
 		Data: models.InferenceAsyncSubmitRespData{TaskID: taskID},
 	}
--- a/pkg/handler/response.go
+++ b/pkg/handler/response.go
@ -1,5 +1,7 @@
 package handler

+import "net/http"
+
 // Response 统一的API响应格式
 type Response struct {
 	Code    int         `json:"code"`    // 状态码
@ -10,7 +12,7 @@ type Response struct {
 // SuccessResponse 成功响应
 func SuccessResponse(data interface{}) Response {
 	return Response{
-		Code:    200,
+		Code:    http.StatusOK,
 		Message: "success",
 		Data:    data,
 	}
--- a/pkg/models/task.go
+++ b/pkg/models/task.go
@ -322,6 +322,19 @@ type StopInferenceTaskResponse struct {
 	Message string `json:"message"` // 消息
 }

+// ReSubmitTaskResponse 重启推理任务响应
+type ReSubmitTaskResponse struct {
+	Code    int              `json:"code"`    // 状态码
+	Message string           `json:"message"` // 消息
+	Data    ReSubmitTaskData `json:"data"`    // 响应数据
+}
+
+// ReSubmitTaskData 重启推理任务响应数据
+type ReSubmitTaskData struct {
+	JobSetID   string `json:"jobSetID"`   // 任务集ID（来自第三方接口返回）
+	LocalJobID string `json:"localJobID"` // 本地任务ID（来自入参 jobs）
+}
+
 // StopInferenceSubmitRequest 停止推理任务提交给第三方的请求
 type StopInferenceSubmitRequest struct {
 	UserID     int        `json:"userID"`     // 用户ID
--- a/pkg/router/router.go
+++ b/pkg/router/router.go
@ -126,6 +126,7 @@ func SetupRouter() *gin.Engine {
 		apiV1.POST("/submitTask", h.SubmitInferenceTask)
 		apiV1.POST("/getTaskStatus", h.QueryStatus)
 		apiV1.POST("/stopInferenceTask", h.StopInferenceTask)
+		apiV1.POST("/reSubmitTask", h.ReSubmitTask)
 		// 异步推理任务提交/查询
 		apiV1.POST("/submitInferenceTaskAsync", h.SubmitInferenceTaskAsync)
 		apiV1.POST("/getInferenceTaskAsync", h.GetInferenceTaskAsync)
--- a/pkg/service/inference/inference_service.go
+++ b/pkg/service/inference/inference_service.go
@ -5,6 +5,7 @@ import (
 	"database/sql"
 	"encoding/json"
 	"fmt"
+	"net/http"
 	"os"
 	"path/filepath"
 	"remote-task-excutor-cli/pkg/client"
@ -77,7 +78,7 @@ func (s *inferenceService) GetTaskStatus(ctx context.Context, authData *models.A
 			// 超时了，返回"启动中"状态，因为对端接口还在准备中
 			fmt.Printf("查询任务状态超时，返回启动中状态: %v\n", err)
 			return &models.InferenceTaskStatusResponse{
-				Code: 200,
+				Code: http.StatusOK,
 				Msg:  "",
 				Data: models.InferenceTaskStatusData{
 					Status: "Init",
@ -107,7 +108,7 @@ func (s *inferenceService) GetTaskStatus(ctx context.Context, authData *models.A

 	// 构建返回响应
 	return &models.InferenceTaskStatusResponse{
-		Code: 200,
+		Code: http.StatusOK,
 		Msg:  "",
 		Data: models.InferenceTaskStatusData{
 			Status: detailResp.Data.Instance.Status,
@ -169,11 +170,46 @@ func (s *inferenceService) StopTask(ctx context.Context, authData *models.AuthDa

 	// 返回成功响应
 	return &models.StopInferenceTaskResponse{
-		Code:    200,
+		Code:    http.StatusOK,
 		Message: "",
 	}, nil
 }

+// ReSubmitTask 重启推理任务：将入参原样转发给第三方 /jsm/v2/jobs/submit，返回 jobSetID 与入参中的 localJobID
+func (s *inferenceService) ReSubmitTask(ctx context.Context, authData *models.AuthData, request *models.SubtaskRequest) (*models.ReSubmitTaskResponse, error) {
+	if request == nil || len(request.JobSetInfo.Jobs) == 0 {
+		return nil, fmt.Errorf("jobSetInfo.jobs 不能为空")
+	}
+
+	s.httpClient.SetHeader("Authorization", "Bearer "+authData.Token)
+
+	resp, err := s.httpClient.PostJSON("/jsm/v2/jobs/submit", request)
+	if err != nil {
+		return nil, fmt.Errorf("调用重启任务接口失败: %w", err)
+	}
+
+	var submitResp models.SubmitTaskResponse
+	if err := json.Unmarshal(resp, &submitResp); err != nil {
+		return nil, fmt.Errorf("解析重启任务响应失败: %w", err)
+	}
+
+	if submitResp.Code != models.ResponseOK {
+		return nil, fmt.Errorf("重启任务失败: %s - %s", submitResp.Code, submitResp.Message)
+	}
+
+	localJobID := request.JobSetInfo.Jobs[0].LocalJobID
+	fmt.Printf("重启推理任务成功，JobSetID: %s, LocalJobID: %s\n", submitResp.Data.JobSetID, localJobID)
+
+	return &models.ReSubmitTaskResponse{
+		Code:    http.StatusOK,
+		Message: "",
+		Data: models.ReSubmitTaskData{
+			JobSetID:   submitResp.Data.JobSetID,
+			LocalJobID: localJobID,
+		},
+	}, nil
+}
+
 func (s *inferenceService) getLocalPath(path string) string {
 	if s.apiCfg == nil {
 		// 如果配置未设置，使用默认值
@ -257,7 +293,7 @@ WHERE task_id = ?
 	}

 	return &models.InferenceAsyncTaskStatusResponse{
-		Code: 200,
+		Code: http.StatusOK,
 		Msg:  "",
 		Data: models.InferenceAsyncTaskStatusData{
 			TaskID:   taskID,
@ -437,7 +473,7 @@ func (s *inferenceService) SubmitInferenceTask(ctx context.Context, authData *mo
 	}

 	return &models.InferenceSubmitTaskResponse{
-		Code: 200,
+		Code: http.StatusOK,
 		Msg:  "",
 		Data: taskData,
 	}, nil
--- a/pkg/service/interface.go
+++ b/pkg/service/interface.go
@ -37,6 +37,7 @@ type InferenceService interface {
 	GetInferenceTaskAsync(ctx context.Context, taskID string) (*models.InferenceAsyncTaskStatusResponse, error)
 	GetTaskStatus(ctx context.Context, authData *models.AuthData, localJobID, jobSetID string) (*models.InferenceTaskStatusResponse, error)
 	StopTask(ctx context.Context, authData *models.AuthData, localJobID, jobSetID string) (*models.StopInferenceTaskResponse, error)
+	ReSubmitTask(ctx context.Context, authData *models.AuthData, request *models.SubtaskRequest) (*models.ReSubmitTaskResponse, error)
 }

 type LogService interface {