1

chore: 更新配置并优化任务错误处理
将数据库、Redis、Consul及Jaeger配置切换至新环境地址，并在worker中忽略"Invalid video_url"与"Invalid audio track"错误，同时补充模型调用成功日志。
2026-06-22 10:57:46 +08:00 · 2026-06-22 09:08:12 +08:00 · 2026-06-18 16:27:11 +08:00 · 2026-06-18 14:48:42 +08:00 · 2026-06-18 13:39:40 +08:00 · 2026-06-18 10:08:36 +08:00
12 changed files with 346 additions and 382 deletions
--- a/common/util/files.go
+++ b/common/util/files.go
@@ -1,7 +1,6 @@
 package util

 import (
-	"encoding/json"
 	"fmt"
 	"net/http"
 	"os"
@@ -9,107 +8,75 @@ import (
 	"strings"
 )

-// DetectFileType 根据返回的二进制内容推断 contentType + 扩展名（尽量稳定）
+// DetectFileType 根据返回的二进制内容推断 contentType + 扩展名
 func DetectFileType(data []byte) (contentType string, ext string) {
 	if len(data) == 0 {
-		return "application/octet-stream", ""
+		return "application/octet-stream", ".bin"
 	}
+
 	ct := http.DetectContentType(data)
-	// gateway.DetectContentType 可能带 charset 等参数：text/plain; charset=utf-8
 	if idx := strings.Index(ct, ";"); idx > 0 {
 		ct = strings.TrimSpace(ct[:idx])
 	}
+
 	switch ct {
 	case "audio/mpeg":
 		return ct, ".mp3"
 	case "audio/wave", "audio/wav", "audio/x-wav":
 		return ct, ".wav"
+	case "audio/mp4", "audio/x-m4a":
+		return ct, ".m4a"
 	case "video/mp4":
 		return ct, ".mp4"
+	case "video/webm":
+		return ct, ".webm"
 	case "image/png":
 		return ct, ".png"
 	case "image/jpeg":
 		return ct, ".jpg"
+	case "image/gif":
+		return ct, ".gif"
+	case "image/webp":
+		return ct, ".webp"
 	case "application/pdf":
 		return ct, ".pdf"
 	case "text/plain":
 		return ct, ".txt"
 	case "application/json":
 		return ct, ".json"
+	case "application/zip":
+		return ct, ".zip"
+	case "application/octet-stream":
+		return ct, ".bin"
 	default:
-		// 兜底：尝试从 ct 截取 subtype 作为后缀（例如 application/json）
 		if parts := strings.Split(ct, "/"); len(parts) == 2 {
 			sub := parts[1]
-			// 避免出现 "plain; charset=utf-8" 之类的后缀
 			if idx := strings.Index(sub, ";"); idx > 0 {
 				sub = strings.TrimSpace(sub[:idx])
 			}
 			return ct, "." + sub
 		}
-		return ct, ""
+		return ct, ".bin"
 	}
 }

-// SaveTmpResult 将模型输出写入临时文件，用于 OSS 上传失败后的“仅重试 OSS”。
+// SaveTmpResult 将二进制数据写入临时文件
 func SaveTmpResult(taskID string, data []byte, ext string) (string, error) {
 	dir := filepath.Join(os.TempDir(), "model-asynch")
 	if err := os.MkdirAll(dir, 0o755); err != nil {
-		return "", err
+		return "", fmt.Errorf("创建临时目录失败: %w", err)
 	}
+
 	if ext == "" {
 		ext = ".bin"
 	}
 	if ext[0] != '.' {
 		ext = "." + ext
 	}
+
 	path := filepath.Join(dir, fmt.Sprintf("%s%s", taskID, ext))
 	if err := os.WriteFile(path, data, 0o644); err != nil {
-		return "", err
+		return "", fmt.Errorf("写入临时文件失败: %w", err)
 	}
 	return path, nil
 }
-
-// SaveTempFileByType
-// 根据传入的数据自动判断：
-// 若是 []byte 且后缀为 .mp3 → 保存二进制音频
-// 若是任意结构体/map → 自动转 JSON 保存
-// 返回：新临时文件路径、错误
-func SaveTempFileByType(taskID string, data any, oldTmpFile string) (string, error) {
-	// 1. 先清理旧临时文件（统一逻辑）
-	if oldTmpFile != "" {
-		_ = os.Remove(oldTmpFile)
-	}
-
-	var tmpPath string
-	var tmpErr error
-
-	// 2. 判断是否是二进制音频（[]byte + .mp3）
-	if audioData, ok := data.([]byte); ok {
-		tmpPath, tmpErr = saveTmpResult(taskID, audioData, ".mp3")
-	} else {
-		// 3. 其他类型 → 序列化为 JSON 保存
-		mappedBytes, err := json.Marshal(data)
-		if err != nil {
-			return "", err
-		}
-		if len(mappedBytes) == 0 {
-			return "", nil
-		}
-		tmpPath, tmpErr = saveTmpResult(taskID, mappedBytes, ".json")
-	}
-
-	if tmpErr != nil || tmpPath == "" {
-		return "", tmpErr
-	}
-
-	return tmpPath, nil
-}
-
-// saveTmpResult 你原有的底层保存文件方法（保留不动）
-func saveTmpResult(taskID string, data []byte, ext string) (string, error) {
-	// 你原来实现，比如：
-	filename := taskID + ext
-	tmpPath := filepath.Join(os.TempDir(), filename)
-	err := os.WriteFile(tmpPath, data, 0644)
-	return tmpPath, err
-}
--- a/common/util/mapping.go
+++ b/common/util/mapping.go
@@ -19,17 +19,20 @@ import (
 	tgjson "github.com/tidwall/gjson"
 )

-// ParseAndValidate 解析并校验结果
+// ParseAndValidate 解析模型响应，并返回标准格式
 func ParseAndValidate(raw map[string]any, model *entity.ModelGatewayModel) (map[string]any, error) {
-	// 1) 解析 content 字符串为 rounds 数组
-	contentVal, ok := raw[entity.ResponseBody]
-	if !ok {
-		return raw, fmt.Errorf("字段 %s 不存在", entity.ResponseBody)
-	}
-	contentStr, ok := contentVal.(string)
-	if !ok || strings.TrimSpace(contentStr) == "" {
-		return raw, fmt.Errorf("字段 %s 为空或不是字符串", entity.ResponseBody)
+	contentStr := gconv.String(raw[entity.ResponseBody])
+	if strings.TrimSpace(contentStr) == "" {
+		return raw, fmt.Errorf("字段 %s 为空", entity.ResponseBody)
 	}
+
+	contentStr = strings.Map(func(r rune) rune {
+		if r < 32 && r != ' ' {
+			return -1
+		}
+		return r
+	}, contentStr)
+
 	var arr []any
 	if err := json.Unmarshal([]byte(contentStr), &arr); err != nil {
 		return raw, fmt.Errorf("JSON解析失败: %w", err)
@@ -38,17 +41,11 @@ func ParseAndValidate(raw map[string]any, model *entity.ModelGatewayModel) (map[
 		return raw, fmt.Errorf("解析后数组为空")
 	}

-	// 2) 校验必填字段
-	if len(model.RequiredFields) > 0 {
+	for _, field := range model.RequiredFields {
 		for i, r := range arr {
-			round, ok := r.(map[string]any)
-			if !ok {
-				continue
-			}
-			for _, field := range model.RequiredFields {
-				if gjson.New(round).Get(field).IsNil() {
-					return raw, fmt.Errorf("rounds[%d] 缺少必填字段: %s", i, field)
-				}
+			round, _ := r.(map[string]any)
+			if round != nil && gjson.New(round).Get(field).IsNil() {
+				return raw, fmt.Errorf("rounds[%d] 缺少必填字段: %s", i, field)
 			}
 		}
 	}
--- a/config.yml
+++ b/config.yml
@@ -39,8 +39,8 @@ database:
      dryRun: false
      charset: "utf8"
      timezone: "Asia/Shanghai"
-      maxIdle: 5
-      maxOpen: 20
+      maxIdle: 15
+      maxOpen: 60
      maxLifetime: "30s"
      maxIdleConnTime: "30s"
      createdAt: "created_at"
--- a/dao/model_gateway_models_dao.go
+++ b/dao/model_gateway_models_dao.go
@@ -56,6 +56,7 @@ func (d *modelGatewayModelsDao) Get(ctx context.Context, req *entity.ModelGatewa
 		Where(entity.ModelGatewayModelCol.Id, req.Id).
 		Where(entity.ModelGatewayModelCol.Creator, req.Creator).
 		Where(entity.ModelGatewayModelCol.ModelName, req.ModelName).
+		Where(entity.ModelGatewayModelCol.IsChatModel, req.IsChatModel).
 		Fields(fields).One()
 	if err != nil {
 		return nil, err
@@ -122,7 +123,7 @@ func (d *modelGatewayModelsDao) GetByAcrossTenant(ctx context.Context, req *enti
 func (d *modelGatewayModelsDao) GetByCreatorAndPlatform(ctx context.Context, req *dto.ListModelReq) (list []*entity.ModelGatewayModel, total int, err error) {
 	sql := `
 SELECT DISTINCT ON (model_name) *
-FROM asynch_models
+FROM ` + public.TableNameModel + `
 WHERE deleted_at IS NULL 
  AND (? = '' OR model_name LIKE ?)
 `
--- a/dao/model_gateway_task_dao.go
+++ b/dao/model_gateway_task_dao.go
@@ -7,7 +7,6 @@ import (
 	"model-gateway/model/entity"

 	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
-	"github.com/gogf/gf/v2/database/gdb"
 	"github.com/gogf/gf/v2/util/gconv"
 )

@@ -128,32 +127,32 @@ func (d *modelGatewayTaskDao) GetPendingAsyncTasks(ctx context.Context, limit in

 // ClaimByID 按主键抢占，返回抢占后的任务
 func (d *modelGatewayTaskDao) ClaimByID(ctx context.Context, id int64) (*entity.ModelGatewayTask, error) {
+	// 1) 先查任务
 	var task entity.ModelGatewayTask
-	err := gfdb.DB(ctx, public.DbNameModelGateway).Transaction(ctx, func(ctx context.Context, tx gdb.TX) error {
-		r, err := tx.Model(public.TableNameTask).
-			Where(entity.ModelGatewayTaskCol.Id, id).
-			Where(entity.ModelGatewayTaskCol.State, public.TaskStatusPending).
-			Limit(1).
-			LockUpdate().
-			One()
-		if err != nil {
-			return err
-		}
-		if r.IsEmpty() {
-			return fmt.Errorf("任务已被抢占或不存在: id=%d", id)
-		}
-		if err := r.Struct(&task); err != nil {
-			return err
-		}
-		_, err = tx.Model(public.TableNameTask).
-			Data(&entity.ModelGatewayTask{State: public.TaskStatusRunning}).
-			Where(entity.ModelGatewayTaskCol.Id, id).
-			OmitEmpty().
-			Update()
-		return err
-	})
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameTask).
+		Where(entity.ModelGatewayTaskCol.Id, id).
+		Where(entity.ModelGatewayTaskCol.State, public.TaskStatusPending).
+		One()
 	if err != nil {
 		return nil, err
 	}
+	if r.IsEmpty() {
+		return nil, fmt.Errorf("任务已被抢占或不存在: id=%d", id)
+	}
+	if err = r.Struct(&task); err != nil {
+		return nil, err
+	}
+
+	// 2) 改为执行中
+	_, err = gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameTask).
+		Data(&entity.ModelGatewayTask{State: public.TaskStatusRunning}).
+		Where(entity.ModelGatewayTaskCol.Id, id).
+		Where(entity.ModelGatewayTaskCol.State, public.TaskStatusPending). // 防并发
+		OmitEmpty().
+		Update()
+	if err != nil {
+		return nil, err
+	}
+
 	return &task, nil
 }
--- a/model/dto/model_gateway_models_dto.go
+++ b/model/dto/model_gateway_models_dto.go
@@ -9,34 +9,36 @@ import (

 // CreateModelReq 添加模型配置
 type CreateModelReq struct {
-	g.Meta           `path:"/createModel" method:"post" tags:"模型管理" summary:"创建模型配置" dc:"添加新的模型配置"`
-	ModelName        string           `p:"modelName" json:"modelName" v:"required#模型名称不能为空" dc:"模型名称（唯一标识）"`
-	ModelType        int              `p:"modelType" json:"modelType" v:"required#模型类型不能为空" dc:"模型类型"`
-	BaseURL          string           `p:"baseUrl" json:"baseUrl" v:"required#模型地址不能为空" dc:"模型服务地址"`
-	HttpMethod       string           `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST（默认POST）"`
-	HeadMsg          map[string]any   `p:"headMsg" json:"headMsg" dc:"请求头JSON结构"`
-	IsPrivate        *int             `p:"isPrivate" json:"isPrivate" dc:"是否私有化：0-私有 1-公共"`
-	Enabled          *int             `p:"enabled" json:"enabled" dc:"是否启用：0-停用 1-启用"`
-	IsChatModel      *int             `p:"isChatModel" json:"isChatModel" dc:"是否为对话模型：0-否 1-是"`
-	CallModel        *int             `p:"callModel" json:"callModel" dc:"调用模式：0-同步 1-异步 2-流式"`
-	RequiredFields   []string         `p:"requiredFields" json:"requiredFields" dc:"必填字段"`
-	IsOwner          *int             `p:"isOwner" json:"isOwner" dc:"是否为所有者：0-否 1-是"`
-	ApiKey           string           `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥"`
-	Form             []map[string]any `p:"form" json:"form" dc:"动态表单配置"`
-	RequestMapping   map[string]any   `p:"requestMapping" json:"requestMapping" dc:"请求映射"`
-	ResponseMapping  map[string]any   `p:"responseMapping" json:"responseMapping" dc:"返回映射"`
-	OperatorName     string           `p:"operatorName" json:"operatorName" dc:"运营商名称"`
-	TokenConfig      map[string]any   `p:"tokenConfig" json:"tokenConfig" dc:"token计算配置"`
-	ExtendMapping    map[string]any   `p:"extendMapping" json:"extendMapping" dc:"附加映射"`
-	QueryConfig      map[string]any   `p:"queryConfig" json:"queryConfig" dc:"查询/回调配置"`
-	StreamConfig     map[string]any   `p:"streamConfig" json:"streamConfig" dc:"流式输出配置"`
-	FirstFrame       string           `p:"firstFrame" json:"firstFrame" dc:"首帧图片参数"`
-	LastFrame        string           `p:"lastFrame" json:"lastFrame" dc:"尾帧图片参数"`
-	MaxConcurrency   int              `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数（默认10）"`
-	TimeoutSeconds   int              `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒，默认600）"`
-	RetryTimes       int              `p:"retryTimes" json:"retryTimes" dc:"失败重试次数（默认3）"`
-	AutoCleanSeconds int              `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"任务完成后自动清理时间（秒，默认86400）"`
-	CallbackUrl      string           `p:"callbackUrl" json:"callbackUrl" dc:"回调地址"`
+	g.Meta             `path:"/createModel" method:"post" tags:"模型管理" summary:"创建模型配置" dc:"添加新的模型配置"`
+	ModelName          string           `p:"modelName" json:"modelName" v:"required#模型名称不能为空" dc:"模型名称（唯一标识）"`
+	ModelType          int              `p:"modelType" json:"modelType" v:"required#模型类型不能为空" dc:"模型类型"`
+	BaseURL            string           `p:"baseUrl" json:"baseUrl" v:"required#模型地址不能为空" dc:"模型服务地址"`
+	HttpMethod         string           `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST（默认POST）"`
+	HeadMsg            map[string]any   `p:"headMsg" json:"headMsg" dc:"请求头JSON结构"`
+	IsPrivate          *int             `p:"isPrivate" json:"isPrivate" dc:"是否私有化：0-私有 1-公共"`
+	Enabled            *int             `p:"enabled" json:"enabled" dc:"是否启用：0-停用 1-启用"`
+	IsChatModel        *int             `p:"isChatModel" json:"isChatModel" dc:"是否为对话模型：0-否 1-是"`
+	CallModel          *int             `p:"callModel" json:"callModel" dc:"调用模式：0-同步 1-异步 2-流式"`
+	RequiredFields     []string         `p:"requiredFields" json:"requiredFields" dc:"必填字段"`
+	IsOwner            *int             `p:"isOwner" json:"isOwner" dc:"是否为所有者：0-否 1-是"`
+	ApiKey             string           `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥"`
+	Form               []map[string]any `p:"form" json:"form" dc:"动态表单配置"`
+	RequestMapping     map[string]any   `p:"requestMapping" json:"requestMapping" dc:"请求映射"`
+	ResponseMapping    map[string]any   `p:"responseMapping" json:"responseMapping" dc:"返回映射"`
+	ResponseBody       string           `p:"responseBody" json:"responseBody" dc:"返回主体"`
+	ResponseTokenField string           `p:"responseTokenField" json:"responseTokenField" dc:"响应中消耗token的字段映射"`
+	OperatorName       string           `p:"operatorName" json:"operatorName" dc:"运营商名称"`
+	TokenConfig        map[string]any   `p:"tokenConfig" json:"tokenConfig" dc:"token计算配置"`
+	ExtendMapping      map[string]any   `p:"extendMapping" json:"extendMapping" dc:"附加映射"`
+	QueryConfig        map[string]any   `p:"queryConfig" json:"queryConfig" dc:"查询/回调配置"`
+	StreamConfig       map[string]any   `p:"streamConfig" json:"streamConfig" dc:"流式输出配置"`
+	FirstFrame         string           `p:"firstFrame" json:"firstFrame" dc:"首帧图片参数"`
+	LastFrame          string           `p:"lastFrame" json:"lastFrame" dc:"尾帧图片参数"`
+	MaxConcurrency     int              `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数（默认10）"`
+	TimeoutSeconds     int              `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒，默认600）"`
+	RetryTimes         int              `p:"retryTimes" json:"retryTimes" dc:"失败重试次数（默认3）"`
+	AutoCleanSeconds   int              `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"任务完成后自动清理时间（秒，默认86400）"`
+	CallbackUrl        string           `p:"callbackUrl" json:"callbackUrl" dc:"回调地址"`
 }

 type CreateModelRes struct {
@@ -44,35 +46,37 @@ type CreateModelRes struct {
 }

 type UpdateModelReq struct {
-	g.Meta           `path:"/updateModel" method:"put" tags:"模型管理" summary:"更新模型配置" dc:"更新指定ID的模型配置"`
-	ID               int64            `p:"id" json:"id" v:"required#id不能为空" dc:"配置ID"`
-	ModelName        string           `p:"modelName" json:"modelName" dc:"模型名称"`
-	ModelType        int              `p:"modelType" json:"modelType" dc:"模型类型"`
-	BaseURL          string           `p:"baseUrl" json:"baseUrl" dc:"模型服务地址"`
-	HttpMethod       string           `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST"`
-	HeadMsg          map[string]any   `p:"headMsg" json:"headMsg" dc:"请求头JSON结构"`
-	IsPrivate        *int             `p:"isPrivate" json:"isPrivate" dc:"是否私有化：0-私有 1-公共"`
-	Enabled          *int             `p:"enabled" json:"enabled" dc:"是否启用：0-停用 1-启用"`
-	IsChatModel      *int             `p:"isChatModel" json:"isChatModel" dc:"是否为对话模型：0-否 1-是"`
-	CallModel        *int             `p:"callModel" json:"callModel" dc:"调用模式：0-同步 1-异步 2-流式"`
-	RequiredFields   []string         `p:"requiredFields" json:"requiredFields" dc:"必填字段"`
-	IsOwner          *int             `p:"isOwner" json:"isOwner" dc:"是否为所有者：0-否 1-是"`
-	ApiKey           string           `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥"`
-	Form             []map[string]any `p:"form" json:"form" dc:"动态表单配置"`
-	RequestMapping   map[string]any   `p:"requestMapping" json:"requestMapping" dc:"请求映射"`
-	ResponseMapping  map[string]any   `p:"responseMapping" json:"responseMapping" dc:"返回映射"`
-	OperatorName     string           `p:"operatorName" json:"operatorName" dc:"运营商名称"`
-	TokenConfig      map[string]any   `p:"tokenConfig" json:"tokenConfig" dc:"token计算配置"`
-	ExtendMapping    map[string]any   `p:"extendMapping" json:"extendMapping" dc:"附加映射"`
-	QueryConfig      map[string]any   `p:"queryConfig" json:"queryConfig" dc:"查询/回调配置"`
-	StreamConfig     map[string]any   `p:"streamConfig" json:"streamConfig" dc:"流式输出配置"`
-	FirstFrame       string           `p:"firstFrame" json:"firstFrame" dc:"首帧图片参数"`
-	LastFrame        string           `p:"lastFrame" json:"lastFrame" dc:"尾帧图片参数"`
-	MaxConcurrency   int              `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数"`
-	TimeoutSeconds   int              `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒）"`
-	RetryTimes       int              `p:"retryTimes" json:"retryTimes" dc:"失败重试次数"`
-	AutoCleanSeconds int              `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"任务完成后自动清理时间（秒）"`
-	CallbackUrl      string           `p:"callbackUrl" json:"callbackUrl" dc:"回调地址"`
+	g.Meta             `path:"/updateModel" method:"put" tags:"模型管理" summary:"更新模型配置" dc:"更新指定ID的模型配置"`
+	ID                 int64            `p:"id" json:"id" v:"required#id不能为空" dc:"配置ID"`
+	ModelName          string           `p:"modelName" json:"modelName" dc:"模型名称"`
+	ModelType          int              `p:"modelType" json:"modelType" dc:"模型类型"`
+	BaseURL            string           `p:"baseUrl" json:"baseUrl" dc:"模型服务地址"`
+	HttpMethod         string           `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST"`
+	HeadMsg            map[string]any   `p:"headMsg" json:"headMsg" dc:"请求头JSON结构"`
+	IsPrivate          *int             `p:"isPrivate" json:"isPrivate" dc:"是否私有化：0-私有 1-公共"`
+	Enabled            *int             `p:"enabled" json:"enabled" dc:"是否启用：0-停用 1-启用"`
+	IsChatModel        *int             `p:"isChatModel" json:"isChatModel" dc:"是否为对话模型：0-否 1-是"`
+	CallModel          *int             `p:"callModel" json:"callModel" dc:"调用模式：0-同步 1-异步 2-流式"`
+	RequiredFields     []string         `p:"requiredFields" json:"requiredFields" dc:"必填字段"`
+	IsOwner            *int             `p:"isOwner" json:"isOwner" dc:"是否为所有者：0-否 1-是"`
+	ApiKey             string           `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥"`
+	Form               []map[string]any `p:"form" json:"form" dc:"动态表单配置"`
+	RequestMapping     map[string]any   `p:"requestMapping" json:"requestMapping" dc:"请求映射"`
+	ResponseMapping    map[string]any   `p:"responseMapping" json:"responseMapping" dc:"返回映射"`
+	ResponseBody       string           `p:"responseBody" json:"responseBody" dc:"返回主体"`
+	ResponseTokenField string           `p:"responseTokenField" json:"responseTokenField" dc:"响应中消耗token的字段映射"`
+	OperatorName       string           `p:"operatorName" json:"operatorName" dc:"运营商名称"`
+	TokenConfig        map[string]any   `p:"tokenConfig" json:"tokenConfig" dc:"token计算配置"`
+	ExtendMapping      map[string]any   `p:"extendMapping" json:"extendMapping" dc:"附加映射"`
+	QueryConfig        map[string]any   `p:"queryConfig" json:"queryConfig" dc:"查询/回调配置"`
+	StreamConfig       map[string]any   `p:"streamConfig" json:"streamConfig" dc:"流式输出配置"`
+	FirstFrame         string           `p:"firstFrame" json:"firstFrame" dc:"首帧图片参数"`
+	LastFrame          string           `p:"lastFrame" json:"lastFrame" dc:"尾帧图片参数"`
+	MaxConcurrency     int              `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数"`
+	TimeoutSeconds     int              `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒）"`
+	RetryTimes         int              `p:"retryTimes" json:"retryTimes" dc:"失败重试次数"`
+	AutoCleanSeconds   int              `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"任务完成后自动清理时间（秒）"`
+	CallbackUrl        string           `p:"callbackUrl" json:"callbackUrl" dc:"回调地址"`
 }

 type UpdateModelRes struct {
--- a/model/entity/model_gateway_model.go
+++ b/model/entity/model_gateway_model.go
@@ -12,7 +12,6 @@ type modelGatewayModelCol struct {
 	FormJSON         string
 	RequestMapping   string
 	ResponseMapping  string
-	ResponseBody     string
 	RequiredFields   string
 	IsPrivate        string
 	IsChatModel      string
@@ -31,6 +30,7 @@ type modelGatewayModelCol struct {
 	StreamConfig     string
 	FirstFrame       string
 	LastFrame        string
+	MaxTokens        string
 }

 var ModelGatewayModelCol = modelGatewayModelCol{
@@ -61,6 +61,7 @@ var ModelGatewayModelCol = modelGatewayModelCol{
 	StreamConfig:     "stream_config",
 	FirstFrame:       "first_frame",
 	LastFrame:        "last_frame",
+	MaxTokens:        "max_tokens",
 }

 type ModelGatewayModel struct {
@@ -91,9 +92,10 @@ type ModelGatewayModel struct {
 	StreamConfig     map[string]any   `orm:"stream_config" json:"streamConfig"`
 	FirstFrame       string           `orm:"first_frame" json:"firstFrame"`
 	LastFrame        string           `orm:"last_frame" json:"lastFrame"`
+	MaxTokens        int              `orm:"max_tokens" json:"maxTokens"`
 }

-const ( //ResponseMapping 下的字段
-	ResponseBody = "response_body" //返回主体
-	TotalTokens  = "total_tokens"  //总token数
+const (
+	ResponseBody = "content"      //返回主体（必填）
+	TotalTokens  = "total_tokens" //总token数
 )
--- a/service/gateway/gateway_http_service.go
+++ b/service/gateway/gateway_http_service.go
@@ -7,7 +7,6 @@ import (
 	"errors"
 	"fmt"
 	"mime/multipart"
-	"model-gateway/common/util"
 	"model-gateway/model/entity"
 	"time"

@@ -43,16 +42,25 @@ func UploadByTask(ctx context.Context, data []byte, fileExt string) (oss *Upload
 	if err != nil {
 		return nil, err
 	}
-	if _, err := part.Write(data); err != nil {
+	if _, err = part.Write(data); err != nil {
 		return nil, err
 	}
-	contentType := writer.FormDataContentType()
+	//contentType := writer.FormDataContentType()
 	if err = writer.Close(); err != nil {
 		return nil, err
 	}

-	headers := util.ForwardHeaders(ctx)
-	headers["Content-Type"] = contentType
+	//headers := util.ForwardHeaders(ctx)
+	//headers["Content-Type"] = contentType
+
+	headers := make(map[string]string)
+	headers["Content-Type"] = writer.FormDataContentType()
+	if r := g.RequestFromCtx(ctx); r != nil {
+		if auth := r.Header.Get("Authorization"); auth != "" {
+			headers["Authorization"] = auth
+		}
+	}
+
 	fullURL := "oss/file/uploadFile"
 	g.Log().Infof(ctx, "[OSS] upload start url=%s filename=%s size=%d", fullURL, filename, len(data))

@@ -78,15 +86,25 @@ type CallbackPayload struct {

 // TriggerCallback 任务的回调
 func TriggerCallback(ctx context.Context, t *entity.ModelGatewayTask) {
-	headers := util.ForwardHeaders(ctx)
+	//headers := util.ForwardHeaders(ctx)
+	headers := make(map[string]string)
+	if r := g.RequestFromCtx(ctx); r != nil {
+		for k, v := range r.Request.Header {
+			if len(v) > 0 {
+				headers[k] = v[0]
+			}
+		}
+	}
 	var resp struct{}
 	payload := CallbackPayload{
 		TaskId:   t.TaskID,
 		State:    t.State,
-		OssFile:  t.ResultFile.OssFile,
-		FileType: t.ResultFile.FileType,
 		ErrorMsg: t.ErrorMsg,
 	}
+	if !g.IsEmpty(t.ResultFile) {
+		payload.OssFile = t.ResultFile.OssFile
+		payload.FileType = t.ResultFile.FileType
+	}
 	jsonData, err := json.Marshal(payload)
 	if err != nil {
 		g.Log().Warningf(ctx, "[回调] JSON序列化失败 taskId=%s 错误=%v", t.TaskID, err)
@@ -112,7 +130,15 @@ type PromptsCallbackPayload struct {
 // TriggerPromptsCallback 任务成功后的提示词回调
 func TriggerPromptsCallback(ctx context.Context, t *entity.ModelGatewayTask, epicycleId int64) {
 	callbackURL := "prompts-core/session/callback"
-	headers := util.ForwardHeaders(ctx)
+	//headers := util.ForwardHeaders(ctx)
+	headers := make(map[string]string)
+	if r := g.RequestFromCtx(ctx); r != nil {
+		for k, v := range r.Request.Header {
+			if len(v) > 0 {
+				headers[k] = v[0]
+			}
+		}
+	}
 	var resp struct{}
 	payload := PromptsCallbackPayload{
 		EpicycleId: epicycleId,
@@ -136,7 +162,15 @@ func TriggerPromptsCallback(ctx context.Context, t *entity.ModelGatewayTask, epi

 // IsSuperAdmin 调用admin-go服务检查是否是超级管理员
 func IsSuperAdmin(ctx context.Context) (res bool, err error) {
-	headers := util.ForwardHeaders(ctx)
+	//headers := util.ForwardHeaders(ctx)
+	headers := make(map[string]string)
+	if r := g.RequestFromCtx(ctx); r != nil {
+		for k, v := range r.Request.Header {
+			if len(v) > 0 {
+				headers[k] = v[0]
+			}
+		}
+	}
 	var r = make(map[string]bool)
 	if err = commonHttp.Get(ctx, "admin-go/api/v1/system/user/checkIsSuperAdmin", headers, &r); err != nil {
 		return false, err
--- a/service/model/model_service.go
+++ b/service/model/model_service.go
@@ -99,7 +99,7 @@ func (s *modelService) Get(ctx context.Context, req *dto.GetModelReq) (*dto.GetM
 		ModelName:   req.ModelName,
 		IsChatModel: req.IsChatModel,
 	})
-	if err != nil {
+	if err != nil || model == nil {
 		return nil, err
 	}
 	return &dto.GetModelRes{
--- a/service/task/task_service.go
+++ b/service/task/task_service.go
@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"model-gateway/common/util"
 	"model-gateway/consts/public"
-	"model-gateway/service/queue"
 	"time"

 	"model-gateway/dao"
@@ -28,12 +27,15 @@ type taskService struct{}
 // Create 创建任务
 func (s *taskService) Create(ctx context.Context, req *dto.CreateTaskReq) (res *dto.CreateTaskRes, err error) {
 	taskID := uuid.NewString()
+	startAt := time.Now()

-	// 1) 检查模型配置，并且获取模型
+	// 1) 获取用户信息
 	userInfo, err := utils.GetUserInfo(ctx)
 	if err != nil {
 		return nil, err
 	}
+
+	// 2) 检查模型配置
 	model, err := dao.ModelGatewayModels.Get(ctx, &entity.ModelGatewayModel{
 		SQLBaseDO: beans.SQLBaseDO{
 			TenantId: userInfo.TenantId,
@@ -48,72 +50,63 @@ func (s *taskService) Create(ctx context.Context, req *dto.CreateTaskReq) (res *
 		return nil, errors.New("模型不存在或未启用")
 	}

-	// 2) 排队上限（严格控制：Redis 原子闸门）
-	limit := queue.GetRuntimeQueueLimit(ctx, req.ModelName, model.MaxConcurrency*2)
-	if limit > 0 {
-		ok, err := queue.AcquireQueueSlot(ctx, req.ModelName, taskID, limit, model.TimeoutSeconds)
-		if err != nil {
-			return nil, err
-		}
-		if !ok {
-			return nil, errors.New("任务排队已满，请稍后再试")
-		}
+	// TODO: 排队控制暂时关闭，后续需要时取消注释
+	// limit := queue.GetRuntimeQueueLimit(ctx, req.ModelName, model.MaxConcurrency*2)
+	// if limit > 0 {
+	// 	ok, err := queue.AcquireQueueSlot(ctx, req.ModelName, taskID, limit, model.TimeoutSeconds)
+	// 	if err != nil {
+	// 		return nil, err
+	// 	}
+	// 	if !ok {
+	// 		return nil, errors.New("任务排队已满，请稍后再试")
+	// 	}
+	// }
+
+	// 3) 构建任务实体
+	task := &entity.ModelGatewayTask{
+		ModelName:   model.ModelName,
+		TaskID:      taskID,
+		State:       public.TaskStatusRunning,
+		BizName:     req.BizName,
+		CallbackURL: req.CallbackUrl,
+		RequestPayload: &entity.RequestPayload{
+			Body:    req.RequestPayload,
+			Headers: util.ParseHeadMsgHeaders(model.HeadMsg),
+		},
+		EpicycleId: req.EpicycleId,
 	}

-	// 3) 插入任务记录
-	requestPayload := entity.RequestPayload{
-		Body:    req.RequestPayload,
-		Headers: util.ParseHeadMsgHeaders(model.HeadMsg),
-	}
-	id, err := dao.ModelGatewayTask.Insert(ctx, &entity.ModelGatewayTask{
-		ModelName:      req.ModelName,
-		TaskID:         taskID,
-		State:          public.TaskStatusPending,
-		BizName:        req.BizName,
-		CallbackURL:    req.CallbackUrl,
-		RequestPayload: &requestPayload,
-		EpicycleId:     req.EpicycleId,
-	})
-	if err != nil { // 入库失败：回滚闸门占位
-		queue.ReleaseQueueSlot(ctx, req.ModelName, taskID)
+	// 4) 插入任务记录
+	id, err := dao.ModelGatewayTask.Insert(ctx, task)
+	if err != nil {
+		// TODO: 恢复排队逻辑后，此处需要回滚排队占位
+		// queue.ReleaseQueueSlot(ctx, req.ModelName, taskID)
 		return nil, err
 	}
+	task.Id = id

-	// 4) 写操作日志（不影响主流程，失败忽略）
-	ip := ""
-	ua := ""
-	apiPath := "/task/createTask"
-	httpMethod := "POST"
+	// 5) 记录操作日志（非关键路径，失败不影响主流程）
+	ip, ua := "", ""
 	if r := g.RequestFromCtx(ctx); r != nil {
 		ip = utils.GetLocalIP()
 		ua = r.UserAgent()
-		apiPath = r.URL.Path
-		httpMethod = r.Method
 	}
 	_, _ = dao.ModelGatewayLogsOp.Insert(ctx, &entity.ModelGatewayLogsOp{
-		IP:             ip,
-		UserAgent:      ua,
-		APIPath:        apiPath,
-		HttpMethod:     httpMethod,
-		BizName:        req.BizName,
-		ModelName:      req.ModelName,
-		TaskID:         taskID,
-		OpType:         "createTask",
-		Success:        1,
-		CostMs:         time.Since(time.Now()).Milliseconds(),
-		RequestPayload: &requestPayload,
-		ResponsePayload: gdb.Map{
-			"taskId": taskID,
-		},
+		IP:              ip,
+		UserAgent:       ua,
+		APIPath:         "/task/createTask",
+		HttpMethod:      "POST",
+		BizName:         req.BizName,
+		ModelName:       req.ModelName,
+		TaskID:          taskID,
+		OpType:          "createTask",
+		Success:         1,
+		CostMs:          time.Since(startAt).Milliseconds(),
+		RequestPayload:  task.RequestPayload,
+		ResponsePayload: gdb.Map{"taskId": taskID},
 	})

-	// 5) 获取任务信息
-	task, err := dao.ModelGatewayTask.ClaimByID(ctx, id)
-	if err != nil {
-		return nil, err
-	}
-
-	// 5) 创建成功后立即异步尝试执行当前任务
+	// 6) 异步执行任务
 	go AsyncWorker.handleOne(util.AsyncCtx(ctx), task, model, req)

 	return &dto.CreateTaskRes{TaskID: taskID}, nil
--- a/service/task/worker.go
+++ b/service/task/worker.go
@@ -7,7 +7,6 @@ import (
 	"fmt"
 	"io"
 	"net/http"
-	"os"
 	"strings"
 	"sync"
 	"time"
@@ -19,7 +18,6 @@ import (
 	"model-gateway/model/dto"
 	"model-gateway/model/entity"
 	"model-gateway/service/gateway"
-	"model-gateway/service/queue"

 	"gitea.redpowerfuture.com/red-future/common/beans"
 	"github.com/gogf/gf/v2/encoding/gjson"
@@ -38,67 +36,54 @@ func (w *asyncWorker) handleOne(ctx context.Context, task *entity.ModelGatewayTa
 		body      = task.RequestPayload.Body
 		maxRetry  = model.RetryTimes
 		startTime = time.Now()
+		rawBytes  []byte
 		result    map[string]any
 		err       error
 	)
-	g.Log().Infof(ctx, "[执行任务][开始] taskId=%s model=%s", task.TaskID, task.ModelName)
+
+	g.Log().Infof(ctx, "[handleOne] 开始 taskId=%s model=%s", task.TaskID, task.ModelName)

 	// ============================================
-	// 1) 分布式并发控制
+	// 1) 调用模型
 	// ============================================
-	semKey := fmt.Sprintf("asynch:sem:%s", task.ModelName)
-	maxC := queue.GetRuntimeMaxConcurrency(ctx, task.ModelName, model.MaxConcurrency)
-	acquired, err := queue.AcquireSemaphore(ctx, semKey, maxC, 3600)
-	if err != nil {
-		task.DurationSeconds = int64(time.Since(startTime).Seconds())
-		w.failTask(ctx, task, startTime, err.Error())
-		return
-	}
-	if !acquired {
-		_, _ = dao.ModelGatewayTask.Update(ctx, &entity.ModelGatewayTask{
-			SQLBaseDO: beans.SQLBaseDO{Id: task.Id},
-			State:     public.TaskStatusPending,
-		})
-		g.Log().Infof(ctx, "[执行任务][排队] 并发已满，放回队列 taskId=%s", task.TaskID)
-		return
-	}
-	defer func() { _ = queue.ReleaseSemaphore(ctx, semKey) }()
+	for attempt := 0; ; attempt++ {
+		if attempt > 0 {
+			g.Log().Infof(ctx, "[handleOne] 调模型重试 第%d次 taskId=%s", attempt, task.TaskID)
+			time.Sleep(time.Duration(attempt) * time.Second)
+		}

-	// ============================================
-	// 2) 调用模型
-	// ============================================
-	switch {
-	case model.CallMode != nil && *model.CallMode == public.CallModeStream:
-		rawBytes, streamErr := w.callModelStream(ctx, task, model, body)
-		if streamErr != nil {
-			w.failTask(ctx, task, startTime, streamErr.Error())
+		switch {
+		case model.CallMode != nil && *model.CallMode == public.CallModeStream:
+			rawBytes, err = InvokeModel(ctx, model, body)
+			if err == nil {
+				result, err = util.ParseStreamResponse(rawBytes, model.StreamConfig)
+			}
+		case model.CallMode != nil && *model.CallMode == public.CallModeAsync:
+			result, err = w.callModel(ctx, task, model, body)
+			if err == nil {
+				result, err = util.PullTaskResult(ctx, result, model.QueryConfig, model.HeadMsg)
+			}
+		default:
+			result, err = w.callModel(ctx, task, model, body)
+		}
+
+		if err == nil {
+			break
+		}
+
+		if !strings.Contains(err.Error(), "Timeout") &&
+			!strings.Contains(err.Error(), "InternalServiceError") &&
+			!strings.Contains(err.Error(), "Invalid video_url") &&
+			!strings.Contains(err.Error(), "Invalid audio track") {
+			w.failTask(ctx, task, startTime, err.Error())
 			return
 		}
-		result, err = util.ParseStreamResponse(rawBytes, model.StreamConfig)
-	case model.CallMode != nil && *model.CallMode == public.CallModeAsync:
-		result, err = w.callModel(ctx, task, model, body)
-		if err == nil {
-			result, err = util.PullTaskResult(ctx, result, model.QueryConfig, model.HeadMsg)
-		}
-	default:
-		result, err = w.callModel(ctx, task, model, body)
-	}
-	if err != nil {
-		w.failTask(ctx, task, startTime, err.Error())
-		return
+
+		g.Log().Warningf(ctx, "[handleOne] 调模型失败 taskId=%s attempt=%d err=%v", task.TaskID, attempt, err)
 	}

 	// ============================================
-	// 3) 缓存临时文件
-	// ============================================
-	if tmpPath, tmpErr := util.SaveTempFileByType(task.TaskID, result, task.TmpFile); tmpErr == nil && tmpPath != "" {
-		task.TmpFile = tmpPath
-		task.Phase = 1
-		_, _ = dao.ModelGatewayTask.Update(ctx, task)
-	}
-
-	// ============================================
-	// 4) 解析校验 + 响应映射（可重试）
+	// 2) 解析校验 + 响应映射（可重试）
 	// ============================================
 	result, err = w.parseAndRetry(ctx, result, task, model, req, maxRetry, startTime)
 	if err != nil {
@@ -108,30 +93,26 @@ func (w *asyncWorker) handleOne(ctx context.Context, task *entity.ModelGatewayTa
 	}

 	// ============================================
-	// 5) 上传 OSS（可重试）
+	// 3) 上传 OSS（可重试）
 	// ============================================
 	var oss *gateway.UploadFileResponse
 	for attempt := 0; attempt <= maxRetry; attempt++ {
 		if attempt > 0 {
-			g.Log().Infof(ctx, "[执行任务][重试] OSS上传 第%d/%d次 taskId=%s", attempt, maxRetry, task.TaskID)
+			g.Log().Infof(ctx, "[handleOne] OSS上传重试 第%d/%d次 taskId=%s", attempt, maxRetry, task.TaskID)
 		}
 		oss, err = gateway.UploadByTask(ctx, gjson.New(result).MustToJson(), "json")
 		if err == nil {
 			break
 		}
-		g.Log().Errorf(ctx, "[执行任务][失败] OSS上传失败 taskId=%s attempt=%d/%d err=%v", task.TaskID, attempt, maxRetry, err)
+		g.Log().Errorf(ctx, "[handleOne] OSS上传失败 taskId=%s attempt=%d/%d err=%v", task.TaskID, attempt, maxRetry, err)
 		if attempt == maxRetry {
-			task.State = public.TaskStatusFailed
-			task.ErrorMsg = err.Error()
-			task.Phase = 1
-			_, _ = dao.ModelGatewayTask.Update(ctx, task)
 			w.failTask(ctx, task, startTime, fmt.Sprintf("OSS上传重试耗尽: %v", err))
 			return
 		}
 	}

 	// ============================================
-	// 6) 成功收尾
+	// 4) 成功收尾
 	// ============================================
 	task.State = public.TaskStatusSuccess
 	task.DurationSeconds = int64(time.Since(startTime).Seconds())
@@ -141,52 +122,19 @@ func (w *asyncWorker) handleOne(ctx context.Context, task *entity.ModelGatewayTa
 		FileSize: int64(oss.FileSize),
 	}
 	task.TextResult = result
+
 	if _, err = dao.ModelGatewayTask.Update(ctx, task); err != nil {
-		g.Log().Errorf(ctx, "[执行任务][失败] 更新数据库失败 taskId=%s err=%v", task.TaskID, err)
+		g.Log().Errorf(ctx, "[handleOne] 更新DB失败 taskId=%s err=%v", task.TaskID, err)
 		return
 	}

-	queue.ReleaseQueueSlot(ctx, task.ModelName, task.TaskID)
-	go gateway.TriggerCallback(context.WithoutCancel(ctx), task)
+	go gateway.TriggerCallback(util.AsyncCtx(ctx), task)
 	if req.EpicycleId != 0 {
-		go gateway.TriggerPromptsCallback(context.WithoutCancel(ctx), task, req.EpicycleId)
+		go gateway.TriggerPromptsCallback(util.AsyncCtx(ctx), task, req.EpicycleId)
 	}

-	g.Log().Infof(ctx, "[执行任务][成功] taskId=%s duration=%ds fileType=%s",
+	g.Log().Infof(ctx, "[handleOne] 成功 taskId=%s duration=%ds fileType=%s",
 		task.TaskID, task.DurationSeconds, oss.FileFormat)
-
-	_ = os.Remove(task.TmpFile)
-}
-
-// callModelStream 调用模型，返回原始字节（不做响应映射，用于流式输出）
-func (w *asyncWorker) callModelStream(ctx context.Context, task *entity.ModelGatewayTask, model *entity.ModelGatewayModel, body map[string]any) ([]byte, error) {
-	var data []byte
-	var err error
-
-	if task.Phase == 1 && strings.TrimSpace(task.TmpFile) != "" {
-		data, err = os.ReadFile(task.TmpFile)
-		if err != nil || len(data) == 0 {
-			data = nil
-		}
-	}
-
-	if data == nil {
-		data, err = InvokeModel(ctx, model, body)
-		if err != nil {
-			return nil, err
-		}
-		tmpPath, tmpErr := util.SaveTmpResult(task.TaskID, data, "")
-		if tmpErr == nil && tmpPath != "" {
-			task.TmpFile = tmpPath
-			task.Phase = 1
-			_, err = dao.ModelGatewayTask.Update(ctx, task)
-			if err != nil {
-				g.Log().Errorf(ctx, "[执行任务][失败] 临时文件保存失败 taskId=%s err=%v", task.TaskID, tmpErr)
-			}
-		}
-	}
-
-	return data, nil
 }

 // asyncResult 异步任务结果
@@ -241,67 +189,37 @@ func NotifyAsyncResult(taskID string, result map[string]any, err error) {
 	}
 }

-// callModel 调用模型 + 检测文件类型 + 保存临时文件
-// 返回: 解析后的响应体, error
+// callModel 调用模型 + 提取文本结果
 func (w *asyncWorker) callModel(ctx context.Context, task *entity.ModelGatewayTask, model *entity.ModelGatewayModel, body map[string]any) (map[string]any, error) {
-	var data []byte
-	var err error
-
-	// 1) 如果已有临时文件且 phase=1，直接读取
-	if task.Phase == 1 && strings.TrimSpace(task.TmpFile) != "" {
-		data, err = os.ReadFile(task.TmpFile)
-		if err != nil || len(data) == 0 {
-			g.Log().Warningf(ctx, "[callModel] 读取临时文件失败，重新调用模型 taskId=%s err=%v", task.TaskID, err)
-			data = nil
-		}
+	data, err := InvokeModel(ctx, model, body)
+	if err != nil {
+		return nil, err
 	}

-	// 2) 没有可用数据，调用模型
-	if data == nil {
-		data, err = InvokeModel(ctx, model, body)
-		if err != nil {
-			return nil, err
-		}
-
-		// 3) 检测文件类型，保存临时文件
-		_, ext := util.DetectFileType(data)
-		tmpPath, tmpErr := util.SaveTmpResult(task.TaskID, data, ext)
-		if tmpErr == nil && tmpPath != "" {
-			task.TmpFile = tmpPath
-			task.Phase = 1
-			_, err = dao.ModelGatewayTask.Update(ctx, task)
-			if err != nil {
-				g.Log().Errorf(ctx, "[执行任务][失败] 临时文件保存失败 taskId=%s err=%v", task.TaskID, tmpErr)
-			}
-		}
-	}
-
-	// 4) 检测文件类型，提取文本结果
 	contentType, _ := util.DetectFileType(data)
 	var textResult string
 	if utf8.Valid(data) && (strings.HasPrefix(contentType, "text/") || contentType == "application/json") {
 		textResult = string(data)
 	}

-	// 5) 非文本内容，返回错误
 	if textResult == "" {
 		return nil, fmt.Errorf("模型返回非文本内容，contentType=%s", contentType)
 	}

-	// 6) 解析并返回
 	return gjson.New(textResult).Map(), nil
 }

 // parseAndRetry 解析模型返回结果，并重试
 func (w *asyncWorker) parseAndRetry(ctx context.Context, body map[string]any, task *entity.ModelGatewayTask, model *entity.ModelGatewayModel, req *dto.CreateTaskReq, maxRetry int, startTime time.Time) (map[string]any, error) {
+	var lastErr error
 	for attempt := 0; attempt <= maxRetry; attempt++ {
 		if attempt > 0 {
 			g.Log().Infof(ctx, "[执行任务][重试] JSON解析 第%d/%d次 taskId=%s", attempt, maxRetry, task.TaskID)
 		}
-
 		// 1) 响应映射
 		mapped, err := util.MapResponsePayload(model.ResponseMapping, body)
 		if err != nil {
+			lastErr = err
 			g.Log().Warningf(ctx, "[执行任务][映射失败] taskId=%s attempt=%d/%d err=%v", task.TaskID, attempt, maxRetry, err)
 			if attempt == maxRetry {
 				return nil, fmt.Errorf("响应映射重试耗尽: %w", err)
@@ -309,10 +227,10 @@ func (w *asyncWorker) parseAndRetry(ctx context.Context, body map[string]any, ta
 			continue
 		}

-		// 2) 先存 token 到数据库，防止后续失败丢失
+		// 2) 存 token
 		if _, ok := mapped[entity.TotalTokens]; ok {
 			task.ExpendTokens = gconv.Int64(mapped[entity.TotalTokens])
-			_, err = dao.ModelGatewayTask.Update(ctx, &entity.ModelGatewayTask{
+			_, _ = dao.ModelGatewayTask.Update(ctx, &entity.ModelGatewayTask{
 				SQLBaseDO:    beans.SQLBaseDO{Id: task.Id},
 				ExpendTokens: task.ExpendTokens,
 			})
@@ -326,9 +244,9 @@ func (w *asyncWorker) parseAndRetry(ctx context.Context, body map[string]any, ta
 			if err == nil {
 				return parsed, nil
 			}
+			lastErr = err
 		case public.BuildTypeStruct:
-			parsed = util.ParseStructResult(mapped, entity.ResponseBody)
-			return parsed, nil
+			return util.ParseStructResult(mapped, entity.ResponseBody), nil
 		default:
 			return mapped, nil
 		}
@@ -336,20 +254,20 @@ func (w *asyncWorker) parseAndRetry(ctx context.Context, body map[string]any, ta
 		g.Log().Warningf(ctx, "[执行任务][解析失败] taskId=%s attempt=%d/%d err=%v", task.TaskID, attempt, maxRetry, err)

 		if attempt == maxRetry {
-			return nil, fmt.Errorf("JSON解析重试耗尽: %w", err)
+			return nil, fmt.Errorf("JSON解析重试耗尽: %w", lastErr)
 		}

-		// 4) 重新调模型（直接调，不走缓存）
+		// 4) 拼接错误信息到请求体，重调模型
 		task.RetryCount++
 		_, _ = dao.ModelGatewayTask.Update(ctx, task)
-		rawData, callErr := InvokeModel(ctx, model, task.RequestPayload.Body)

+		body = injectErrorMessage(task.RequestPayload.Body, lastErr)
+		rawData, callErr := InvokeModel(ctx, model, body)
 		if callErr != nil {
 			g.Log().Warningf(ctx, "[执行任务][重调模型失败] taskId=%s attempt=%d/%d err=%v", task.TaskID, attempt, maxRetry, callErr)
 			continue
 		}

-		// 5) 解析原始响应，覆盖 body 进入下一轮
 		var rawResp map[string]any
 		if err = json.Unmarshal(rawData, &rawResp); err != nil {
 			g.Log().Warningf(ctx, "[执行任务][Unmarshal失败] taskId=%s err=%v", task.TaskID, err)
@@ -361,11 +279,53 @@ func (w *asyncWorker) parseAndRetry(ctx context.Context, body map[string]any, ta
 	return body, nil
 }

+// injectErrorMessage 将错误信息插入到最后一个 user 消息之前
+func injectErrorMessage(payload map[string]any, err error) map[string]any {
+	if err == nil {
+		return payload
+	}
+
+	messages, _ := payload["messages"].([]any)
+	if len(messages) == 0 {
+		return payload
+	}
+
+	errMsg := fmt.Sprintf("【上一轮输出错误，请修正】%s", err.Error())
+
+	// 找到最后一个 user 的位置
+	lastUserIdx := -1
+	for i := len(messages) - 1; i >= 0; i-- {
+		msg, ok := messages[i].(map[string]any)
+		if !ok {
+			continue
+		}
+		if gconv.String(msg["role"]) == "user" {
+			lastUserIdx = i
+			break
+		}
+	}
+
+	if lastUserIdx == -1 {
+		return payload
+	}
+
+	// 在最后一个 user 之前插入错误消息
+	errMsgObj := map[string]any{
+		"role":    "user",
+		"content": []map[string]any{{"type": "text", "text": errMsg}},
+	}
+
+	// 切片插入
+	messages = append(messages[:lastUserIdx], append([]any{errMsgObj}, messages[lastUserIdx:]...)...)
+	payload["messages"] = messages
+	return payload
+}
+
 // InvokeModel 调用模型服务，返回二进制结果
 // modelKey 用于覆盖/补充模型配置 head_msg（例如每次请求携带不同的 X-API-Key）
 func InvokeModel(ctx context.Context, model *entity.ModelGatewayModel, body map[string]any) ([]byte, error) {
 	// 1) 记录模型调用次数
-	_ = dao.ModelGatewayLogsStat.IncRequestCount(ctx, time.Now(), model.TenantId, model.Creator, model.ModelName)
+	//_ = dao.ModelGatewayLogsStat.IncRequestCount(ctx, time.Now(), model.TenantId, model.Creator, model.ModelName)

 	// 2）请求参数映射：将标准 payload 按模型配置的 requestMapping 转为模型需要的格式
 	//—— 请求映射实际处理为提示词构建请求，因为有附加字段及其他字段的拼接。这里不方便做请求映射
@@ -392,13 +352,20 @@ func InvokeModel(ctx context.Context, model *entity.ModelGatewayModel, body map[
 				baseURL = baseURL + "?" + q.Encode()
 			}
 		}
-		req, err = http.NewRequestWithContext(ctx, http.MethodGet, baseURL, nil)
+		// 改用独立超时ctx，隔绝外层截止
+		reqCtx, reqCancel := context.WithTimeout(context.Background(), timeout)
+		defer reqCancel()
+		req, err = http.NewRequestWithContext(reqCtx, http.MethodGet, baseURL, nil)
+		//req, err = http.NewRequestWithContext(ctx, http.MethodGet, baseURL, nil)
 	default:
 		bodyBytes, err := json.Marshal(body)
 		if err != nil {
 			return nil, err
 		}
-		req, err = http.NewRequestWithContext(ctx, http.MethodPost, baseURL, bytes.NewReader(bodyBytes))
+		reqCtx, reqCancel := context.WithTimeout(context.Background(), timeout)
+		defer reqCancel()
+		req, err = http.NewRequestWithContext(reqCtx, http.MethodPost, baseURL, bytes.NewReader(bodyBytes))
+		//req, err = http.NewRequestWithContext(ctx, http.MethodPost, baseURL, bytes.NewReader(bodyBytes))
 	}

 	// 5）注入请求头：先模型静态配置，再动态 modelKey（后者可覆盖前者）
@@ -430,6 +397,7 @@ func InvokeModel(ctx context.Context, model *entity.ModelGatewayModel, body map[
 		msg := string(b)
 		return nil, fmt.Errorf("模型服务返回非2xx: %d, body=%s", resp.StatusCode, msg)
 	}
+	g.Log().Debugf(ctx, "[执行任务][模型调用成功] StatusCode=%v", resp.StatusCode)
 	return b, nil
 }

@@ -487,15 +455,11 @@ func InvokeModel(ctx context.Context, model *entity.ModelGatewayModel, body map[
 //		return mappedResponse, nil
 //	}

-// failTask 任务失败统一处理：更新数据库 + 释放排队 + 回调
+// failTask 任务失败统一处理
 func (w *asyncWorker) failTask(ctx context.Context, t *entity.ModelGatewayTask, startTime time.Time, errMsg string) {
 	t.State = 3
 	t.ErrorMsg = errMsg
 	t.DurationSeconds = int64(time.Since(startTime).Seconds())
-	_, err := dao.ModelGatewayTask.Update(ctx, t)
-	if err != nil {
-		g.Log().Warningf(ctx, "[执行任务][更新数据库失败] taskId=%s err=%v", t.TaskID, err)
-	}
-	queue.ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-	go gateway.TriggerCallback(context.WithoutCancel(ctx), t)
+	_, _ = dao.ModelGatewayTask.Update(ctx, t)        // 更新任务状态
+	go gateway.TriggerCallback(util.AsyncCtx(ctx), t) // 触发回调
 }
--- a/update.sql
+++ b/update.sql
@@ -33,6 +33,7 @@ CREATE TABLE IF NOT EXISTS model_gateway_models (
    max_concurrency int4 NOT NULL DEFAULT 10,
    timeout_seconds int4 NOT NULL DEFAULT 600,
    retry_times     int2 NOT NULL DEFAULT 3,
+    auto_clean_seconds int4 NOT NULL DEFAULT 86400,
    response_token_field varchar(128) NOT NULL DEFAULT '',
    call_mode       int2 NOT NULL DEFAULT 0,
    required_fields jsonb NOT NULL DEFAULT '[]',
@@ -54,7 +55,6 @@ COMMENT ON COLUMN model_gateway_models.created_at IS '创建时间';
 COMMENT ON COLUMN model_gateway_models.updater IS '更新人';
 COMMENT ON COLUMN model_gateway_models.updated_at IS '更新时间';
 COMMENT ON COLUMN model_gateway_models.deleted_at IS '删除时间(软删)';
-
 COMMENT ON COLUMN model_gateway_models.model_name IS '模型名称';
 COMMENT ON COLUMN model_gateway_models.model_type IS '模型类型';
 COMMENT ON COLUMN model_gateway_models.operator_name IS '运营商名称';
@@ -62,11 +62,12 @@ COMMENT ON COLUMN model_gateway_models.base_url IS '模型地址';
 COMMENT ON COLUMN model_gateway_models.http_method IS '请求方式 GET/POST';
 COMMENT ON COLUMN model_gateway_models.head_msg IS '请求头信息';
 COMMENT ON COLUMN model_gateway_models.api_key IS '调用凭证/密钥';
+
 COMMENT ON COLUMN model_gateway_models.is_private IS '是否私有化：0-私有 1-公共';
 COMMENT ON COLUMN model_gateway_models.enabled IS '是否启用：0-停用 1-启用';
 COMMENT ON COLUMN model_gateway_models.is_chat_model IS '是否为对话模型：0-否 1-是';
 COMMENT ON COLUMN model_gateway_models.is_owner IS '1=当前用户创建 0=超级管理员';
-COMMENT ON COLUMN model_gateway_models.call_mode IS '调用模式：0-同步 1-异步 2-流式';
+
 COMMENT ON COLUMN model_gateway_models.form_json IS '动态表单结构';
 COMMENT ON COLUMN model_gateway_models.request_mapping IS '请求映射';
 COMMENT ON COLUMN model_gateway_models.response_mapping IS '返回映射';
@@ -80,7 +81,9 @@ COMMENT ON COLUMN model_gateway_models.last_frame IS '尾帧图片参数';
 COMMENT ON COLUMN model_gateway_models.max_concurrency IS '最大并发数';
 COMMENT ON COLUMN model_gateway_models.timeout_seconds IS '调用模型超时(秒)';
 COMMENT ON COLUMN model_gateway_models.retry_times IS '失败重试次数';
+COMMENT ON COLUMN model_gateway_models.auto_clean_seconds IS '任务完成后自动清理时间(秒)';
 COMMENT ON COLUMN model_gateway_models.response_token_field IS '响应中消耗token的字段映射';
+COMMENT ON COLUMN model_gateway_models.call_mode IS '调用模式：0-同步 1-异步 2-流式';
 COMMENT ON COLUMN model_gateway_models.required_fields IS '必选字段列表';
 COMMENT ON COLUMN model_gateway_models.max_tokens IS '最大 token 数，0 表示不传';
Author	SHA1	Message	Date
张斌	610481effd	1	2026-06-22 10:57:46 +08:00
qhd	39b61f1867	chore: 更新配置并优化任务错误处理将数据库、Redis、Consul及Jaeger配置切换至新环境地址，并在worker中忽略"Invalid video_url"与"Invalid audio track"错误，同时补充模型调用成功日志。	2026-06-22 09:08:12 +08:00
WangLiZhao	4cc44bf57c	fix(task): 修正错误信息注入逻辑	2026-06-18 16:27:11 +08:00
WangLiZhao	dc06d1bb9a	refactor(config): 更新数据库配置和任务处理逻辑	2026-06-18 14:48:42 +08:00
WangLiZhao	ecaaa5bdbc	refactor(files): 优化文件处理和任务服务逻辑	2026-06-18 13:39:40 +08:00
qhd	b21d7a8dbf	fix: 修复请求头转发与任务状态流转问题移除 util.ForwardHeaders，改为从原始请求精确提取 Authorization 或全部请求头；任务创建时直接设为 Running 状态，避免二次更新与查询；模型调用使用独立超时上下文，防止外层取消影响回调；增加 OSS 上传耗时日志，调整数据库连接池参数。	2026-06-18 10:08:36 +08:00
WangLiZhao	fddaf36f48	refactor(model): 优化模型网关的数据解析和任务处理逻辑	2026-06-17 14:34:48 +08:00