chore(media): 更新数据库连接凭据

移除 replace 本地依赖，改用 common v0.0.33；重名时直接返回已存在的模型。

.gitignore

@@ -1 +1,3 @@
-/.idea/*
+/.idea/*
+/.superpowers/
+/docs/superpowers/

Dockerfile

@@ -1,43 +1,23 @@
-# 多阶段构建 - 第一阶段：编译（使用已安装的镜像）
+# 阶段1: 构建
-FROM golang:1.26-alpine3.23 AS builder
+FROM golang:alpine AS builder
 
 RUN apk add --no-cache git ca-certificates tzdata
 
+ENV TZ=Asia/Shanghai
+RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
+
 ENV GO111MODULE=on
 ENV GOPROXY=https://goproxy.cn,direct
 ENV CGO_ENABLED=0
 ENV GOTOOLCHAIN=auto
-ENV GOPRIVATE=gitea.com/red-future/common
-
-# 配置git使用私有Gitea仓库（带Token认证）
-RUN git config --global url."http://x-token-auth:619679cd366aefea3a50f0622d842a41f2209e08595767bba49c3836ef57d415@116.204.74.41:3000/red-future/common.git".insteadOf "https://gitea.com/red-future/common.git" && \
-    git config --global credential.helper store
-
 WORKDIR /build
 
-# 复制父目录的 common 模块(因为 go.mod 中使用了本地 replace)
-#COPY ../common /build/common
 COPY . .
 
 RUN go mod download && go mod tidy
 
 RUN go build -ldflags="-s -w" -o main ./main.go
 
-# 第二阶段：运行
-FROM alpine:3.23
-
-ENV TIME_ZONE=Asia/Shanghai
-RUN apk add --no-cache ca-certificates tzdata && \
-    ln -sf /usr/share/zoneinfo/$TIME_ZONE /etc/localtime
-
-WORKDIR /app
-
-# 复制编译好的二进制文件
-COPY --from=builder /build/main .
-COPY --from=builder /build/config.yml ./
-
-# 创建日志目录
-RUN mkdir -p /logs /app/resource/log/run /app/resource/log/server
 
 EXPOSE 3004
 

config.yml

@@ -7,10 +7,10 @@ server:
 database:
   default:
     - type: "pgsql"
-      host: "116.204.74.41"
+      host: "192.168.0.83"
       port: "15432"
-      user: "postgres"
+      user: "sql9f15b63fd203b36e"
-      pass: "Bjang09@686^*^"
+      pass: "1ec94b1acdaf57b66030242d418fad5a"
       name: "model-gateway"
       prefix: ""                  # (可选)表名前缀
       role: "master"              # (可选)数据库主从角色(master/slave)，默认为master。如果不使用应用主从机制请不配置或留空即可。
@@ -26,20 +26,55 @@ database:
       updatedAt: "updated_at"     # (可选)自动更新时间字段名称
       deletedAt: "deleted_at"     # (可选)软删除时间字段名称
       timeMaintainDisabled: false # (可选)是否完全关闭时间更新特性，为true时CreatedAt/UpdatedAt/DeletedAt都将失效
+  model_gateway:
+    - type: "pgsql"
+      host: "192.168.0.83"
+      port: "15432"
+      user: "sql9f15b63fd203b36e"
+      pass: "1ec94b1acdaf57b66030242d418fad5a"
+      name: "model-gateway"
+      prefix: "model_gateway_"
+      role: "master"
+      debug: true
+      dryRun: false
+      charset: "utf8"
+      timezone: "Asia/Shanghai"
+      maxIdle: 15
+      maxOpen: 60
+      maxLifetime: "30s"
+      maxIdleConnTime: "30s"
+      createdAt: "created_at"
+      updatedAt: "updated_at"
+      deletedAt: "deleted_at"
+      timeMaintainDisabled: false
 
 redis:
   default:
-    address: 116.204.74.41:6379
+    address: 192.168.0.83:6379
     db: 0
 
 consul:
-  address: 116.204.74.41:8500
+  address: 192.168.0.83:8500
 
 jaeger:
-  addr: 116.204.74.41:4318
+  addr: 192.168.0.83:4318
+
+nats:
+  addr: 192.168.0.83
+  port: 4222
+
+# schema_mapping 自动构建专用 LLM（OpenAI 兼容；密钥不要写死进代码）
+schemaMapping:
+  baseUrl: "https://ark.cn-beijing.volces.com/api/v3/chat/completions"
+  modelName: "doubao-seed-2-0-lite-260428"
+  apiKey: "ark-9df744e8-a0de-4c54-9db3-18379bccd523-e6733"
 
 # 本地调试用：可选自动执行 worker/cleaner（默认关闭）
 asynch:
+  queryPending:
+    enabled: false
+    intervalSeconds: 10  # 每10秒轮询一次
+    limit: 10            # 每次查10条
   worker:
     enabled: false
     intervalSeconds: 5
@@ -48,11 +83,3 @@ asynch:
   cleaner:
     enabled: false
     intervalSeconds: 30
-
-modelType:
-  types:
-    1: "推理模型"
-    2: "图片模型"
-    3: "音频模型"
-    4: "向量化模型"
-    5: "全模态模型"

consts/model/model_supplier.go

@@ -0,0 +1,113 @@
+package model
+
+import (
+	"model-gateway/consts/public"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// 供应商编码常量
+const (
+	SupplierAliyun     = 1  // 阿里云百炼
+	SupplierVolcengine = 2  // 火山引擎
+	SupplierTencent    = 3  // 腾讯云
+	SupplierHuawei     = 4  // 华为云
+	SupplierBaidu      = 5  // 百度智能云
+	SupplierOpenAI     = 6  // OpenAI
+	SupplierAzure      = 7  // 微软 Azure
+	SupplierAWS        = 8  // 亚马逊 AWS
+	SupplierGoogle     = 9  // Google
+	SupplierDeepSeek   = 10 // DeepSeek
+	SupplierMoonshot   = 11 // Moonshot（月之暗面）
+	SupplierZhipu      = 12 // 智谱AI
+	SupplierBaichuan   = 13 // 百川智能
+	SupplierMinimax    = 14 // MiniMax
+	SupplierXunfei     = 15 // 科大讯飞
+	SupplierOthers     = 16 // 其他
+)
+
+// SupplierType 供应商编码类型
+type SupplierType *int8
+
+// SupplierItem 供应商项
+type SupplierItem struct {
+	Code SupplierType `json:"code"`
+	Desc string       `json:"desc"`
+}
+
+// 名称映射【唯一文案维护】
+var supplierNameMap = map[int]string{
+	SupplierAliyun:     "阿里云百炼",
+	SupplierVolcengine: "火山引擎",
+	SupplierTencent:    "腾讯云",
+	SupplierHuawei:     "华为云",
+	SupplierBaidu:      "百度智能云",
+	SupplierOpenAI:     "OpenAI",
+	SupplierAzure:      "微软 Azure",
+	SupplierAWS:        "亚马逊 AWS",
+	SupplierGoogle:     "Google",
+	SupplierDeepSeek:   "DeepSeek",
+	SupplierMoonshot:   "Moonshot（月之暗面）",
+	SupplierZhipu:      "智谱AI",
+	SupplierBaichuan:   "百川智能",
+	SupplierMinimax:    "MiniMax",
+	SupplierXunfei:     "科大讯飞",
+	SupplierOthers:     "其他",
+}
+
+// 供应商展示顺序
+var supplierOrder = []int{
+	// 国内云厂商
+	SupplierAliyun, SupplierVolcengine, SupplierTencent, SupplierHuawei, SupplierBaidu,
+	// 海外头部
+	SupplierOpenAI, SupplierGoogle, SupplierAWS, SupplierAzure,
+	// 国内AI厂商
+	SupplierDeepSeek, SupplierMoonshot, SupplierZhipu, SupplierBaichuan, SupplierMinimax, SupplierXunfei,
+	// 兜底
+	SupplierOthers,
+}
+
+// 全局供应商实例
+var (
+	SupplierItemAliyun     = newSupplierItem(gconv.PtrInt8(SupplierAliyun))
+	SupplierItemVolcengine = newSupplierItem(gconv.PtrInt8(SupplierVolcengine))
+	SupplierItemTencent    = newSupplierItem(gconv.PtrInt8(SupplierTencent))
+	SupplierItemHuawei     = newSupplierItem(gconv.PtrInt8(SupplierHuawei))
+	SupplierItemBaidu      = newSupplierItem(gconv.PtrInt8(SupplierBaidu))
+	SupplierItemOpenAI     = newSupplierItem(gconv.PtrInt8(SupplierOpenAI))
+	SupplierItemAzure      = newSupplierItem(gconv.PtrInt8(SupplierAzure))
+	SupplierItemAWS        = newSupplierItem(gconv.PtrInt8(SupplierAWS))
+	SupplierItemGoogle     = newSupplierItem(gconv.PtrInt8(SupplierGoogle))
+	SupplierItemDeepSeek   = newSupplierItem(gconv.PtrInt8(SupplierDeepSeek))
+	SupplierItemMoonshot   = newSupplierItem(gconv.PtrInt8(SupplierMoonshot))
+	SupplierItemZhipu      = newSupplierItem(gconv.PtrInt8(SupplierZhipu))
+	SupplierItemBaichuan   = newSupplierItem(gconv.PtrInt8(SupplierBaichuan))
+	SupplierItemMinimax    = newSupplierItem(gconv.PtrInt8(SupplierMinimax))
+	SupplierItemXunfei     = newSupplierItem(gconv.PtrInt8(SupplierXunfei))
+	SupplierItemOthers     = newSupplierItem(gconv.PtrInt8(SupplierOthers))
+)
+
+func newSupplierItem(code SupplierType) SupplierItem {
+	val := int(*code)
+	return SupplierItem{
+		Code: code,
+		Desc: supplierNameMap[val],
+	}
+}
+
+// GetSupplierDescByCode 根据编码获取供应商名称
+func GetSupplierDescByCode(code int) string {
+	return supplierNameMap[code]
+}
+
+// GetSupplierOptionList 获取供应商下拉列表
+func GetSupplierOptionList() []*public.Option {
+	var list []*public.Option
+	for _, code := range supplierOrder {
+		list = append(list, &public.Option{
+			Value: code,
+			Label: supplierNameMap[code],
+		})
+	}
+	return list
+}

consts/model/model_type.go

@@ -0,0 +1,224 @@
+package model
+
+import (
+	"model-gateway/consts/public"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// 模型类型编码常量
+const (
+	TypeInference = 100 // 推理模型
+	TypeImage     = 200 // 图片模型
+	TypeAudio     = 300 // 音频模型
+	TypeVector    = 400 // 向量模型
+	TypeOmni      = 500 // 多模态模型
+	TypeVideo     = 600 // 视频模型
+	TypeCode      = 700 // 代码模型
+
+	//// 推理子类型
+	//InferenceSubChat     = 101 // 对话/补全
+	//InferenceSubReason   = 102 // 思维链/深度推理
+	//InferenceSubFunction = 103 // 函数调用
+	//
+	//// 图片子类型
+	//ImageSubTextToImage      = 201 // 文生图
+	//ImageSubImageToImage     = 202 // 图生图
+	//ImageSubImageEdit        = 203 // 图片编辑
+	//ImageSubImageVariation   = 204 // 图片变体
+	//ImageSubImageTextToImage = 205 // 图文生图
+	//
+	//// 音频子类型
+	//AudioSubTextToSpeech   = 301 // 文生音
+	//AudioSubSpeechToText   = 302 // 音生文
+	//AudioSubSpeechToSpeech = 303 // 音生音
+	//AudioSubVoiceClone     = 304 // 声音克隆
+	//
+	//// 向量子类型
+	//VectorSubEmbedding = 401 // 文本嵌入
+	//VectorSubRerank    = 402 // 重排序
+	//
+	//// 多模态子类型
+	//OmniSubTextImageAudio = 501 // 文图音理解
+	//OmniSubVision         = 502 // 视觉理解
+	//OmniSubVideoUnderstand = 503 // 视频理解
+	//
+	//// 视频子类型
+	//VideoSubTextToVideo      = 601 // 文生视频
+	//VideoSubImageToVideo     = 602 // 图生视频
+	//VideoSubImageTextToVideo = 603 // 图文生视频
+	//VideoSubVideoToVideo     = 604 // 视频生视频
+	//
+	//// 代码子类型
+	//CodeSubGeneration = 701 // 代码生成
+	//CodeSubCompletion = 702 // 代码补全
+	//CodeSubReview     = 703 // 代码审查
+)
+
+// ModelType 编码类型
+type ModelType *int
+
+// ModelTypeItem 模型类型项
+type ModelTypeItem struct {
+	Code ModelType `json:"code"`
+	Desc string    `json:"desc"`
+}
+
+// TypeTree 树形结构
+type TypeTree struct {
+	Value    int              `json:"value"`
+	Label    string           `json:"label"`
+	Children []*public.Option `json:"children"`
+}
+
+// 名称映射表【唯一文案维护入口】
+var typeNameMap = map[int]string{
+	TypeInference: "推理模型",
+	TypeImage:     "图片模型",
+	TypeAudio:     "音频模型",
+	TypeVector:    "向量模型",
+	TypeOmni:      "多模态模型",
+	TypeVideo:     "视频模型",
+	TypeCode:      "代码模型",
+
+	//InferenceSubChat:     "对话/补全",
+	//InferenceSubReason:   "思维链/深度推理",
+	//InferenceSubFunction: "函数调用",
+	//
+	//ImageSubTextToImage:      "文生图",
+	//ImageSubImageToImage:     "图生图",
+	//ImageSubImageEdit:        "图片编辑",
+	//ImageSubImageVariation:   "图片变体",
+	//ImageSubImageTextToImage: "图文生图",
+	//
+	//AudioSubTextToSpeech:   "文生音",
+	//AudioSubSpeechToText:   "音生文",
+	//AudioSubSpeechToSpeech: "音生音",
+	//AudioSubVoiceClone:     "声音克隆",
+	//
+	//VectorSubEmbedding: "文本嵌入",
+	//VectorSubRerank:    "重排序",
+	//
+	//OmniSubTextImageAudio:  "文图音理解",
+	//OmniSubVision:          "视觉理解",
+	//OmniSubVideoUnderstand: "视频理解",
+	//
+	//VideoSubTextToVideo:      "文生视频",
+	//VideoSubImageToVideo:     "图生视频",
+	//VideoSubImageTextToVideo: "图文生视频",
+	//VideoSubVideoToVideo:     "视频生视频",
+	//
+	//CodeSubGeneration: "代码生成",
+	//CodeSubCompletion: "代码补全",
+	//CodeSubReview:     "代码审查",
+}
+
+// 父子级映射（仅存有子项的分类）
+var parentChildMap = map[int][]int{
+	//TypeInference: {InferenceSubChat, InferenceSubReason, InferenceSubFunction},
+	//TypeImage:     {ImageSubTextToImage, ImageSubImageToImage, ImageSubImageEdit, ImageSubImageVariation, ImageSubImageTextToImage},
+	//TypeAudio:     {AudioSubTextToSpeech, AudioSubSpeechToText, AudioSubSpeechToSpeech, AudioSubVoiceClone},
+	//TypeVector:    {VectorSubEmbedding, VectorSubRerank},
+	//TypeOmni:      {OmniSubTextImageAudio, OmniSubVision, OmniSubVideoUnderstand},
+	//TypeVideo:     {VideoSubTextToVideo, VideoSubImageToVideo, VideoSubImageTextToVideo, VideoSubVideoToVideo},
+	//TypeCode:      {CodeSubGeneration, CodeSubCompletion, CodeSubReview},
+}
+
+// 一级分类展示顺序
+var parentTypeOrder = []int{
+	TypeInference, TypeImage, TypeAudio, TypeVector, TypeOmni, TypeVideo, TypeCode,
+}
+
+// 全局实例：一级 + 全部二级子类型，统一通过 newItem 构造，文案仅维护在 typeNameMap
+var (
+	// 一级类型
+	ModelTypeInference = newItem(gconv.PtrInt(TypeInference))
+	ModelTypeImage     = newItem(gconv.PtrInt(TypeImage))
+	ModelTypeAudio     = newItem(gconv.PtrInt(TypeAudio))
+	ModelTypeVector    = newItem(gconv.PtrInt(TypeVector))
+	ModelTypeOmni      = newItem(gconv.PtrInt(TypeOmni))
+	ModelTypeVideo     = newItem(gconv.PtrInt(TypeVideo))
+	ModelTypeCode      = newItem(gconv.PtrInt(TypeCode))
+
+	//// 推理二级子类型
+	//ModelInferenceSubChat     = newItem(gconv.PtrInt(InferenceSubChat))
+	//ModelInferenceSubReason   = newItem(gconv.PtrInt(InferenceSubReason))
+	//ModelInferenceSubFunction = newItem(gconv.PtrInt(InferenceSubFunction))
+	//
+	//// 图片二级子类型
+	//ModelImageSubTextToImage      = newItem(gconv.PtrInt(ImageSubTextToImage))
+	//ModelImageSubImageToImage     = newItem(gconv.PtrInt(ImageSubImageToImage))
+	//ModelImageSubImageEdit        = newItem(gconv.PtrInt(ImageSubImageEdit))
+	//ModelImageSubImageVariation   = newItem(gconv.PtrInt(ImageSubImageVariation))
+	//ModelImageSubImageTextToImage = newItem(gconv.PtrInt(ImageSubImageTextToImage))
+	//
+	//// 音频二级子类型
+	//ModelAudioSubTextToSpeech   = newItem(gconv.PtrInt(AudioSubTextToSpeech))
+	//ModelAudioSubSpeechToText   = newItem(gconv.PtrInt(AudioSubSpeechToText))
+	//ModelAudioSubSpeechToSpeech = newItem(gconv.PtrInt(AudioSubSpeechToSpeech))
+	//ModelAudioSubVoiceClone     = newItem(gconv.PtrInt(AudioSubVoiceClone))
+	//
+	//// 向量二级子类型
+	//ModelVectorSubEmbedding = newItem(gconv.PtrInt(VectorSubEmbedding))
+	//ModelVectorSubRerank    = newItem(gconv.PtrInt(VectorSubRerank))
+	//
+	//// 多模态二级子类型
+	//ModelOmniSubTextImageAudio  = newItem(gconv.PtrInt(OmniSubTextImageAudio))
+	//ModelOmniSubVision          = newItem(gconv.PtrInt(OmniSubVision))
+	//ModelOmniSubVideoUnderstand = newItem(gconv.PtrInt(OmniSubVideoUnderstand))
+	//
+	//// 视频二级子类型
+	//ModelVideoSubTextToVideo      = newItem(gconv.PtrInt(VideoSubTextToVideo))
+	//ModelVideoSubImageToVideo     = newItem(gconv.PtrInt(VideoSubImageToVideo))
+	//ModelVideoSubImageTextToVideo = newItem(gconv.PtrInt(VideoSubImageTextToVideo))
+	//ModelVideoSubVideoToVideo     = newItem(gconv.PtrInt(VideoSubVideoToVideo))
+	//
+	//// 代码二级子类型
+	//ModelCodeSubGeneration = newItem(gconv.PtrInt(CodeSubGeneration))
+	//ModelCodeSubCompletion = newItem(gconv.PtrInt(CodeSubCompletion))
+	//ModelCodeSubReview     = newItem(gconv.PtrInt(CodeSubReview))
+)
+
+// newItem 构造方法：自动从 typeNameMap 读取描述
+func newItem(code ModelType) ModelTypeItem {
+	return ModelTypeItem{
+		Code: code,
+		Desc: typeNameMap[*code],
+	}
+}
+
+// GetDescByCode 根据编码获取名称
+func GetDescByCode(code int) string {
+	return typeNameMap[code]
+}
+
+// GetTypeTreeList 生成树形数据
+func GetTypeTreeList() []*TypeTree {
+	var list []*TypeTree
+	for _, parentCode := range parentTypeOrder {
+		tree := &TypeTree{
+			Value:    parentCode,
+			Label:    typeNameMap[parentCode],
+			Children: make([]*public.Option, 0),
+		}
+		if childCodes, ok := parentChildMap[parentCode]; ok {
+			for _, c := range childCodes {
+				tree.Children = append(tree.Children, &public.Option{
+					Value: c,
+					Label: typeNameMap[c],
+				})
+			}
+		}
+		list = append(list, tree)
+	}
+	return list
+}
+
+// GetAllTypeOption 全量平铺选项
+func GetAllTypeOption() []*public.Option {
+	var list []*public.Option
+	for code, label := range typeNameMap {
+		list = append(list, &public.Option{Value: code, Label: label})
+	}
+	return list
+}

consts/model/response_type.go

@@ -0,0 +1,27 @@
+package model
+
+import "github.com/gogf/gf/v2/util/gconv"
+
+var (
+	ResponseTypeSync   = newResponseType(gconv.PtrInt8(1), "sync")   // 同步
+	ResponseTypeAsync  = newResponseType(gconv.PtrInt8(2), "async")  // 异步
+	ResponseTypeStream = newResponseType(gconv.PtrInt8(3), "stream") // 流
+)
+
+type ResponseType *int8
+
+type responseType struct {
+	code ResponseType
+	desc string
+}
+
+func (s responseType) Code() ResponseType {
+	return s.code
+}
+func (s responseType) Desc() string {
+	return s.desc
+}
+
+func newResponseType(code ResponseType, desc string) responseType {
+	return responseType{code: code, desc: desc}
+}

consts/public/public.go

@@ -0,0 +1,136 @@
+package public
+
+const GmqMsgPluginsName = "gmq_model_msg"
+
+const KnowledgeLockEsKey = "knowledge:lock:knowledgeIdEs-%v"
+const KnowledgeLockSqlKey = "knowledge:lock:knowledgeIdSql-%v"
+const KnowledgeContentHashEsKey = "knowledge:knowledgeId:contentHashEs-%v"
+const KnowledgeContentHashSqlKey = "knowledge:knowledgeId:contentHashSql-%v"
+
+// Option 通用下拉选项
+type Option struct {
+	Value int    `json:"value"`
+	Label string `json:"label"`
+}
+
+const (
+	CallModeSync   = 0 // 同步调用
+	CallModeAsync  = 1 // 异步调用
+	CallModeStream = 2 // 流式调用
+)
+
+const (
+	TaskStatusPending    = 0 // 排队中
+	TaskStatusRunning    = 1 // 执行中
+	TaskStatusSuccess    = 2 // 成功
+	TaskStatusFailed     = 3 // 失败
+	TaskStatusDownloaded = 4 // 已下载
+)
+
+const (
+	BuildTypePrompt = 1 //提示词构建
+	BuildTypeNode   = 2 //节点构建
+	BuildTypeStruct = 3 //结构构建
+)
+
+// ModelType 模型类型常量
+const (
+	ModelTypeInference = 100 // 推理模型
+
+	ModelTypeImage               = 200 // 图片模型
+	ImageSubTypeTextToImage      = 201 // 图片模型-文生图
+	ImageSubTypeImageToImage     = 202 // 图片模型-图生图
+	ImageSubTypeImageEdit        = 203 // 图片模型-图片编辑
+	ImageSubTypeImageVariation   = 204 // 图片模型-图片变体
+	ImageSubTypeImageTextToImage = 205 // 图片模型-图文生图
+
+	ModelTypeAudio             = 300 // 音频模型
+	AudioSubTypeTextToSpeech   = 301 // 音频模型-文生音
+	AudioSubTypeSpeechToText   = 302 // 音频模型-音生文
+	AudioSubTypeSpeechToSpeech = 303 // 音频模型-音生音
+
+	ModelTypeVector        = 400 // 向量化模型
+	VectorSubTypeEmbedding = 401 // 向量化模型-文本嵌入
+	VectorSubTypeRerank    = 402 // 向量化模型-重排序
+
+	ModelTypeOmni             = 500 // 全模态模型
+	OmniSubTypeTextImageAudio = 501 // 全模态模型-文图音
+	OmniSubTypeVision         = 502 // 全模态模型-视觉理解
+
+	ModelTypeVideo               = 600 // 视频模型
+	VideoSubTypeTextToVideo      = 601 // 视频模型-文生视频
+	VideoSubTypeImageToVideo     = 602 // 视频模型-图生视频
+	VideoSubTypeImageTextToVideo = 603 // 视频模型-图文生视频
+	VideoSubTypeVideoToVideo     = 604 // 视频模型-视频生视频
+)
+
+// ModelTypeName 模型类型名称映射
+var ModelTypeName = map[int]string{
+	ModelTypeInference: "推理模型",
+
+	ModelTypeImage:               "图片模型",
+	ImageSubTypeTextToImage:      "图片模型-文生图",
+	ImageSubTypeImageToImage:     "图片模型-图生图",
+	ImageSubTypeImageEdit:        "图片模型-图片编辑",
+	ImageSubTypeImageVariation:   "图片模型-图片变体",
+	ImageSubTypeImageTextToImage: "图片模型-图文生图",
+
+	ModelTypeAudio:             "音频模型",
+	AudioSubTypeTextToSpeech:   "音频模型-文生音",
+	AudioSubTypeSpeechToText:   "音频模型-音生文",
+	AudioSubTypeSpeechToSpeech: "音频模型-音生音",
+
+	ModelTypeVector:        "向量化模型",
+	VectorSubTypeEmbedding: "向量化模型-文本嵌入",
+	VectorSubTypeRerank:    "向量化模型-重排序",
+
+	ModelTypeOmni:             "全模态模型",
+	OmniSubTypeTextImageAudio: "全模态模型-文图音",
+	OmniSubTypeVision:         "全模态模型-视觉理解",
+
+	ModelTypeVideo:               "视频模型",
+	VideoSubTypeTextToVideo:      "视频模型-文生视频",
+	VideoSubTypeImageToVideo:     "视频模型-图生视频",
+	VideoSubTypeImageTextToVideo: "视频模型-图文生视频",
+	VideoSubTypeVideoToVideo:     "视频模型-视频生视频",
+}
+
+// 运营商常量
+const (
+	OperatorAliyun     = "阿里云百炼"
+	OperatorVolcengine = "火山引擎"
+	OperatorTencent    = "腾讯云"
+	OperatorHuawei     = "华为云"
+	OperatorBaidu      = "百度智能云"
+	OperatorOpenAI     = "OpenAI"
+	OperatorAzure      = "Azure OpenAI"
+	OperatorAWS        = "AWS Bedrock"
+	OperatorGoogle     = "Google Cloud"
+	OperatorDeepSeek   = "DeepSeek"
+	OperatorMoonshot   = "Moonshot"
+	OperatorZhipu      = "智谱AI"
+	OperatorBaichuan   = "百川智能"
+	OperatorMinimax    = "MiniMax"
+	OperatorXunfei     = "科大讯飞"
+	OperatorOthers     = "其他"
+)
+
+// OperatorList 运营商列表（供前端下拉框使用）
+var OperatorList = []string{
+	OperatorAliyun,
+	OperatorVolcengine,
+	OperatorTencent,
+	OperatorHuawei,
+	OperatorBaidu,
+	OperatorOpenAI,
+	OperatorAzure,
+	OperatorAWS,
+	OperatorGoogle,
+	OperatorDeepSeek,
+	OperatorMoonshot,
+	OperatorZhipu,
+	OperatorBaichuan,
+	OperatorMinimax,
+	OperatorXunfei,
+	OperatorOthers,
+}

consts/public/table_name.go

@@ -1,8 +1,13 @@
 package public
 
 const (
-	TableNameModel = "asynch_models"   // 模型表
+	DbNameModelGateway = "model_gateway" //数据库名称
-	TableNameTask  = "asynch_task"     // 任务表
+)
-	TableNameOpLog = "logs_model_op"   // 操作日志表
+
-	TableNameStat  = "logs_model_stat" // 按天统计表（请求次数）
+const (
+	TableNameModelManage      = "model_manage"
+	TableNameModelSession     = "model_session"
+	TableNameModelTaskStart   = "model_task_start"
+	TableNameModelTaskEnd     = "model_task_end"
+	TableNameModelErrorMemory = "model_error_memory"
 )

controller/base.go

@@ -1 +0,0 @@
-package controller

controller/model_call_controller.go

@@ -0,0 +1,40 @@
+package controller
+
+import (
+	"context"
+	"encoding/json"
+	"model-gateway/model/dto"
+	"model-gateway/service"
+	"net/http"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// ModelCall 模型调用控制器
+var ModelCall = new(modelCall)
+
+type modelCall struct{}
+
+// ModelCall 模型调用
+func (c *modelCall) ModelCall(ctx context.Context, req *dto.ModelCallReq) (res *dto.ModelCallRes, err error) {
+	return service.ModelCall.ModelCall(ctx, req)
+}
+
+// CreateSessionStream 创建模型会话(流式)
+func (c *modelCall) CreateSessionStream(ctx context.Context, req *dto.ModelCallStreamReq) (res *beans.ResponseEmpty, err error) {
+	r := g.RequestFromCtx(ctx)
+	w := r.Response.RawWriter()
+	err = service.ModelCall.ModelCallStream(ctx, w, req)
+	if err != nil {
+		h := w.Header()
+		h.Set("Content-Type", "application/json; charset=utf-8")
+		errResp, _ := json.Marshal(map[string]interface{}{
+			"code":    http.StatusInternalServerError,
+			"message": err.Error(),
+		})
+		w.WriteHeader(http.StatusInternalServerError)
+		_, _ = w.Write(errResp)
+	}
+	return
+}

controller/model_controller.go

@@ -1,87 +0,0 @@
-package controller
-
-import (
-	"context"
-
-	"model-gateway/model/dto"
-	"model-gateway/model/entity"
-	"model-gateway/service"
-
-	"gitea.com/red-future/common/beans"
-)
-
-type model struct{}
-
-// Model 模型配置控制器
-var Model = new(model)
-
-// CreateModel 添加配置
-func (c *model) CreateModel(ctx context.Context, req *dto.CreateModelReq) (res *dto.CreateModelRes, err error) {
-	return service.Model.Create(ctx, req)
-}
-
-// UpdateModel 更改配置
-func (c *model) UpdateModel(ctx context.Context, req *dto.UpdateModelReq) (res *beans.ResponseEmpty, err error) {
-	err = service.Model.Update(ctx, req)
-	return
-}
-
-// DeleteModel 删除配置
-func (c *model) DeleteModel(ctx context.Context, req *dto.DeleteModelReq) (res *beans.ResponseEmpty, err error) {
-	err = service.Model.Delete(ctx, req.ID)
-	return
-}
-
-// GetModel 获取配置详情（按 modelName）
-func (c *model) GetModel(ctx context.Context, req *dto.GetModelReq) (res *dto.GetModelRes, err error) {
-	model, err := service.Model.Get(ctx, req.ID)
-	if err != nil {
-		return nil, err
-	}
-	if model == nil {
-		return nil, nil
-	}
-	return &dto.GetModelRes{Model: model}, nil
-}
-
-// ListModel 配置列表
-func (c *model) ListModel(ctx context.Context, req *dto.ListModelReq) (res *dto.ListModelRes, err error) {
-	list, total, err := service.Model.List(ctx, req)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.ListModelRes{
-		List:  list,
-		Total: total,
-	}, nil
-}
-
-// AutoTune 动态调参（由上层定时任务每小时触发一次）
-func (c *model) AutoTune(ctx context.Context, req *dto.AutoTuneReq) (res *dto.AutoTuneRes, err error) {
-	windowSeconds := 3600
-	if req != nil && req.WindowSeconds > 0 {
-		windowSeconds = req.WindowSeconds
-	}
-	list, err := service.AutoTune(ctx, windowSeconds)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.AutoTuneRes{List: list}, nil
-}
-
-func (c *model) ListType(ctx context.Context, req *dto.ListTypeReq) (res dto.TypeItem, err error) {
-	modelType := service.GetModelTypesFromConfig(ctx)
-	res.Type = modelType
-	return res, nil
-}
-
-// UpdateChatModel 更新是否为聊天模型
-func (c *model) UpdateChatModel(ctx context.Context, req *dto.UpdateChatModelReq) (res *beans.ResponseEmpty, err error) {
-	err = service.Model.UpdateChatModel(ctx, req)
-	return
-}
-
-// GetIsChatModel 获取是否为聊天模型
-func (c *model) GetIsChatModel(ctx context.Context, req *dto.GetIsChatModelReq) (res *entity.AsynchModel, err error) {
-	return service.Model.GetIsChatModel(ctx)
-}

controller/model_error_memory_controller.go

@@ -0,0 +1,25 @@
+package controller
+
+import (
+	"context"
+	"model-gateway/model/dto"
+	"model-gateway/service"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+// ModelErrorMemory 错误重试记忆控制器
+var ModelErrorMemory = new(modelErrorMemory)
+
+type modelErrorMemory struct{}
+
+// List 错误重试记忆列表
+func (c *modelErrorMemory) List(ctx context.Context, req *dto.GetErrorMemoryListReq) (res *dto.GetErrorMemoryListRes, err error) {
+	return service.ModelErrorMemory.List(ctx, req)
+}
+
+// Delete 删除错误重试记忆
+func (c *modelErrorMemory) Delete(ctx context.Context, req *dto.DeleteErrorMemoryReq) (res *beans.ResponseEmpty, err error) {
+	err = service.ModelErrorMemory.Delete(ctx, req)
+	return
+}

controller/model_manage_controller.go

@@ -0,0 +1,65 @@
+package controller
+
+import (
+	"context"
+	"model-gateway/model/dto"
+	"model-gateway/service"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+// ModelManage 模型配置控制器
+var ModelManage = new(modelManage)
+
+type modelManage struct{}
+
+// CreateModel 添加配置
+func (c *modelManage) CreateModel(ctx context.Context, req *dto.CreateModelManageReq) (res *dto.CreateModelManageRes, err error) {
+	return service.ModelManage.Create(ctx, req)
+}
+
+// UpdateModel 更改配置
+func (c *modelManage) UpdateModel(ctx context.Context, req *dto.UpdateModelManageReq) (res *dto.GetModelManageRes, err error) {
+	return service.ModelManage.Update(ctx, req)
+}
+
+// DeleteModel 删除配置
+func (c *modelManage) DeleteModel(ctx context.Context, req *dto.DeleteModelManageReq) (res *beans.ResponseEmpty, err error) {
+	err = service.ModelManage.Delete(ctx, req)
+	return
+}
+
+// GetModel 获取配置
+func (c *modelManage) GetModel(ctx context.Context, req *dto.GetModelManageReq) (res *dto.GetModelManageRes, err error) {
+	return service.ModelManage.Get(ctx, req)
+}
+
+// GetChatModel 获取聊天模型
+func (c *modelManage) GetChatModel(ctx context.Context, req *dto.GetChatModelReq) (res *dto.GetChatModelRes, err error) {
+	return service.ModelManage.GetChatModel(ctx, req)
+}
+
+// ListModel 配置列表
+func (c *modelManage) ListModel(ctx context.Context, req *dto.ListModelManageReq) (res *dto.ListModelManageRes, err error) {
+	return service.ModelManage.List(ctx, req)
+}
+
+// CheckChatModel 检查是否为聊天模型
+func (c *modelManage) CheckChatModel(ctx context.Context, req *dto.CheckChatModelReq) (res *dto.CheckChatModelRes, err error) {
+	return service.ModelManage.CheckChatModel(ctx, req)
+}
+
+// ListType 模型类型列表
+func (c *modelManage) ListType(ctx context.Context, req *dto.ModelTypeReq) (res *dto.ModelTypeRes, err error) {
+	return service.ModelManage.GetModelType(ctx, req)
+}
+
+// ListOperator 运营商列表
+func (c *modelManage) ListOperator(ctx context.Context, req *dto.ModelSupplierReq) (res *dto.ModelSupplierRes, err error) {
+	return service.ModelManage.GetModelSupplier(ctx, req)
+}
+
+// BuildSchemaMapping 自动构建 Schema 映射
+func (c *modelManage) BuildSchemaMapping(ctx context.Context, req *dto.BuildSchemaMappingReq) (res *dto.BuildSchemaMappingRes, err error) {
+	return service.SchemaMapping.BuildSchemaMapping(ctx, req)
+}

controller/stat_controller.go

@@ -1,18 +0,0 @@
-package controller
-
-import (
-	"context"
-
-	"model-gateway/model/dto"
-	"model-gateway/service"
-)
-
-type stat struct{}
-
-// Stat 统计控制器
-var Stat = new(stat)
-
-// ListModelStat 统计列表
-func (c *stat) ListModelStat(ctx context.Context, req *dto.ListModelStatReq) (res *dto.ListModelStatRes, err error) {
-	return service.Stat.List(ctx, req)
-}

controller/task_controller.go

@@ -1,57 +0,0 @@
-package controller
-
-import (
-	"context"
-
-	"model-gateway/model/dto"
-	"model-gateway/service"
-)
-
-type task struct{}
-
-// Task 任务控制器
-var Task = new(task)
-
-// CreateTask 根据 modelName 创建异步任务，返回 taskId
-func (c *task) CreateTask(ctx context.Context, req *dto.CreateTaskReq) (res *dto.CreateTaskRes, err error) {
-	return service.Task.Create(ctx, req)
-}
-
-// GetTaskResult 获取任务结果（只返回 oss 地址 + state）
-func (c *task) GetTaskResult(ctx context.Context, req *dto.GetTaskResultReq) (res *dto.GetTaskResultRes, err error) {
-	return service.Task.GetResult(ctx, req.TaskID)
-}
-
-// GetTaskBatch 批量查询任务（成功任务标记为已下载）
-func (c *task) GetTaskBatch(ctx context.Context, req *dto.GetTaskBatchReq) (res *dto.GetTaskBatchRes, err error) {
-	return service.Task.GetBatch(ctx, req)
-}
-
-// ListTask 任务列表分页查询
-func (c *task) ListTask(ctx context.Context, req *dto.ListTaskReq) (res *dto.ListTaskRes, err error) {
-	return service.Task.List(ctx, req)
-}
-
-// RunWork 手动触发一次 worker（由上层定时任务调用）
-func (c *task) RunWork(ctx context.Context, req *dto.RunWorkReq) (res *dto.RunWorkRes, err error) {
-	batchSize, goroutines := 10, 1
-	if req != nil {
-		if req.BatchSize > 0 {
-			batchSize = req.BatchSize
-		}
-		if req.Goroutines > 0 {
-			goroutines = req.Goroutines
-		}
-	}
-	n, err := service.AsyncWorker.RunOnce(ctx, batchSize, goroutines)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.RunWorkRes{Claimed: n}, nil
-}
-
-// CleanWork 手动触发一次 cleaner（由上层定时任务调用）
-func (c *task) CleanWork(ctx context.Context, req *dto.CleanWorkReq) (res *dto.CleanWorkRes, err error) {
-	service.Cleaner.RunOnce(ctx)
-	return &dto.CleanWorkRes{Ok: true}, nil
-}

dao/model_dao.go

@@ -1,232 +0,0 @@
-package dao
-
-import (
-	"context"
-	"fmt"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/dto"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-	"gitea.com/red-future/common/utils"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-var Model = &modelDao{}
-
-type modelDao struct{}
-
-func (d *modelDao) Insert(ctx context.Context, req *dto.CreateModelReq) (id int64, err error) {
-	asyncModel := new(entity.AsynchModel)
-	err = gconv.Struct(req, &asyncModel)
-	if err != nil {
-		return
-	}
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).Data(asyncModel).Insert()
-	if err != nil {
-		return 0, err
-	}
-	return r.LastInsertId()
-}
-
-func (d *modelDao) Update(ctx context.Context, m *dto.UpdateModelReq) (rows int64, err error) {
-	// 触发 gfdb 的 updateHook 自动填充 updater，需要显式带 updater 字段
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		OmitEmpty().
-		Where(entity.AsynchModelCol.Id, m.ID).
-		Data(m).
-		Update()
-	if err != nil {
-		return 0, err
-	}
-	return r.RowsAffected()
-}
-
-func (d *modelDao) DeleteByID(ctx context.Context, id string) (rows int64, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		Where(entity.AsynchModelCol.Id, id).
-		Delete()
-	if err != nil {
-		return 0, err
-	}
-	return r.RowsAffected()
-}
-
-func (d *modelDao) GetByModelName(ctx context.Context, modelName string) (m *entity.AsynchModel, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		Where(entity.AsynchModelCol.ModelName, modelName).
-		One()
-	if err != nil {
-		return nil, err
-	}
-	if r.IsEmpty() {
-		return nil, nil
-	}
-	err = r.Struct(&m)
-	return
-}
-
-func (d *modelDao) Get(ctx context.Context, id int64) (m *entity.AsynchModel, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		NoTenantId(ctx).
-		Where(entity.AsynchModelCol.Id, id).
-		One()
-	if err != nil {
-		return nil, err
-	}
-	if r.IsEmpty() {
-		return nil, nil
-	}
-	err = r.Struct(&m)
-	return
-}
-
-func (d *modelDao) Count(ctx context.Context, req *dto.GetModelReq) (count int, err error) {
-	count, err = gfdb.DB(ctx).Model(ctx, public.TableNameModel).OmitEmpty().
-		Where(entity.AsynchModelCol.Creator, req.Creator).
-		Where(entity.AsynchModelCol.Id, req.ID).Count()
-	return
-}
-
-func (d *modelDao) List(ctx context.Context, pageNum, pageSize int, modelNameLike string, modelType int, isPrivate int) (list []*entity.AsynchModel, total int64, err error) {
-	model := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		OrderDesc(entity.AsynchModelCol.CreatedAt)
-	if modelNameLike != "" {
-		model = model.WhereLike(entity.AsynchModelCol.ModelName, "%"+modelNameLike+"%")
-	}
-	if modelType != 0 {
-		model = model.Where(entity.AsynchModelCol.ModelType, modelType)
-	}
-	if isPrivate != 0 {
-		model = model.Where(entity.AsynchModelCol.IsPrivate, isPrivate)
-	}
-	if pageNum > 0 && pageSize > 0 {
-		model = model.Page(pageNum, pageSize)
-	}
-	r, totalInt, err := model.AllAndCount(false)
-	if err != nil {
-		return nil, 0, err
-	}
-	total = gconv.Int64(totalInt)
-	err = r.Structs(&list)
-	return
-}
-
-// ListByCreatorAndPlatform 普通用户：平台公共(tenant_id=0) + 自己创建的(creator=xxx)
-func (d *modelDao) ListByCreatorAndPlatform(ctx context.Context, creator string, pageNum, pageSize int, modelNameLike string) (list []*entity.AsynchModel, total int64, err error) {
-	// 构建 Where 条件
-	whereSQL := "deleted_at IS NULL AND (tenant_id = 1 OR creator = ?)" //1 代表超级管理员
-	args := []any{creator}
-
-	if modelNameLike != "" {
-		whereSQL += " AND model_name LIKE ?"
-		args = append(args, "%"+modelNameLike+"%")
-	}
-
-	// 查总数
-	countSQL := fmt.Sprintf("SELECT COUNT(1) FROM %s WHERE %s", public.TableNameModel, whereSQL)
-	countResult, err := gfdb.DB(ctx).GetAll(ctx, countSQL, args...)
-	if err != nil {
-		return nil, 0, err
-	}
-	if len(countResult) > 0 {
-		total = gconv.Int64(countResult[0]["count"])
-	}
-
-	// 查列表
-	querySQL := fmt.Sprintf("SELECT * FROM %s WHERE %s ORDER BY created_at DESC", public.TableNameModel, whereSQL)
-	if pageNum > 0 && pageSize > 0 {
-		offset := (pageNum - 1) * pageSize
-		querySQL += fmt.Sprintf(" LIMIT %d OFFSET %d", pageSize, offset)
-	}
-
-	r, err := gfdb.DB(ctx).GetAll(ctx, querySQL, args...)
-	if err != nil {
-		return nil, 0, err
-	}
-
-	err = r.Structs(&list)
-	return
-}
-func (d *modelDao) GetByCreatorAndPlatform(ctx context.Context, req *dto.ListModelReq) (list []*entity.AsynchModel, total int, err error) {
-	// 基础 SQL
-	sql := `
-SELECT DISTINCT ON (model_name) *
-FROM asynch_models
-WHERE deleted_at IS NULL 
-  AND (? = '' OR model_name LIKE ?)
-  AND (? = 0 OR model_type = ?)
-`
-	args := []any{
-		req.ModelName, "%" + req.ModelName + "%",
-		req.ModelType, req.ModelType,
-	}
-	if !g.IsEmpty(req.IsPrivate) {
-		sql += ` AND is_private = ? `
-		args = append(args, req.IsPrivate)
-	}
-	if req.IsOwner != nil && *req.IsOwner == 0 {
-		sql += ` AND creator = ? AND is_owner = ? `
-		args = append(args, req.Creator)
-		args = append(args, req.IsOwner)
-	} else if req.IsOwner != nil && *req.IsOwner == 1 {
-		if req.Enabled != nil && *req.Enabled == 1 {
-			sql += ` AND ((creator = ? AND is_owner = ? AND enabled=1) OR (is_owner = 0 AND enabled=1)) `
-		} else if req.Enabled != nil && *req.Enabled == 0 {
-			sql += ` AND ((creator = ? AND is_owner = ? AND enabled=0) OR (is_owner = 0 AND enabled=1)) `
-		} else {
-			sql += ` AND ((creator = ? AND is_owner = ?) OR (is_owner = 0 AND enabled=1)) `
-		}
-		args = append(args, req.Creator)
-		args = append(args, req.IsOwner)
-	}
-
-	// 最后拼接排序
-	sql += ` ORDER BY model_name, is_owner DESC, created_at DESC`
-
-	r, err := gfdb.DB(ctx).GetAll(ctx, sql, args...)
-	if err != nil {
-		return nil, 0, err
-	}
-
-	err = r.Structs(&list)
-	if err != nil {
-		return nil, 0, err
-	}
-
-	total = len(list)
-	return
-}
-
-func (d *modelDao) GetByIsChatModel(ctx context.Context) (m *entity.AsynchModel, err error) {
-	userInfo, err := utils.GetUserInfo(ctx)
-	if err != nil {
-		return nil, err
-	}
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		Where(entity.AsynchModelCol.IsChatModel, 1).
-		Where(entity.AsynchModelCol.Creator, userInfo.UserName).
-		One()
-	if err != nil {
-		return nil, err
-	}
-	if r.IsEmpty() {
-		return nil, nil
-	}
-	err = r.Struct(&m)
-	return
-}
-
-// ListAll 用于分组展示：查询全部模型（不按类型过滤，类型拆分在 service 层处理）
-func (d *modelDao) ListAll(ctx context.Context) (list []*entity.AsynchModel, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		OrderDesc(entity.AsynchModelCol.CreatedAt).
-		All()
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}

dao/model_dao_bg.go

@@ -1,32 +0,0 @@
-package dao
-
-import (
-	"context"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-)
-
-// GetByModelNameForTenant 后台任务使用：按 tenant_id + model_name 查询，不依赖 gfdb Hook/Trace/用户上下文
-func (d *modelDao) GetByModelNameForTenant(ctx context.Context, tenantId uint64, modelName string) (m *entity.AsynchModel, err error) {
-	r, err := gfdb.DB(ctx).GetAll(ctx,
-		"SELECT * FROM "+public.TableNameModel+" WHERE tenant_id=? AND model_name=? AND deleted_at IS NULL LIMIT 1",
-		tenantId, modelName,
-	)
-	if err != nil {
-		return nil, err
-	}
-	if r.IsEmpty() {
-		return nil, nil
-	}
-	var list []*entity.AsynchModel
-	if err := r.Structs(&list); err != nil {
-		return nil, err
-	}
-	if len(list) == 0 {
-		return nil, nil
-	}
-	return list[0], nil
-}

dao/model_error_memory_dao.go

@@ -0,0 +1,70 @@
+package dao
+
+import (
+	"context"
+	"model-gateway/consts/public"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+)
+
+var ModelErrorMemory = &modelErrorMemoryDao{}
+
+type modelErrorMemoryDao struct{}
+
+// GetByKey 按记忆键查询（未命中返回 (nil, nil)）
+// 错误记忆为全局表：NoTenantId 绕过租户过滤，跨租户共享；r.IsEmpty() 兜底 miss 契约，
+// 避免对空记录 r.Struct(&res) 上浮 sql.ErrNoRows 导致调用方 fail-closed。
+func (d *modelErrorMemoryDao) GetByKey(ctx context.Context, key string) (res *entity.ModelErrorMemory, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelErrorMemory).
+		NoTenantId(ctx).
+		Where(entity.ModelErrorMemoryCol.MemoryKey, key).
+		One()
+	if err != nil {
+		return
+	}
+	if r.IsEmpty() {
+		return nil, nil
+	}
+	err = r.Struct(&res)
+	return
+}
+
+// Upsert 存在则更新 retryable/reason/analyzed_by，不存在则插入
+func (d *modelErrorMemoryDao) Upsert(ctx context.Context, m *entity.ModelErrorMemory) (err error) {
+	model := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelErrorMemory)
+	// Count 同样全局化：跨租户已存在的记忆键需命中更新分支，而非重复插入
+	n, err := model.NoTenantId(ctx).Where(entity.ModelErrorMemoryCol.MemoryKey, m.MemoryKey).Count()
+	if err != nil {
+		return
+	}
+	if n > 0 {
+		_, err = model.Where(entity.ModelErrorMemoryCol.MemoryKey, m.MemoryKey).Data(map[string]any{
+			entity.ModelErrorMemoryCol.Retryable:  m.Retryable,
+			entity.ModelErrorMemoryCol.Reason:     m.Reason,
+			entity.ModelErrorMemoryCol.AnalyzedBy: m.AnalyzedBy,
+		}).Update()
+		return
+	}
+	_, err = model.Insert(m)
+	return
+}
+
+// List 分页查询（按 id 倒序）；全局表，管理端列表展示所有租户记忆
+func (d *modelErrorMemoryDao) List(ctx context.Context, page, pageSize int) (list []entity.ModelErrorMemory, total int64, err error) {
+	model := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelErrorMemory).NoTenantId(ctx)
+	n, err := model.Count()
+	if err != nil {
+		return
+	}
+	total = int64(n)
+	err = model.Page(page, pageSize).OrderDesc(entity.ModelErrorMemoryCol.Id).Scan(&list)
+	return
+}
+
+// Delete 按 id 删除（软删除）
+func (d *modelErrorMemoryDao) Delete(ctx context.Context, id int64) (err error) {
+	_, err = gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelErrorMemory).
+		Where(entity.ModelErrorMemoryCol.Id, id).Delete()
+	return
+}

dao/model_manage_dao.go

@@ -0,0 +1,229 @@
+package dao
+
+import (
+	"context"
+	"fmt"
+	"model-gateway/consts/public"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"github.com/gogf/gf/v2/database/gdb"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelManage = &modelManageDao{}
+
+type modelManageDao struct{}
+
+// Insert 插入
+func (d *modelManageDao) Insert(ctx context.Context, req *dto.CreateModelManageReq) (id int64, err error) {
+	var e = new(entity.ModelManage)
+	err = gconv.Struct(req, &e)
+	if err != nil {
+		return
+	}
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).Insert(e)
+	if err != nil {
+		return
+	}
+	return r.LastInsertId()
+}
+
+// Update 更新
+func (d *modelManageDao) Update(ctx context.Context, req *dto.UpdateModelManageReq) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).OmitEmpty().Data(req).Where(entity.ModelManageCol.Id, req.Id).Update()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}
+
+// Delete 删除
+func (d *modelManageDao) Delete(ctx context.Context, req *dto.DeleteModelManageReq) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).Where(entity.ModelManageCol.Id, req.Id).Delete()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}
+
+func (d *modelManageDao) Get(ctx context.Context, req *dto.GetModelManage, fields ...string) (res *entity.ModelManage, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).Cache(ctx).
+		OmitEmpty().
+		Where(entity.ModelManageCol.ModelName, req.ModelName).
+		Where(entity.ModelManageCol.ChatModel, req.ChatModel).
+		Where(entity.ModelManageCol.Creator, req.Creator).
+		Fields(fields).One()
+	if err != nil {
+		return
+	}
+	err = r.Struct(&res)
+	return
+}
+
+// GetByCreatorAndName 按创建人+模型名精确查询（无缓存），用于同一用户下的同名唯一性校验。
+// 走 Model 链（自动过滤软删除），与 Get 一致但不带 Cache，避免缓存过期导致重复放行。
+func (d *modelManageDao) GetByCreatorAndName(ctx context.Context, creator, modelName string) (res *entity.ModelManage, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).
+		OmitEmpty().
+		Where(entity.ModelManageCol.ModelName, modelName).
+		Where(entity.ModelManageCol.Creator, creator).
+		One()
+	if err != nil {
+		return
+	}
+	err = r.Struct(&res)
+	return
+}
+
+// CountReferences 统计引用某系统模型的引用行数（Model 链自动过滤软删）
+func (d *modelManageDao) CountReferences(ctx context.Context, systemModelId int64) (count int, err error) {
+	count, err = gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).
+		Where(entity.ModelManageCol.RefSystemModelId, systemModelId).
+		Count()
+	return
+}
+
+// UpdateReferencesName 系统模型改名时同步引用行的 model_name（保列表 DISTINCT ON 去重正确）
+func (d *modelManageDao) UpdateReferencesName(ctx context.Context, systemModelId int64, newName string) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelManage).
+		Data(gdb.Map{entity.ModelManageCol.ModelName: newName}).
+		Where(entity.ModelManageCol.RefSystemModelId, systemModelId).
+		Update()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}
+
+func (d *modelManageDao) GetNotTenantId(ctx context.Context, req *dto.GetModelManageReq, fields ...string) (res *entity.ModelManage, err error) {
+	// 获取表前缀
+	prefix := g.Cfg().MustGet(ctx, fmt.Sprintf("database.%s.0.prefix", public.DbNameModelGateway)).String()
+	table := prefix + public.TableNameModelManage
+	// 动态拼接 SELECT 列
+	var field string
+	if !g.IsEmpty(fields) {
+		for k, v := range fields {
+			if k == len(fields)-1 {
+				field = field + v
+			} else {
+				field = field + v + ","
+			}
+		}
+	} else {
+		field = "*"
+	}
+	// 动态拼接 WHERE 条件
+	var whereCondition string
+	var queryParams []interface{}
+	if !g.IsEmpty(req.Id) {
+		whereCondition = fmt.Sprintf(" AND %s=(?) ", entity.ModelManageCol.Id)
+		queryParams = append(queryParams, req.Id)
+	}
+	whereCondition = whereCondition + " AND " + entity.ModelManageCol.DeletedAt + " IS NULL "
+
+	sql := `SELECT ` + field + ` FROM ` + table + ` WHERE 1=1 ` + whereCondition + ``
+	// 执行查询
+	result, err := gfdb.DB(ctx, public.DbNameModelGateway).GetOne(ctx, sql, queryParams...)
+	if err != nil {
+		return nil, err
+	}
+	err = result.Struct(&res)
+	return
+}
+func (d *modelManageDao) ListNotTenantId(ctx context.Context, req *dto.ListModelManageReq, fields ...string) (res []*entity.ModelManage, total int, err error) {
+	// 获取表前缀
+	prefix := g.Cfg().MustGet(ctx, fmt.Sprintf("database.%s.0.prefix", public.DbNameModelGateway)).String()
+	table := prefix + public.TableNameModelManage
+
+	// 动态拼接 SELECT 列
+	var field string
+	if !g.IsEmpty(fields) {
+		for k, v := range fields {
+			if k == len(fields)-1 {
+				field = field + v
+			} else {
+				field = field + v + ","
+			}
+		}
+	} else {
+		field = "*"
+	}
+
+	// 动态拼接 WHERE 条件
+	var whereCondition string
+	var queryParams []interface{}
+
+	if !g.IsEmpty(req.ModelName) {
+		whereCondition += fmt.Sprintf(" AND %s=(?) ", entity.ModelManageCol.ModelName)
+		queryParams = append(queryParams, req.ModelName)
+	}
+	if !g.IsEmpty(req.ModelType) {
+		whereCondition += fmt.Sprintf(" AND %s=(?) ", entity.ModelManageCol.ModelType)
+		queryParams = append(queryParams, req.ModelType)
+	}
+	if !g.IsEmpty(req.Creator) {
+		whereCondition += fmt.Sprintf(" AND (%s=(?) OR %s=true) ", entity.ModelManageCol.Creator, entity.ModelManageCol.SystemModel)
+		queryParams = append(queryParams, req.Creator)
+	}
+	whereCondition = whereCondition + " AND " + entity.ModelManageCol.DeletedAt + " IS NULL "
+
+	// 1. 统计去重后总条数
+	countSql := fmt.Sprintf(
+		`SELECT COUNT(DISTINCT %s) FROM %s WHERE 1=1 %s`,
+		entity.ModelManageCol.ModelName,
+		table,
+		whereCondition,
+	)
+	countResult, err := gfdb.DB(ctx, public.DbNameModelGateway).GetOne(ctx, countSql, queryParams...)
+	if err != nil {
+		return nil, 0, err
+	}
+	type crr struct {
+		Count int64 `db:"count"`
+	}
+	var cr crr
+	if err = countResult.Struct(&cr); err != nil {
+		return nil, 0, err
+	}
+	total = int(cr.Count)
+	// 2. 分页处理
+	limitSql := ""
+	if req.Page != nil {
+		pageNum := int(req.Page.PageNum)
+		pageSize := int(req.Page.PageSize)
+		offset := (pageNum - 1) * pageSize
+		limitSql = fmt.Sprintf(" LIMIT ? OFFSET ? ")
+		// PG 语法 LIMIT 条数 OFFSET 偏移量
+		queryParams = append(queryParams, pageSize, offset)
+	}
+
+	// 排序优先级：1.分组字段ModelName  2.SystemModel升序(false在前，保留用户数据)  3.创建时间倒序
+	orderSql := fmt.Sprintf(
+		" ORDER BY %s, %s ASC, %s DESC ",
+		entity.ModelManageCol.ModelName,
+		entity.ModelManageCol.SystemModel,
+		entity.ModelManageCol.CreatedAt,
+	)
+
+	// PG DISTINCT ON 按模型名去重，同名只取第一条(用户数据)
+	sql := fmt.Sprintf(
+		`SELECT DISTINCT ON (%s) %s FROM %s WHERE 1=1 %s %s %s`,
+		entity.ModelManageCol.ModelName,
+		field,
+		table,
+		whereCondition,
+		orderSql,
+		limitSql,
+	)
+
+	// 执行查询
+	result, err := gfdb.DB(ctx, public.DbNameModelGateway).GetAll(ctx, sql, queryParams...)
+	if err != nil {
+		return nil, 0, err
+	}
+	err = result.Structs(&res)
+	return
+}

dao/model_session_dao.go

@@ -0,0 +1,42 @@
+package dao
+
+import (
+	"context"
+	"model-gateway/consts/public"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelSession = &modelSessionDao{}
+
+type modelSessionDao struct{}
+
+// Insert 插入
+func (d *modelSessionDao) Insert(ctx context.Context, req *dto.CreateModelSessionReq) (id int64, err error) {
+	m := new(entity.ModelSession)
+	err = gconv.Struct(req, &m)
+	if err != nil {
+		return
+	}
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelSession).Insert(m)
+	if err != nil {
+		return
+	}
+	return r.LastInsertId()
+}
+
+// Update 更新（按ID）
+func (d *modelSessionDao) Update(ctx context.Context, req *dto.UpdateModelSessionReq) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelSession).
+		OmitEmpty().
+		Data(req).
+		Where(entity.ModelSessionCol.Id, req.Id).
+		Update()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}

dao/model_task_end_dao.go

@@ -0,0 +1,29 @@
+package dao
+
+import (
+	"context"
+	"model-gateway/consts/public"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelTaskEnd = &modelTaskEndDao{}
+
+type modelTaskEndDao struct{}
+
+// Insert 插入
+func (d *modelTaskEndDao) Insert(ctx context.Context, req *dto.CreateModelTaskEndReq) (id int64, err error) {
+	m := new(entity.ModelTaskEnd)
+	err = gconv.Struct(req, &m)
+	if err != nil {
+		return
+	}
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelTaskEnd).Insert(m)
+	if err != nil {
+		return
+	}
+	return r.LastInsertId()
+}

dao/model_task_start_dao.go

@@ -0,0 +1,97 @@
+package dao
+
+import (
+	"context"
+	"fmt"
+	"model-gateway/consts/public"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelTaskStart = &modelTaskStartDao{}
+
+type modelTaskStartDao struct{}
+
+// Insert 插入
+func (d *modelTaskStartDao) Insert(ctx context.Context, req *dto.CreateModelTaskStartReq) (id int64, err error) {
+	m := new(entity.ModelTaskStart)
+	err = gconv.Struct(req, &m)
+	if err != nil {
+		return
+	}
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelTaskStart).Insert(m)
+	if err != nil {
+		return
+	}
+	return r.LastInsertId()
+}
+
+// Update 更新（按ID）
+func (d *modelTaskStartDao) Update(ctx context.Context, req *dto.UpdateModelTaskStartReq) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelTaskStart).
+		OmitEmpty().
+		Data(req).
+		Where(entity.ModelTaskStartCol.Id, req.Id).
+		Update()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}
+
+func (d *modelTaskStartDao) Delete(ctx context.Context, req *dto.DeleteModelTaskStartReq) (rows int64, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameModelGateway).Model(ctx, public.TableNameModelTaskStart).
+		Where(entity.ModelTaskStartCol.Id, req.Id).
+		Delete()
+	if err != nil {
+		return
+	}
+	return r.RowsAffected()
+}
+
+func (d *modelTaskStartDao) ListByLimitNotTenantId(ctx context.Context, req *dto.GetModelTaskStartListReq, fields ...string) (res []entity.ModelTaskStart, err error) {
+	// 获取表前缀
+	prefix := g.Cfg().MustGet(ctx, fmt.Sprintf("database.%s.0.prefix", public.DbNameModelGateway)).String()
+	table := prefix + public.TableNameModelTaskStart
+	// 动态拼接 SELECT 列
+	var field string
+	if !g.IsEmpty(fields) {
+		for k, v := range fields {
+			if k == len(fields)-1 {
+				field = field + v
+			} else {
+				field = field + v + ","
+			}
+		}
+	} else {
+		field = "*"
+	}
+	// 动态拼接 WHERE 条件
+	var whereCondition string
+	whereCondition = whereCondition + fmt.Sprintf(" AND %s != '' ", entity.ModelTaskStartCol.TaskId)
+	whereCondition = whereCondition + fmt.Sprintf(" AND %s IS NULL ", entity.ModelTaskStartCol.DeletedAt)
+	// 排序
+	orderSql := fmt.Sprintf(" ORDER BY %s ASC ", entity.ModelTaskStartCol.CreatedAt)
+	// 分页
+	limitSql := ""
+	if req.Page != nil {
+		pageNum := int(req.Page.PageNum)
+		pageSize := int(req.Page.PageSize)
+		offset := (pageNum - 1) * pageSize
+		limitSql = fmt.Sprintf(" LIMIT %d OFFSET %d ", pageSize, offset)
+	}
+	// 查询
+	sql := `SELECT ` + field + ` FROM ` + table + ` WHERE 1=1 ` + whereCondition + orderSql + limitSql + ``
+	// 执行查询
+	result, err := gfdb.DB(ctx, public.DbNameModelGateway).GetAll(ctx, sql)
+	if err != nil {
+		return nil, err
+	}
+	err = result.Structs(&res)
+
+	return
+}

dao/op_log_dao.go

@@ -1,22 +0,0 @@
-package dao
-
-import (
-	"context"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-)
-
-type opLogDao struct{}
-
-var OpLog = &opLogDao{}
-
-func (d *opLogDao) Insert(ctx context.Context, log *entity.LogsModelOp) (id int64, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameOpLog).Data(log).Insert()
-	if err != nil {
-		return 0, err
-	}
-	return r.LastInsertId()
-}

dao/stat_dao.go

@@ -1,60 +0,0 @@
-package dao
-
-import (
-	"context"
-	"fmt"
-	"time"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-	"github.com/gogf/gf/v2/os/gtime"
-)
-
-type statDao struct{}
-
-var Stat = &statDao{}
-
-// IncRequestCount 原子累加（支持分布式/多协程）：按天+租户+创建人+模型 +1
-func (d *statDao) IncRequestCount(ctx context.Context, day time.Time, tenantId int64, creator, modelName string) error {
-	sql := fmt.Sprintf(`
-INSERT INTO %s(day, tenant_id, creator, model_name, request_count, created_at, updated_at)
-VALUES(?, ?, ?, ?, 1, NOW(), NOW())
-ON CONFLICT (day, tenant_id, creator, model_name)
-DO UPDATE SET request_count = %s.request_count + 1, updated_at = NOW()`,
-		public.TableNameStat, public.TableNameStat,
-	)
-	_, err := gfdb.DB(ctx).Exec(ctx, sql, gtime.New(day).Format("Y-m-d"), tenantId, creator, modelName)
-	return err
-}
-
-func (d *statDao) List(ctx context.Context, pageNum, pageSize int, startDay, endDay string, tenantId *int64, creator, modelName string) (list []*entity.LogsModelStat, total int64, err error) {
-	m := gfdb.DB(ctx).Model(ctx, public.TableNameStat).Where("1=1")
-	if startDay != "" {
-		m = m.Where("day >= ?", startDay)
-	}
-	if endDay != "" {
-		m = m.Where("day <= ?", endDay)
-	}
-	if tenantId != nil {
-		m = m.Where("tenant_id = ?", *tenantId)
-	}
-	if creator != "" {
-		m = m.WhereLike("creator", "%"+creator+"%")
-	}
-	if modelName != "" {
-		m = m.WhereLike("model_name", "%"+modelName+"%")
-	}
-	m = m.OrderDesc("day").OrderDesc("request_count")
-	if pageNum > 0 && pageSize > 0 {
-		m = m.Page(pageNum, pageSize)
-	}
-	r, totalInt, err := m.AllAndCount(false)
-	if err != nil {
-		return nil, 0, err
-	}
-	total = int64(totalInt)
-	err = r.Structs(&list)
-	return
-}

dao/task_dao.go

@@ -1,250 +0,0 @@
-package dao
-
-import (
-	"context"
-	"fmt"
-	"time"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-	"github.com/gogf/gf/v2/database/gdb"
-	"github.com/gogf/gf/v2/os/gtime"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-var Task = &taskDao{}
-
-type taskDao struct{}
-
-func (d *taskDao) Insert(ctx context.Context, t *entity.AsynchTask) (id int64, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).Data(t).Insert()
-	if err != nil {
-		return 0, err
-	}
-	return r.LastInsertId()
-}
-
-func (d *taskDao) GetByTaskID(ctx context.Context, taskID string) (t *entity.AsynchTask, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.TaskID, taskID).
-		One()
-	if err != nil {
-		return nil, err
-	}
-	if r.IsEmpty() {
-		return nil, nil
-	}
-	err = r.Struct(&t)
-	return
-}
-
-// ListByTaskIDs 批量查询任务（会受 gfdb 的租户 Hook 影响，只返回当前租户数据）
-func (d *taskDao) ListByTaskIDs(ctx context.Context, taskIDs []string) (list []*entity.AsynchTask, err error) {
-	if len(taskIDs) == 0 {
-		return nil, nil
-	}
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		WhereIn(entity.AsynchTaskCol.TaskID, taskIDs).
-		All()
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// MarkDownloadedByID 将成功任务标记为已下载(state=4)，并写入过期时间
-func (d *taskDao) MarkDownloadedByID(ctx context.Context, id int64, expireAt *gtime.Time) error {
-	data := gdb.Map{
-		entity.AsynchTaskCol.State:    4,
-		entity.AsynchTaskCol.ExpireAt: expireAt,
-		entity.AsynchTaskCol.Updater:  "",
-	}
-	_, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.Id, id).
-		Where(entity.AsynchTaskCol.State, 2).
-		Data(data).
-		Update()
-	return err
-}
-
-func (d *taskDao) UpdateRunning(ctx context.Context, id int64) error {
-	now := gtime.Now()
-	data := gdb.Map{
-		entity.AsynchTaskCol.State:     1,
-		entity.AsynchTaskCol.StartedAt: now,
-		entity.AsynchTaskCol.Updater:   "",
-	}
-	_, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.Id, id).
-		Data(data).
-		Update()
-	return err
-}
-
-func (d *taskDao) UpdateSuccess(ctx context.Context, id int64, ossFile, fileType string, fileSize int64, expireAt *gtime.Time) error {
-	now := gtime.Now()
-	data := gdb.Map{
-		entity.AsynchTaskCol.State:      2,
-		entity.AsynchTaskCol.OssFile:    ossFile,
-		entity.AsynchTaskCol.FileType:   fileType,
-		entity.AsynchTaskCol.FileSize:   fileSize,
-		entity.AsynchTaskCol.ErrorMsg:   "",
-		entity.AsynchTaskCol.FinishedAt: now,
-		entity.AsynchTaskCol.ExpireAt:   expireAt,
-		entity.AsynchTaskCol.Updater:    "",
-	}
-	_, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.Id, id).
-		Data(data).
-		Update()
-	return err
-}
-
-func (d *taskDao) UpdateFailed(ctx context.Context, id int64, errorMsg string) error {
-	now := gtime.Now()
-	data := gdb.Map{
-		entity.AsynchTaskCol.State:      3,
-		entity.AsynchTaskCol.ErrorMsg:   errorMsg,
-		entity.AsynchTaskCol.FinishedAt: now,
-		entity.AsynchTaskCol.Updater:    "",
-	}
-	_, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.Id, id).
-		Data(data).
-		Update()
-	return err
-}
-
-func (d *taskDao) SoftDeleteByTaskID(ctx context.Context, taskID string) (rows int64, err error) {
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.TaskID, taskID).
-		Delete()
-	if err != nil {
-		return 0, err
-	}
-	return r.RowsAffected()
-}
-
-// CountActiveByModel 统计某模型排队中/执行中的任务数，用于 queue_limit 限制（近似值）
-func (d *taskDao) CountActiveByModel(ctx context.Context, modelName string) (int64, error) {
-	n, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.ModelName, modelName).
-		WhereIn(entity.AsynchTaskCol.State, []int{0, 1}).
-		Count()
-	return int64(n), err
-}
-
-// List 任务分页查询（受 gfdb 租户 Hook 影响）
-func (d *taskDao) List(ctx context.Context, pageNum, pageSize int, modelNameLike, taskIDLike string, state *int) (list []*entity.AsynchTask, total int64, err error) {
-	m := gfdb.DB(ctx).Model(ctx, public.TableNameTask).Where("deleted_at IS NULL")
-	if modelNameLike != "" {
-		m = m.WhereLike(entity.AsynchTaskCol.ModelName, "%"+modelNameLike+"%")
-	}
-	if taskIDLike != "" {
-		m = m.WhereLike(entity.AsynchTaskCol.TaskID, "%"+taskIDLike+"%")
-	}
-	if state != nil {
-		m = m.Where(entity.AsynchTaskCol.State, *state)
-	}
-	m = m.OrderDesc(entity.AsynchTaskCol.CreatedAt)
-	if pageNum > 0 && pageSize > 0 {
-		m = m.Page(pageNum, pageSize)
-	}
-	r, totalInt, err := m.AllAndCount(false)
-	if err != nil {
-		return nil, 0, err
-	}
-	total = gconv.Int64(totalInt)
-	err = r.Structs(&list)
-	return
-}
-
-// ClaimPending 抢占 pending 任务（state=0），并在同一事务中更新为 running（state=1）
-// 使用 PostgreSQL: FOR UPDATE SKIP LOCKED 避免多 worker 重复消费
-func (d *taskDao) ClaimPending(ctx context.Context, batchSize int) (tasks []*entity.AsynchTask, err error) {
-	if batchSize <= 0 {
-		batchSize = 1
-	}
-	err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) error {
-		sql := fmt.Sprintf(
-			`SELECT id, tenant_id, model_name, task_id, input_ref, request_payload
-			   FROM %s
-			  WHERE deleted_at IS NULL AND state = 0
-			  ORDER BY created_at ASC
-			  LIMIT %d
-			  FOR UPDATE SKIP LOCKED`,
-			public.TableNameTask,
-			batchSize,
-		)
-		r, err := tx.GetAll(sql)
-		if err != nil {
-			return err
-		}
-		if r.IsEmpty() {
-			tasks = nil
-			return nil
-		}
-		if err := r.Structs(&tasks); err != nil {
-			return err
-		}
-		// 更新为 running
-		now := time.Now()
-		for _, t := range tasks {
-			// tx.Model 不走 gfdb Hook，这里手动更新必要字段
-			_, err = tx.Exec(
-				fmt.Sprintf(`UPDATE %s SET state=1, started_at=?, updated_at=? WHERE id=?`, public.TableNameTask),
-				now, now, t.Id,
-			)
-			if err != nil {
-				return err
-			}
-		}
-		return nil
-	})
-	return
-}
-
-// ListExpiredSuccess 获取已成功且过期的任务
-func (d *taskDao) ListExpiredSuccess(ctx context.Context, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 100
-	}
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		Where(entity.AsynchTaskCol.State, 2).
-		Where(entity.AsynchTaskCol.ExpireAt+" IS NOT NULL").
-		Where(entity.AsynchTaskCol.ExpireAt+" < ?", gtime.Now()).
-		Limit(limit).
-		All()
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// ListTimeoutTasks 获取超时的排队/执行中任务
-func (d *taskDao) ListTimeoutTasks(ctx context.Context, timeout time.Duration, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 100
-	}
-	deadline := gtime.New(time.Now().Add(-timeout))
-	r, err := gfdb.DB(ctx).Model(ctx, public.TableNameTask).
-		WhereIn(entity.AsynchTaskCol.State, []int{0, 1}).
-		Where(entity.AsynchTaskCol.UpdatedAt+" < ?", deadline).
-		Limit(limit).
-		All()
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// DebugPing 用于启动时检测数据库连通性（可选）
-func (d *taskDao) DebugPing(ctx context.Context) error {
-	_, err := gfdb.DB(ctx).GetAll(ctx, "SELECT 1")
-	return err
-}

dao/task_dao_bg.go

@@ -1,286 +0,0 @@
-package dao
-
-import (
-	"context"
-	"fmt"
-	"time"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-	"github.com/gogf/gf/v2/database/gdb"
-	"github.com/gogf/gf/v2/os/gtime"
-)
-
-// ClaimPendingGlobal 后台任务使用：全局抢占 pending 任务（不加 tenant 过滤）
-func (d *taskDao) ClaimPendingGlobal(ctx context.Context, batchSize int) (tasks []*entity.AsynchTask, err error) {
-	if batchSize <= 0 {
-		batchSize = 1
-	}
-	err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) error {
-		sql := fmt.Sprintf(
-			`SELECT id, tenant_id, creator, model_name, task_id, biz_name, callback_url, model_key, input_ref, request_payload, phase, tmp_file
-			   FROM %s
-			  WHERE deleted_at IS NULL AND state = 0
-			  ORDER BY enqueue_at ASC
-			  LIMIT %d
-			  FOR UPDATE SKIP LOCKED`,
-			public.TableNameTask,
-			batchSize,
-		)
-		r, err := tx.GetAll(sql)
-		if err != nil {
-			return err
-		}
-		if r.IsEmpty() {
-			tasks = nil
-			return nil
-		}
-		if err := r.Structs(&tasks); err != nil {
-			return err
-		}
-		now := time.Now()
-		for _, t := range tasks {
-			_, err = tx.Exec(
-				fmt.Sprintf(`UPDATE %s SET state=1, started_at=?, updated_at=? WHERE id=?`, public.TableNameTask),
-				now, now, t.Id,
-			)
-			if err != nil {
-				return err
-			}
-		}
-		return nil
-	})
-	return
-}
-
-// ClaimPendingByTaskIDGlobal 按 task_id 定向抢占单个 pending 任务（不加 tenant 过滤）
-// 用于 createTask 创建成功后立即异步尝试执行当前任务，避免只依赖后续 runWork 扫描队列。
-func (d *taskDao) ClaimPendingByTaskIDGlobal(ctx context.Context, taskID string) (task *entity.AsynchTask, err error) {
-	if taskID == "" {
-		return nil, nil
-	}
-	err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) error {
-		sql := fmt.Sprintf(
-			`SELECT id, tenant_id, creator, model_name, task_id, biz_name, callback_url, model_key, input_ref, request_payload, phase, tmp_file
-			   FROM %s
-			  WHERE deleted_at IS NULL AND state = 0 AND task_id = ?
-			  LIMIT 1
-			  FOR UPDATE SKIP LOCKED`,
-			public.TableNameTask,
-		)
-		r, err := tx.GetOne(sql, taskID)
-		if err != nil {
-			return err
-		}
-		if r.IsEmpty() {
-			task = nil
-			return nil
-		}
-		if err := r.Struct(&task); err != nil {
-			return err
-		}
-		now := time.Now()
-		_, err = tx.Exec(
-			fmt.Sprintf(`UPDATE %s SET state=1, started_at=?, updated_at=? WHERE id=?`, public.TableNameTask),
-			now, now, task.Id,
-		)
-		return err
-	})
-	return
-}
-
-func (d *taskDao) UpdateSuccessGlobal(ctx context.Context, id int64, ossFile, fileType, textResult string, fileSize int64, expireAt *gtime.Time, expendTokens int) error {
-	now := gtime.Now()
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s
-SET state=2,
-    oss_file=?,
-    file_type=?,
-    text_result=?,
-	expend_tokens=?,
-    file_size=?,
-    error_msg='',
-    finished_at=?,
-    duration_seconds=EXTRACT(EPOCH FROM (? - created_at))::BIGINT,
-    expire_at=NULL,
-    phase=0,
-    tmp_file='',
-    updated_at=?
-WHERE id=?`, public.TableNameTask),
-		ossFile, fileType, textResult, expendTokens, fileSize, now, now, now, id,
-	)
-	return err
-}
-
-func (d *taskDao) UpdateFailedGlobal(ctx context.Context, id int64, errorMsg string) error {
-	now := gtime.Now()
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s
-SET state=3,
-    error_msg=?,
-    finished_at=?,
-    duration_seconds=EXTRACT(EPOCH FROM (? - created_at))::BIGINT,
-    phase=0,
-    tmp_file='',
-    updated_at=?
-WHERE id=?`, public.TableNameTask),
-		errorMsg, now, now, now, id,
-	)
-	return err
-}
-
-// UpdateFailedKeepTmpGlobal OSS 上传失败：保留 phase/tmp_file，下一轮仅重试 OSS 上传
-func (d *taskDao) UpdateFailedKeepTmpGlobal(ctx context.Context, id int64, errorMsg string) error {
-	now := gtime.Now()
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s SET state=3, error_msg=?, finished_at=?, phase=1, updated_at=? WHERE id=?`, public.TableNameTask),
-		errorMsg, now, now, id,
-	)
-	return err
-}
-
-// UpdateTmpAfterModelGlobal 模型调用成功后，写入临时文件路径并标记 phase=1
-func (d *taskDao) UpdateTmpAfterModelGlobal(ctx context.Context, id int64, tmpFile string) error {
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s SET phase=1, tmp_file=?, updated_at=NOW() WHERE id=?`, public.TableNameTask),
-		tmpFile, id,
-	)
-	return err
-}
-
-func (d *taskDao) SoftDeleteByTaskIDGlobal(ctx context.Context, taskID string) error {
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s SET deleted_at=NOW(), updated_at=NOW() WHERE task_id=? AND deleted_at IS NULL`, public.TableNameTask),
-		taskID,
-	)
-	return err
-}
-
-func (d *taskDao) RollbackToPendingGlobal(ctx context.Context, id int64) error {
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s SET state=0, enqueue_at=NOW(), updated_at=NOW() WHERE id=? AND state=1`, public.TableNameTask),
-		id,
-	)
-	return err
-}
-
-// ListExpiredDownloadedGlobal 获取已下载(state=4)且过期的任务，用于清理
-func (d *taskDao) ListExpiredDownloadedGlobal(ctx context.Context, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 200
-	}
-	r, err := gfdb.DB(ctx).GetAll(ctx,
-		fmt.Sprintf(`SELECT * FROM %s WHERE deleted_at IS NULL AND state=4 AND expire_at IS NOT NULL AND expire_at < ? LIMIT ?`, public.TableNameTask),
-		gtime.Now(), limit,
-	)
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// ListFailedRetryableGlobal 获取失败(state=3)且仍可重试的任务
-// retry_count 不含首次执行；retry_times 表示失败后最多再重试 N 次
-func (d *taskDao) ListFailedRetryableGlobal(ctx context.Context, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 200
-	}
-	r, err := gfdb.DB(ctx).GetAll(ctx,
-		fmt.Sprintf(`
-SELECT t.*,
-       m.retry_queue_max_seconds AS retry_queue_max_seconds
-  FROM %s t
-  JOIN %s m
-    ON t.tenant_id = m.tenant_id
-   AND t.model_name = m.model_name
- WHERE t.deleted_at IS NULL
-   AND t.state = 3
-   AND t.retry_count < m.retry_times
- ORDER BY t.updated_at ASC
- LIMIT ?`, public.TableNameTask, public.TableNameModel),
-		limit,
-	)
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// RequeueForRetryGlobal 将任务重新入队（state=0），并将 retry_count +1
-// enqueueAt 用于控制重试任务在队列中的位置：
-// - enqueueAt 越早，越靠前（ClaimPendingGlobal 按 enqueue_at ASC 抢占）
-func (d *taskDao) RequeueForRetryGlobal(ctx context.Context, id int64, enqueueAt time.Time) error {
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`UPDATE %s SET state=0, retry_count=retry_count+1, enqueue_at=?, updated_at=NOW() WHERE id=? AND state=3 AND deleted_at IS NULL`, public.TableNameTask),
-		enqueueAt, id,
-	)
-	return err
-}
-
-// ListFailedExhaustedGlobal 获取失败(state=3)且超过重试次数的任务，用于硬删除
-func (d *taskDao) ListFailedExhaustedGlobal(ctx context.Context, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 200
-	}
-	r, err := gfdb.DB(ctx).GetAll(ctx,
-		fmt.Sprintf(`
-SELECT t.*
-  FROM %s t
-  JOIN %s m
-    ON t.tenant_id = m.tenant_id
-   AND t.model_name = m.model_name
- WHERE t.deleted_at IS NULL
-   AND t.state = 3
-   AND t.retry_count >= m.retry_times
- ORDER BY t.updated_at ASC
- LIMIT ?`, public.TableNameTask, public.TableNameModel),
-		limit,
-	)
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}
-
-// HardDeleteByIDGlobal 硬删除任务记录
-func (d *taskDao) HardDeleteByIDGlobal(ctx context.Context, id int64) error {
-	_, err := gfdb.DB(ctx).Exec(ctx,
-		fmt.Sprintf(`DELETE FROM %s WHERE id=?`, public.TableNameTask),
-		id,
-	)
-	return err
-}
-
-// ListTimeoutTasksGlobal 根据模型配置 expected_seconds 判定超时任务：
-// - state in (0,1)
-// - 模型 expected_seconds > 0
-// - now - created_at >= expected_seconds
-func (d *taskDao) ListTimeoutTasksGlobal(ctx context.Context, limit int) (list []*entity.AsynchTask, err error) {
-	if limit <= 0 {
-		limit = 200
-	}
-	r, err := gfdb.DB(ctx).GetAll(ctx,
-		fmt.Sprintf(`
-SELECT t.*
-  FROM %s t
-  JOIN %s m
-    ON t.tenant_id = m.tenant_id
-   AND t.model_name = m.model_name
- WHERE t.deleted_at IS NULL
-   AND t.state IN (0,1)
-   AND m.expected_seconds > 0
-   AND t.created_at < (NOW() - (m.expected_seconds || ' seconds')::interval)
- LIMIT ?`, public.TableNameTask, public.TableNameModel),
-		limit,
-	)
-	if err != nil {
-		return nil, err
-	}
-	err = r.Structs(&list)
-	return
-}

go.mod

@@ -1,22 +1,14 @@
 module model-gateway
 
-go 1.26.0
+go 1.26.1
 
 require (
-	gitea.com/red-future/common v0.0.19
+	gitea.redpowerfuture.com/red-future/common v0.0.33
-	github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.0
+	github.com/bjang03/gmq v0.0.3
-	github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.0
+	github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.2
-	github.com/gogf/gf/v2 v2.10.0
+	github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.2
-	github.com/google/uuid v1.6.0
+	github.com/gogf/gf/v2 v2.10.2
-	github.com/tidwall/gjson v1.14.2
+	golang.org/x/sync v0.19.0
-)
-
-require (
-	github.com/r3labs/diff/v2 v2.15.1 // indirect
-	github.com/tidwall/match v1.1.1 // indirect
-	github.com/tidwall/pretty v1.2.0 // indirect
-	github.com/vmihailenco/msgpack v4.0.4+incompatible // indirect
-	google.golang.org/appengine v1.6.7 // indirect
 )
 
 require (
@@ -33,9 +25,13 @@ require (
 	github.com/emirpasic/gods/v2 v2.0.0-alpha // indirect
 	github.com/fatih/color v1.18.0 // indirect
 	github.com/fsnotify/fsnotify v1.9.0 // indirect
+	github.com/gabriel-vasile/mimetype v1.4.13 // indirect
 	github.com/go-ego/gse v1.0.2 // indirect
 	github.com/go-logr/logr v1.4.3 // indirect
 	github.com/go-logr/stdr v1.2.2 // indirect
+	github.com/go-playground/locales v0.14.1 // indirect
+	github.com/go-playground/universal-translator v0.18.1 // indirect
+	github.com/go-playground/validator/v10 v10.30.1 // indirect
 	github.com/gogf/gf/contrib/registry/consul/v2 v2.9.5 // indirect
 	github.com/gogf/gf/contrib/trace/otlphttp/v2 v2.9.5 // indirect
 	github.com/gogo/protobuf v1.3.2 // indirect
@@ -45,6 +41,7 @@ require (
 	github.com/golang/protobuf v1.5.4 // indirect
 	github.com/golang/snappy v1.0.0 // indirect
 	github.com/google/flatbuffers v1.12.1 // indirect
+	github.com/google/uuid v1.6.0 // indirect
 	github.com/gorilla/websocket v1.5.4-0.20250319132907-e064f32e3674 // indirect
 	github.com/grokify/html-strip-tags-go v0.1.0 // indirect
 	github.com/grpc-ecosystem/grpc-gateway/v2 v2.27.2 // indirect
@@ -57,7 +54,8 @@ require (
 	github.com/hashicorp/go-rootcerts v1.0.2 // indirect
 	github.com/hashicorp/golang-lru v1.0.2 // indirect
 	github.com/hashicorp/serf v0.10.1 // indirect
-	github.com/klauspost/compress v1.18.2 // indirect
+	github.com/klauspost/compress v1.18.4 // indirect
+	github.com/leodido/go-urn v1.4.0 // indirect
 	github.com/lib/pq v1.10.9 // indirect
 	github.com/magiconair/properties v1.8.10 // indirect
 	github.com/mattn/go-colorable v0.1.14 // indirect
@@ -65,15 +63,25 @@ require (
 	github.com/mattn/go-runewidth v0.0.16 // indirect
 	github.com/mitchellh/go-homedir v1.1.0 // indirect
 	github.com/mitchellh/mapstructure v1.5.0 // indirect
+	github.com/nats-io/nats.go v1.49.0 // indirect
+	github.com/nats-io/nkeys v0.4.15 // indirect
+	github.com/nats-io/nuid v1.0.1 // indirect
 	github.com/olekukonko/errors v1.1.0 // indirect
 	github.com/olekukonko/ll v0.0.9 // indirect
 	github.com/olekukonko/tablewriter v1.1.0 // indirect
 	github.com/pkg/errors v0.9.1 // indirect
-	github.com/redis/go-redis/v9 v9.12.1 // indirect
+	github.com/r3labs/diff/v2 v2.15.1 // indirect
+	github.com/rabbitmq/amqp091-go v1.10.0 // indirect
+	github.com/redis/go-redis/v9 v9.18.0 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
-	github.com/tidwall/sjson v1.2.5
+	github.com/spf13/cast v1.10.0 // indirect
+	github.com/tidwall/gjson v1.18.0 // indirect
+	github.com/tidwall/match v1.1.1 // indirect
+	github.com/tidwall/pretty v1.2.1 // indirect
+	github.com/tidwall/sjson v1.2.5 // indirect
 	github.com/tiger1103/gfast-token v1.0.10 // indirect
 	github.com/vcaesar/cedar v0.30.0 // indirect
+	github.com/vmihailenco/msgpack v4.0.4+incompatible // indirect
 	go.mongodb.org/mongo-driver/v2 v2.4.0 // indirect
 	go.opencensus.io v0.23.0 // indirect
 	go.opentelemetry.io/auto/sdk v1.1.0 // indirect
@@ -84,10 +92,13 @@ require (
 	go.opentelemetry.io/otel/sdk v1.38.0 // indirect
 	go.opentelemetry.io/otel/trace v1.38.0 // indirect
 	go.opentelemetry.io/proto/otlp v1.7.1 // indirect
+	go.uber.org/atomic v1.11.0 // indirect
+	golang.org/x/crypto v0.48.0 // indirect
 	golang.org/x/exp v0.0.0-20250305212735-054e65f0b394 // indirect
-	golang.org/x/net v0.48.0 // indirect
+	golang.org/x/net v0.49.0 // indirect
-	golang.org/x/sys v0.39.0 // indirect
+	golang.org/x/sys v0.41.0 // indirect
-	golang.org/x/text v0.32.0 // indirect
+	golang.org/x/text v0.34.0 // indirect
+	google.golang.org/appengine v1.6.7 // indirect
 	google.golang.org/genproto/googleapis/api v0.0.0-20250825161204-c5933d9347a5 // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20250825161204-c5933d9347a5 // indirect
 	google.golang.org/grpc v1.75.0 // indirect

go.sum

@@ -1,6 +1,6 @@
 cloud.google.com/go v0.26.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
-gitea.com/red-future/common v0.0.19 h1:9/WrfCFUCeFUYwuhBYF+JOQi5F5xuOy+gVnf2ZvHZu4=
+gitea.redpowerfuture.com/red-future/common v0.0.33 h1:AhWJ6l9zrjc1U0UEfyIZu8wkkVFxNe0hfuA51vOnOIo=
-gitea.com/red-future/common v0.0.19/go.mod h1:6/nqIucVzmjOyqDTIq71feYBXXFNBy0rFwzaQ0/Ueoo=
+gitea.redpowerfuture.com/red-future/common v0.0.33/go.mod h1:FtI9KJJSKo4/K0emjVkbL8yoSIPHJdZXr27vnScQpmM=
 github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03qcyfWMU=
 github.com/BurntSushi/toml v1.5.0 h1:W5quZX/G/csjUnuI8SUYlsHs9M38FC7znL0lIO+DvMg=
 github.com/BurntSushi/toml v1.5.0/go.mod h1:ukJfTF/6rtPPRCnwkur4qwRxa8vTRFBF0uk2lLoLwho=
@@ -19,6 +19,8 @@ github.com/beorn7/perks v0.0.0-20180321164747-3a771d992973/go.mod h1:Dwedo/Wpr24
 github.com/beorn7/perks v1.0.0/go.mod h1:KWe93zE9D1o94FZ5RNwFwVgaQK1VOXiVxmqh+CedLV8=
 github.com/beorn7/perks v1.0.1/go.mod h1:G2ZrVWU2WbWT9wwq4/hrbKbnv/1ERSJQ0ibhJ6rlkpw=
 github.com/bgentry/speakeasy v0.1.0/go.mod h1:+zsyZBPWlz7T6j88CTgSN5bM796AkVf0kBD4zp0CCIs=
+github.com/bjang03/gmq v0.0.3 h1:Yn9GZP1okOc8uh0f/1FFTooV5/mbO4pKrkcK9mTMjok=
+github.com/bjang03/gmq v0.0.3/go.mod h1:Y7TwWGuV4Cw97WUDaM7x+NC4kyFx1z44WAvNwJV3HV8=
 github.com/bsm/ginkgo/v2 v2.12.0 h1:Ny8MWAHyOepLGlLKYmXG4IEkioBysk6GpaRTLC8zwWs=
 github.com/bsm/ginkgo/v2 v2.12.0/go.mod h1:SwYbGRRDovPVboqFv0tPTcG1sN61LM1Z4ARdbAV9g4c=
 github.com/bsm/gomega v1.27.10 h1:yeMWxP2pV2fG3FgAODIY8EiRE3dy0aeFYt4l7wh6yKA=
@@ -63,8 +65,12 @@ github.com/fatih/color v1.9.0/go.mod h1:eQcE1qtQxscV5RaZvpXrrb8Drkc3/DdQ+uUYCNjL
 github.com/fatih/color v1.13.0/go.mod h1:kLAiJbzzSOZDVNGyDpeOxJ47H46qBXwg5ILebYFFOfk=
 github.com/fatih/color v1.18.0 h1:S8gINlzdQ840/4pfAwic/ZE0djQEH3wM94VfqLTZcOM=
 github.com/fatih/color v1.18.0/go.mod h1:4FelSpRwEGDpQ12mAdzqdOukCy4u8WUtOY6lkT/6HfU=
+github.com/frankban/quicktest v1.14.6 h1:7Xjx+VpznH+oBnejlPUj8oUpdxnVs4f8XU8WnHkI4W8=
+github.com/frankban/quicktest v1.14.6/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
 github.com/fsnotify/fsnotify v1.9.0 h1:2Ml+OJNzbYCTzsxtv8vKSFD9PbJjmhYF14k/jKC7S9k=
 github.com/fsnotify/fsnotify v1.9.0/go.mod h1:8jBTzvmWwFyi3Pb8djgCCO5IBqzKJ/Jwo8TRcHyHii0=
+github.com/gabriel-vasile/mimetype v1.4.13 h1:46nXokslUBsAJE/wMsp5gtO500a4F3Nkz9Ufpk2AcUM=
+github.com/gabriel-vasile/mimetype v1.4.13/go.mod h1:d+9Oxyo1wTzWdyVUPMmXFvp4F9tea18J8ufA774AB3s=
 github.com/go-ego/gse v1.0.2 h1:+27lYFPhQEhA9igtdOsJPRKYL/k3TwYsxBF5jr6KFv4=
 github.com/go-ego/gse v1.0.2/go.mod h1:Fy35G+q7VV7Et1zIKO8o/sW1kkugV3znXap/lF/11zc=
 github.com/go-kit/kit v0.8.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
@@ -76,17 +82,25 @@ github.com/go-logr/logr v1.4.3 h1:CjnDlHq8ikf6E492q6eKboGOC0T8CDaOvkHCIg8idEI=
 github.com/go-logr/logr v1.4.3/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
 github.com/go-logr/stdr v1.2.2 h1:hSWxHoqTgW2S2qGc0LTAI563KZ5YKYRhT3MFKZMbjag=
 github.com/go-logr/stdr v1.2.2/go.mod h1:mMo/vtBO5dYbehREoey6XUKy/eSumjCCveDpRre4VKE=
+github.com/go-playground/assert/v2 v2.2.0 h1:JvknZsQTYeFEAhQwI4qEt9cyV5ONwRHC+lYKSsYSR8s=
+github.com/go-playground/assert/v2 v2.2.0/go.mod h1:VDjEfimB/XKnb+ZQfWdccd7VUvScMdVu0Titje2rxJ4=
+github.com/go-playground/locales v0.14.1 h1:EWaQ/wswjilfKLTECiXz7Rh+3BjFhfDFKv/oXslEjJA=
+github.com/go-playground/locales v0.14.1/go.mod h1:hxrqLVvrK65+Rwrd5Fc6F2O76J/NuW9t0sjnWqG1slY=
+github.com/go-playground/universal-translator v0.18.1 h1:Bcnm0ZwsGyWbCzImXv+pAJnYK9S473LQFuzCbDbfSFY=
+github.com/go-playground/universal-translator v0.18.1/go.mod h1:xekY+UJKNuX9WP91TpwSH2VMlDf28Uj24BCp08ZFTUY=
+github.com/go-playground/validator/v10 v10.30.1 h1:f3zDSN/zOma+w6+1Wswgd9fLkdwy06ntQJp0BBvFG0w=
+github.com/go-playground/validator/v10 v10.30.1/go.mod h1:oSuBIQzuJxL//3MelwSLD5hc2Tu889bF0Idm9Dg26cM=
 github.com/go-stack/stack v1.8.0/go.mod h1:v0f6uXyyMGvRgIKkXu+yp6POWl0qKG85gN/melR3HDY=
-github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.0 h1:39+jbTenm7KBj4hO2C8ANAxVHpX/7OuRDs1VcGC9ylA=
+github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.2 h1:u8EpP24GkprogROnJ7htMov9Fc66pTP1eVYrWxiCYOs=
-github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.0/go.mod h1:B0s0fVzn0W220E8UTpSGzrrGKsop5KcB90twBeLCiz0=
+github.com/gogf/gf/contrib/drivers/pgsql/v2 v2.10.2/go.mod h1:GmvM3r8GVByVMi4RD2+MCs5+CfxVXPMeT8mVDkAaAXE=
-github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.0 h1:N/F9CuDdUZLoM1nVRqrDE/33pDZuhVxpNY4wYdeIaBs=
+github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.2 h1:iTQegT+lEg/wDKvj2mi3W1wrdrwFarjokf88EXVVgu4=
-github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.0/go.mod h1:x6uoJGfZOtirIRQls8xUlYzC6f7T/eULPUa9er368X0=
+github.com/gogf/gf/contrib/nosql/redis/v2 v2.10.2/go.mod h1:ZRw3GNz5cq4uYrW4TPSVyrYWaoqzujKdWro/AOcGBaE=
 github.com/gogf/gf/contrib/registry/consul/v2 v2.9.5 h1:eUqwJ/qNH8lJ6yssiqskazgp1ACQuNU6zXlLOZVuXTQ=
 github.com/gogf/gf/contrib/registry/consul/v2 v2.9.5/go.mod h1:sjQyMry9+0POYZCA6lHXBxO77WoNKkruJpRB4xKqk5k=
 github.com/gogf/gf/contrib/trace/otlphttp/v2 v2.9.5 h1:tHUEZYB5GTqEYYVDYnlGobf1xISARKDE4KHVlgjwTec=
 github.com/gogf/gf/contrib/trace/otlphttp/v2 v2.9.5/go.mod h1:cfzTn2HS9RDX8f5pUVkbGxUWcSosouqfNQ1G6cY0V88=
-github.com/gogf/gf/v2 v2.10.0 h1:rzDROlyqGMe/eM6dCalSR8dZOuMIdLhmxKSH1DGhbFs=
+github.com/gogf/gf/v2 v2.10.2 h1:46IO0Uc8e85/FqdftJFskfDejJLBL0JBnGS5qOftUu8=
-github.com/gogf/gf/v2 v2.10.0/go.mod h1:Svl1N+E8G/QshU2DUbh/3J/AJauqCgUnxHurXWR4Qx0=
+github.com/gogf/gf/v2 v2.10.2/go.mod h1:Svl1N+E8G/QshU2DUbh/3J/AJauqCgUnxHurXWR4Qx0=
 github.com/gogo/protobuf v1.1.1/go.mod h1:r8qH/GZQm5c6nD/R0oafs1akxWv10x8SbQlK7atdtwQ=
 github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
 github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
@@ -185,8 +199,10 @@ github.com/json-iterator/go v1.1.9/go.mod h1:KdQUCv79m/52Kvf8AW2vK1V8akMuk1QjK/u
 github.com/julienschmidt/httprouter v1.2.0/go.mod h1:SYymIcj16QtmaHHD7aYtjjsJG7VTCxuUUipMqKk8s4w=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
-github.com/klauspost/compress v1.18.2 h1:iiPHWW0YrcFgpBYhsA6D1+fqHssJscY/Tm/y2Uqnapk=
+github.com/klauspost/compress v1.18.4 h1:RPhnKRAQ4Fh8zU2FY/6ZFDwTVTxgJ/EMydqSTzE9a2c=
-github.com/klauspost/compress v1.18.2/go.mod h1:R0h/fSBs8DE4ENlcrlib3PsXS61voFxhIs2DeRhCvJ4=
+github.com/klauspost/compress v1.18.4/go.mod h1:R0h/fSBs8DE4ENlcrlib3PsXS61voFxhIs2DeRhCvJ4=
+github.com/klauspost/cpuid/v2 v2.2.11 h1:0OwqZRYI2rFrjS4kvkDnqJkKHdHaRnCm68/DY4OxRzU=
+github.com/klauspost/cpuid/v2 v2.2.11/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
 github.com/konsorten/go-windows-terminal-sequences v1.0.1/go.mod h1:T0+1ngSBFLxvqU3pZ+m/2kptfBszLMUkC4ZK/EgS/cQ=
 github.com/kr/logfmt v0.0.0-20140226030751-b84e30acd515/go.mod h1:+0opPa2QZZtGFBFZlji/RkVcI2GknAs/DXo4wKdlNEc=
 github.com/kr/pretty v0.1.0/go.mod h1:dAy3ld7l9f0ibDNOQOHHMYYIIbhfbHSm3C4ZsoJORNo=
@@ -196,6 +212,8 @@ github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
 github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
+github.com/leodido/go-urn v1.4.0 h1:WT9HwE9SGECu3lg4d/dIA+jxlljEa1/ffXKmRjqdmIQ=
+github.com/leodido/go-urn v1.4.0/go.mod h1:bvxc+MVxLKB4z00jd1z+Dvzr47oO32F/QSNjSBOlFxI=
 github.com/lib/pq v1.10.9 h1:YXG7RB+JIjhP29X+OtkiDnYaXQwpS4JEWq7dtCCRUEw=
 github.com/lib/pq v1.10.9/go.mod h1:AlVN5x4E4T544tWzH6hKfbfQvm3HdbOxrmggDNAPY9o=
 github.com/magiconair/properties v1.8.10 h1:s31yESBquKXCV9a/ScB3ESkOjUYYv+X0rg8SYxI99mE=
@@ -232,6 +250,12 @@ github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJ
 github.com/modern-go/reflect2 v0.0.0-20180701023420-4b7aa43c6742/go.mod h1:bx2lNnkwVCuqBIxFjflWJWanXIb3RllmbCylyMrvgv0=
 github.com/modern-go/reflect2 v1.0.1/go.mod h1:bx2lNnkwVCuqBIxFjflWJWanXIb3RllmbCylyMrvgv0=
 github.com/mwitkow/go-conntrack v0.0.0-20161129095857-cc309e4a2223/go.mod h1:qRWi+5nqEBWmkhHvq77mSJWrCKwh8bxhgT7d/eI7P4U=
+github.com/nats-io/nats.go v1.49.0 h1:yh/WvY59gXqYpgl33ZI+XoVPKyut/IcEaqtsiuTJpoE=
+github.com/nats-io/nats.go v1.49.0/go.mod h1:fDCn3mN5cY8HooHwE2ukiLb4p4G4ImmzvXyJt+tGwdw=
+github.com/nats-io/nkeys v0.4.15 h1:JACV5jRVO9V856KOapQ7x+EY8Jo3qw1vJt/9Jpwzkk4=
+github.com/nats-io/nkeys v0.4.15/go.mod h1:CpMchTXC9fxA5zrMo4KpySxNjiDVvr8ANOSZdiNfUrs=
+github.com/nats-io/nuid v1.0.1 h1:5iA8DT8V7q8WK2EScv2padNa/rTESc1KdnPw4TC2paw=
+github.com/nats-io/nuid v1.0.1/go.mod h1:19wcPz3Ph3q0Jbyiqsd0kePYG7A95tJPxeL+1OSON2c=
 github.com/olekukonko/errors v1.1.0 h1:RNuGIh15QdDenh+hNvKrJkmxxjV4hcS50Db478Ou5sM=
 github.com/olekukonko/errors v1.1.0/go.mod h1:ppzxA5jBKcO1vIpCXQ9ZqgDh8iwODz6OXIGKU8r5m4Y=
 github.com/olekukonko/ll v0.0.9 h1:Y+1YqDfVkqMWuEQMclsF9HUR5+a82+dxJuL1HHSRpxI=
@@ -264,8 +288,10 @@ github.com/prometheus/procfs v0.0.2/go.mod h1:TjEm7ze935MbeOT/UhFTIMYKhuLP4wbCsT
 github.com/prometheus/procfs v0.0.8/go.mod h1:7Qr8sr6344vo1JqZ6HhLceV9o3AJ1Ff+GxbHq6oeK9A=
 github.com/r3labs/diff/v2 v2.15.1 h1:EOrVqPUzi+njlumoqJwiS/TgGgmZo83619FNDB9xQUg=
 github.com/r3labs/diff/v2 v2.15.1/go.mod h1:I8noH9Fc2fjSaMxqF3G2lhDdC0b+JXCfyx85tWFM9kc=
-github.com/redis/go-redis/v9 v9.12.1 h1:k5iquqv27aBtnTm2tIkROUDp8JBXhXZIVu1InSgvovg=
+github.com/rabbitmq/amqp091-go v1.10.0 h1:STpn5XsHlHGcecLmMFCtg7mqq0RnD+zFr4uzukfVhBw=
-github.com/redis/go-redis/v9 v9.12.1/go.mod h1:huWgSWd8mW6+m0VPhJjSSQ+d6Nh1VICQ6Q5lHuCH/Iw=
+github.com/rabbitmq/amqp091-go v1.10.0/go.mod h1:Hy4jKW5kQART1u+JkDTF9YYOQUHXqMuhrgxOEeS7G4o=
+github.com/redis/go-redis/v9 v9.18.0 h1:pMkxYPkEbMPwRdenAzUNyFNrDgHx9U+DrBabWNfSRQs=
+github.com/redis/go-redis/v9 v9.18.0/go.mod h1:k3ufPphLU5YXwNTUcCRXGxUoF1fqxnhFQmscfkCoDA0=
 github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
 github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
 github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
@@ -276,6 +302,8 @@ github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529 h1:nn5Wsu0esKSJiIVhscUt
 github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529/go.mod h1:DxrIzT+xaE7yg65j358z/aeFdxmN0P9QXhEzd20vsDc=
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
+github.com/spf13/cast v1.10.0 h1:h2x0u2shc1QuLHfxi+cTJvs30+ZAHOGRic8uyGTDWxY=
+github.com/spf13/cast v1.10.0/go.mod h1:jNfB8QC9IA6ZuY2ZjDp0KtFO2LZZlg4S/7bzP6qqeHo=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/objx v0.1.1/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/objx v0.5.0 h1:1zr/of2m5FGMsad5YfcqgdqdWrIhu+EBEJRhR1U7z/c=
@@ -288,12 +316,14 @@ github.com/stretchr/testify v1.6.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/
 github.com/stretchr/testify v1.7.2/go.mod h1:R6va5+xMeoiuVRoj+gSkQ7d3FALtqAAGI1FQKckRals=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
-github.com/tidwall/gjson v1.14.2 h1:6BBkirS0rAHjumnjHF6qgy5d2YAJ1TLIaFE2lzfOLqo=
 github.com/tidwall/gjson v1.14.2/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
+github.com/tidwall/gjson v1.18.0 h1:FIDeeyB800efLX89e5a8Y0BNH+LOngJyGrIWxG2FKQY=
+github.com/tidwall/gjson v1.18.0/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
 github.com/tidwall/match v1.1.1 h1:+Ho715JplO36QYgwN9PGYNhgZvoUSc9X2c80KVTi+GA=
 github.com/tidwall/match v1.1.1/go.mod h1:eRSPERbgtNPcGhD8UCthc6PmLEQXEWd3PRB5JTxsfmM=
-github.com/tidwall/pretty v1.2.0 h1:RWIZEg2iJ8/g6fDDYzMpobmaoGh5OLl4AXtGUGPcqCs=
 github.com/tidwall/pretty v1.2.0/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
+github.com/tidwall/pretty v1.2.1 h1:qjsOFOWWQl+N3RsoF5/ssm1pHmJJwhjlSbZ51I6wMl4=
+github.com/tidwall/pretty v1.2.1/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
 github.com/tidwall/sjson v1.2.5 h1:kLy8mja+1c9jlljvWTlSazM7cKDRfJuR/bOJhcY5NcY=
 github.com/tidwall/sjson v1.2.5/go.mod h1:Fvgq9kS/6ociJEDnK0Fk1cpYF4FIW6ZF7LAe+6jwd28=
 github.com/tiger1103/gfast-token v1.0.10 h1:fNiBE/Dq5iTHvTGlCx3DmXa2o4hr0NtumFpffZ39k6s=
@@ -307,6 +337,8 @@ github.com/vmihailenco/msgpack v4.0.4+incompatible h1:dSLoQfGFAo3F6OoNhwUmLwVgaU
 github.com/vmihailenco/msgpack v4.0.4+incompatible/go.mod h1:fy3FlTQTDXWkZ7Bh6AcGMlsjHatGryHQYUTf1ShIgkk=
 github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
+github.com/zeebo/xxh3 v1.0.2 h1:xZmwmqxHZA8AI603jOQ0tMqmBr9lPeFwGg6d+xy9DC0=
+github.com/zeebo/xxh3 v1.0.2/go.mod h1:5NWz9Sef7zIDm2JHfFlcQvNekmcEl9ekUZQQKCYaDcA=
 go.mongodb.org/mongo-driver/v2 v2.4.0 h1:Oq6BmUAAFTzMeh6AonuDlgZMuAuEiUxoAD1koK5MuFo=
 go.mongodb.org/mongo-driver/v2 v2.4.0/go.mod h1:jHeEDJHJq7tm6ZF45Issun9dbogjfnPySb1vXA7EeAI=
 go.opencensus.io v0.23.0 h1:gqCw0LfLxScz8irSi8exQc7fyQ0fKQU/qnC/X8+V/1M=
@@ -329,6 +361,8 @@ go.opentelemetry.io/otel/trace v1.38.0 h1:Fxk5bKrDZJUH+AMyyIXGcFAPah0oRcT+LuNtJr
 go.opentelemetry.io/otel/trace v1.38.0/go.mod h1:j1P9ivuFsTceSWe1oY+EeW3sc+Pp42sO++GHkg4wwhs=
 go.opentelemetry.io/proto/otlp v1.7.1 h1:gTOMpGDb0WTBOP8JaO72iL3auEZhVmAQg4ipjOVAtj4=
 go.opentelemetry.io/proto/otlp v1.7.1/go.mod h1:b2rVh6rfI/s2pHWNlB7ILJcRALpcNDzKhACevjI+ZnE=
+go.uber.org/atomic v1.11.0 h1:ZvwS0R+56ePWxUNi+Atn9dWONBPp/AUETXlHW0DxSjE=
+go.uber.org/atomic v1.11.0/go.mod h1:LUxbIzbOniOlMKjJjyPfpl4v+PKK2cNJn91OQbhoJI0=
 go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
 go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
 golang.org/x/crypto v0.0.0-20180904163835-0709b304e793/go.mod h1:6SG95UA2DQfeDnfUPMdvaQW0Q7yPrPDi9nlGo2tz2b4=
@@ -336,6 +370,8 @@ golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACk
 golang.org/x/crypto v0.0.0-20190923035154-9ee001bba392/go.mod h1:/lpIB1dKB+9EgE3H3cr1v9wB50oz8l4C4h62xy7jSTY=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
+golang.org/x/crypto v0.48.0 h1:/VRzVqiRSggnhY7gNRxPauEQ5Drw9haKdM0jqfcCFts=
+golang.org/x/crypto v0.48.0/go.mod h1:r0kV5h3qnFPlQnBSrULhlsRfryS2pmewsg+XfMgkVos=
 golang.org/x/exp v0.0.0-20190121172915-509febef88a4/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
 golang.org/x/exp v0.0.0-20250305212735-054e65f0b394 h1:nDVHiLt8aIbd/VzvPWN6kSOPE7+F/fNFDSXLVYkE/Iw=
 golang.org/x/exp v0.0.0-20250305212735-054e65f0b394/go.mod h1:sIifuuw/Yco/y6yb6+bDNfyeQ/MdPUy/hKEMYQV17cM=
@@ -344,8 +380,8 @@ golang.org/x/lint v0.0.0-20190227174305-5b3e6a55c961/go.mod h1:wehouNa3lNwaWXcvx
 golang.org/x/lint v0.0.0-20190313153728-d0100b6bd8b3/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
-golang.org/x/mod v0.30.0 h1:fDEXFVZ/fmCKProc/yAXXUijritrDzahmwwefnjoPFk=
+golang.org/x/mod v0.32.0 h1:9F4d3PHLljb6x//jOyokMv3eX+YDeepZSEo3mFJy93c=
-golang.org/x/mod v0.30.0/go.mod h1:lAsf5O2EvJeSFMiBxXDki7sCgAxEUcZHXoXMKT4GJKc=
+golang.org/x/mod v0.32.0/go.mod h1:SgipZ/3h2Ci89DlEtEXWUk/HteuRin+HHhN+WbNhguU=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20180826012351-8a410e7b638d/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20181114220301-adae6a3d119a/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -361,8 +397,8 @@ golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwY
 golang.org/x/net v0.0.0-20201110031124-69a78807bb2b/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
 golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
 golang.org/x/net v0.0.0-20210410081132-afb366fc7cd1/go.mod h1:9tjilg8BloeKEkVJvy7fQ90B1CfIiPueXVOjqfkSzI8=
-golang.org/x/net v0.48.0 h1:zyQRTTrjc33Lhh0fBgT/H3oZq9WuvRR5gPC70xpDiQU=
+golang.org/x/net v0.49.0 h1:eeHFmOGUTtaaPSGNmjBKpbng9MulQsJURQUAfUwY++o=
-golang.org/x/net v0.48.0/go.mod h1:+ndRgGjkh8FGtu1w1FGbEC31if4VrNVMuKTgcAAnQRY=
+golang.org/x/net v0.49.0/go.mod h1:/ysNB2EvaqvesRkuLAyjI1ycPZlQHM3q01F02UY/MV8=
 golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
 golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -397,15 +433,15 @@ golang.org/x/sys v0.0.0-20220503163025-988cb79eb6c6/go.mod h1:oPkhp1MJrh7nUepCBc
 golang.org/x/sys v0.0.0-20220728004956-3c1f35247d10/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20221010170243-090e33056c14/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.39.0 h1:CvCKL8MeisomCi6qNZ+wbb0DN9E5AATixKsvNtMoMFk=
+golang.org/x/sys v0.41.0 h1:Ivj+2Cp/ylzLiEU89QhWblYnOE9zerudt9Ftecq2C6k=
-golang.org/x/sys v0.39.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/sys v0.41.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.32.0 h1:ZD01bjUt1FQ9WJ0ClOL5vxgxOI/sVCNgX1YtKwcY0mU=
+golang.org/x/text v0.34.0 h1:oL/Qq0Kdaqxa1KbNeMKwQq0reLCCaFtqu2eNuSeNHbk=
-golang.org/x/text v0.32.0/go.mod h1:o/rUWzghvpD5TXrTIBuJU77MTaN0ljMWE47kxGJQ7jY=
+golang.org/x/text v0.34.0/go.mod h1:homfLqTYRFyVYemLBFl5GgL/DWEiH5wcsQ5gSh1yziA=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20190114222345-bf090417da8b/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20190226205152-f727befe758c/go.mod h1:9Yl7xja0Znq3iFh3HoIrodX9oNMXvdceNzlUR8zjMvY=
@@ -415,8 +451,8 @@ golang.org/x/tools v0.0.0-20190907020128-2ca718005c18/go.mod h1:b+2E5dAYhXwXZwtn
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
-golang.org/x/tools v0.39.0 h1:ik4ho21kwuQln40uelmciQPp9SipgNDdrafrYA4TmQQ=
+golang.org/x/tools v0.41.0 h1:a9b8iMweWG+S0OBnlU36rzLp20z1Rp10w+IY2czHTQc=
-golang.org/x/tools v0.39.0/go.mod h1:JnefbkDPyD8UU2kI5fuf8ZX4/yUeh9W877ZeBONxUqQ=
+golang.org/x/tools v0.41.0/go.mod h1:XSY6eDqxVNiYgezAVqqCeihT4j1U2CCsqvH3WhQpnlg=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
@@ -426,6 +462,8 @@ gonum.org/v1/gonum v0.16.0/go.mod h1:fef3am4MQ93R2HHpKnLk4/Tbh/s0+wqD5nfa6Pnwy4E
 google.golang.org/appengine v1.1.0/go.mod h1:EbEs0AVv82hx2wNQdGPgUI5lhzA/G0D9YwlJXL52JkM=
 google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
 google.golang.org/appengine v1.6.6/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
+google.golang.org/appengine v1.6.7 h1:FZR1q0exgwxzPzp/aF+VccGrSfxfPpkBqjIIEq3ru6c=
+google.golang.org/appengine v1.6.7/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
 google.golang.org/genproto v0.0.0-20180817151627-c66870c02cf8/go.mod h1:JiN7NxoALGmiZfu7CAH4rXhgtRTLTxftemlI0sWmxmc=
 google.golang.org/genproto v0.0.0-20190819201941-24fa4b261c55/go.mod h1:DMBHOl98Agz4BDEuKkezgsaosCRResVns1a3J2ZsMNc=
 google.golang.org/genproto v0.0.0-20200526211855-cb27e3aa2013/go.mod h1:NbSheEEYHJ7i3ixzK3sjbqSGDJWnxyFXZblF3eUsNvo=

main.go

@@ -2,90 +2,67 @@ package main
 
 import (
 	"context"
+	"model-gateway/consts/public"
+	"model-gateway/service"
+	"model-gateway/service/utils"
 	"os"
 	"os/signal"
 	"syscall"
 	"time"
 
 	"model-gateway/controller"
-	"model-gateway/service"
 
-	"gitea.com/red-future/common/http"
+	"gitea.redpowerfuture.com/red-future/common/http"
-	"gitea.com/red-future/common/jaeger"
+	"gitea.redpowerfuture.com/red-future/common/jaeger"
-	_ "gitea.com/red-future/common/swagger"
+	_ "gitea.redpowerfuture.com/red-future/common/swagger"
+	gmq "github.com/bjang03/gmq/core/gmq"
+	"github.com/bjang03/gmq/mq"
 	_ "github.com/gogf/gf/contrib/drivers/pgsql/v2"
 	_ "github.com/gogf/gf/contrib/nosql/redis/v2"
 	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/os/gtimer"
 )
 
 func main() {
-	ctx, cancel := context.WithCancel(context.Background())
+	ctx := context.Background()
-	defer cancel()
 	defer jaeger.ShutDown(ctx)
 
+	// 初始化全局协程池（最大 goroutine 数，从配置文件读取，默认 100）
+	workerNum := g.Cfg().MustGet(ctx, "pool.workerNum", utils.DefaultWorkerNum).Int()
+	utils.Init(workerNum)
+	g.Log().Infof(ctx, "[main] 全局协程池已初始化, workerNum=%d", workerNum)
+
 	// 注册路由
 	http.RouteRegister([]interface{}{
-		controller.Model,
+		controller.ModelCall,
-		controller.Task,
+		controller.ModelManage,
-		controller.Stat,
+		controller.ModelErrorMemory,
 	})
 
-	// 本地调试：可选自动触发 worker/cleaner（由配置文件控制）
+	gmq.GmqRegister(public.GmqMsgPluginsName, &mq.NatsConn{
-	startAutoRunner(ctx)
+		NatsConfig: mq.NatsConfig{
+			Addr:     g.Config().MustGet(ctx, "nats.addr").String(),
+			Port:     g.Config().MustGet(ctx, "nats.port").String(),
+			Username: g.Config().MustGet(ctx, "nats.username").String(),
+			Password: g.Config().MustGet(ctx, "nats.password").String(),
+		},
+	})
 
-	// 监听退出信号，确保 Ctrl+C 能完整退出（停止 worker/cleaner 并关闭 http server）
+	gtimer.AddSingleton(ctx, 10*time.Second, func(ctx context.Context) {
+		err := service.ModelTaskEndService.GetTaskStartList(ctx)
+		if err != nil {
+			g.Log().Error(ctx, "模型视频任务处理失败 err: %v", err)
+		}
+	})
+
+	// 监听退出信号，确保 Ctrl+C 能完整退出（停掉定时器与协程池，等任务执行完成再关闭）
 	quit := make(chan os.Signal, 1)
 	signal.Notify(quit, os.Interrupt, syscall.SIGTERM)
 	<-quit
 
-	g.Log().Infof(ctx, "[main] 收到退出信号，开始优雅退出...")
+	utils.Shutdown()
-	cancel()
+	g.Log().Infof(ctx, "[main] 全局协程池已关闭")
-	// 关闭 http server（RouteRegister 内部是 go Httpserver.Run() 启动的）
-	_ = http.Httpserver.Shutdown()
-}
 
-func startAutoRunner(ctx context.Context) {
+	// 收到退出信号后，关闭全局协程池，等待所有已提交的任务执行完成
-	// worker
+	g.Log().Info(ctx, "服务正在关闭...")
-	if g.Cfg().MustGet(ctx, "asynch.worker.enabled").Bool() {
-		interval := g.Cfg().MustGet(ctx, "asynch.worker.intervalSeconds").Int()
-		if interval <= 0 {
-			interval = 5
-		}
-		batchSize := g.Cfg().MustGet(ctx, "asynch.worker.batchSize").Int()
-		goroutines := g.Cfg().MustGet(ctx, "asynch.worker.goroutines").Int()
-		ticker := time.NewTicker(time.Duration(interval) * time.Second)
-		go func() {
-			defer ticker.Stop()
-			for {
-				select {
-				case <-ctx.Done():
-					return
-				case <-ticker.C:
-					if _, err := service.AsyncWorker.RunOnce(ctx, batchSize, goroutines); err != nil {
-						g.Log().Warningf(ctx, "[auto-worker] run once failed: %v", err)
-					}
-				}
-			}
-		}()
-	}
-
-	// cleaner
-	if g.Cfg().MustGet(ctx, "asynch.cleaner.enabled").Bool() {
-		interval := g.Cfg().MustGet(ctx, "asynch.cleaner.intervalSeconds").Int()
-		if interval <= 0 {
-			interval = 30
-		}
-		ticker := time.NewTicker(time.Duration(interval) * time.Second)
-		go func() {
-			defer ticker.Stop()
-			for {
-				select {
-				case <-ctx.Done():
-					return
-				case <-ticker.C:
-					service.Cleaner.RunOnce(ctx)
-				}
-			}
-		}()
-	}
 }

model/domain/schema_mapping.go

@@ -0,0 +1,52 @@
+package domain
+
+// ChatFieldsReq 对话/推理模型业务字段映射
+// 适用于 推理模型(100) 和 多模态模型(500) 的子类型
+type ChatFieldsReq struct {
+	MaxTokens              string `json:"max_tokens" dc:"模型支持的最大输出 token 数"`
+	Stream                 string `json:"stream" dc:"是否流式输出"`
+	Tools                  string `json:"tools" dc:"工具"`
+	ToolId                 string `json:"tool_id" dc:"工具 ID"`
+	ToolPrompt             string `json:"tool_prompt" dc:"工具提示词,用于描述工具的需求"`
+	UserPrompt             string `json:"user_prompt" dc:"用户提示词,用于描述用户的需求"`
+	SystemPrompt           string `json:"system_prompt" dc:"系统提示词,用于描述系统的需求"`
+	AssistantPrompt        string `json:"assistant_prompt" dc:"助手提示词,用于描述助手的需求"`
+	ReferenceImage         string `json:"reference_image" dc:"参考图片，用于生成角色形象和风格一致性参考"`
+	ReferenceVideo         string `json:"reference_video" dc:"参考视频，用于生成动作和场景一致性参考"`
+	ReferenceAudio         string `json:"reference_audio" dc:"参考音频，用于生成声音和风格一致性参考"`
+	ImgReferenceTemplate   string `json:"img_reference_template" dc:"prompt 中引用参考图片的标签格式，用 %d 作为编号占位符"`
+	VideoReferenceTemplate string `json:"video_reference_template" dc:"prompt 中引用参考视频的标签格式，用 %d 作为编号占位符"`
+	AudioReferenceTemplate string `json:"audio_reference_template" dc:"prompt 中引用参考音频的标签格式，用 %d 作为编号占位符"`
+}
+
+// ChatFieldsRes 对话/推理模型业务字段映射
+// 适用于 推理模型(100) 和 多模态模型(500) 的子类型
+type ChatFieldsRes struct {
+	Tools            string `json:"tools" dc:"工具"`
+	ReasoningContent string `json:"reasoning_content" dc:"推理内容"`
+}
+
+// VideoFields 视频模型业务字段映射
+// 适用于 视频模型(600) 及其子类型
+type VideoFields struct {
+	SystemPrompt           string `json:"system_prompt" dc:"系统提示词,用于描述系统的需求"`
+	MinDuration            string `json:"min_duration" dc:"模型支持的最小视频时长（秒）"`
+	MaxDuration            string `json:"max_duration" dc:"模型支持的最大视频时长（秒）"`
+	FirstFrame             string `json:"first_frame" dc:"视频的首帧/初始画面，传入一张图片作为视频第一帧画面"`
+	ReferenceImage         string `json:"reference_image" dc:"参考图片，用于生成角色形象和风格一致性参考"`
+	ReferenceVideo         string `json:"reference_video" dc:"参考视频，用于生成动作和场景一致性参考"`
+	MaxMediaItems          string `json:"max_media_items" dc:"模型允许传入的最大参考媒体数量"`
+	ImgReferenceTemplate   string `json:"img_reference_template" dc:"prompt 中引用参考图片的标签格式，用 %d 作为编号占位符"`
+	VideoReferenceTemplate string `json:"video_reference_template" dc:"prompt 中引用参考视频的标签格式，用 %d 作为编号占位符"`
+	AudioReferenceTemplate string `json:"audio_reference_template" dc:"prompt 中引用参考音频的标签格式，用 %d 作为编号占位符"`
+	Fps                    string `json:"fps" dc:"视频帧率"`
+	Resolution             string `json:"resolution" dc:"视频分辨率，如 1920x1080"`
+	NegativePrompt         string `json:"negative_prompt" dc:"反向提示词，描述不希望出现的内容"`
+	CfgScale               string `json:"cfg_scale" dc:"CFG 引导比例，控制对 prompt 的遵从程度"`
+}
+
+// VideoFieldsRes 视频模型业务字段映射
+// 适用于 视频模型(600) 及其子类型
+type VideoFieldsRes struct {
+	Duration int64 `json:"duration" dc:"视频时长（秒）"`
+}

model/dto/model_call_dto.go

@@ -0,0 +1,149 @@
+package dto
+
+import (
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// ModelCallReq 模型调用请求
+type ModelCallReq struct {
+	g.Meta         `path:"/modelCall" method:"post" tags:"模型管理" summary:"模型调用" dc:"模型调用"`
+	ModelId        int64          `json:"modelId" v:"required#modelId不能为空" dc:"模型ID"`
+	BizName        string         `json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
+	SessionId      string         `json:"sessionId" v:"required#sessionId不能为空" dc:"会话ID"`
+	RequestParams  map[string]any `json:"requestParams" dc:"请求参数（模板字段）"`
+	BusinessParams map[string]any `json:"businessParams" dc:"业务参数（按业务字段名传，按 RequestBusinessFieldMapping 写入请求体）"`
+	MsgTopic       string         `json:"msgTopic" dc:"消息主题（异步必要参数）"`
+}
+
+type ModelCallRes struct {
+	TaskId           int64          `json:"id" dc:"任务ID"`
+	ModelId          int64          `json:"modelId" dc:"生效模型ID（引用行=解析后的系统模型ID，计价按此）"`
+	MediaType        string         `json:"mediaType" dc:"输入媒体类型（shop词汇: text/audio/video）"`
+	TotalTokens      int64          `json:"totalTokens" dc:"总token"`
+	PromptTokens     int64          `json:"promptTokens" dc:"输入token"`
+	CompletionTokens int64          `json:"completionTokens" dc:"输出token"`
+	Tools            []ModelTool    `json:"tools" dc:"工具"`
+	ReasoningContent string         `json:"reasoningContent" dc:"思考内容"`
+	Content          map[string]any `json:"content" dc:"内容"`
+	Cost             float64        `json:"cost" dc:"费用（元）"`
+	ErrorMsg         string         `json:"errorMsg" dc:"错误消息"`
+}
+
+type ModelTool struct {
+	Id       string `json:"id" dc:"工具ID"`
+	Type     string `json:"type" dc:"工具类型"`
+	Function struct {
+		Name      string `json:"name"`
+		Arguments string `json:"arguments"`
+	} `json:"function"`
+}
+
+// ModelCallStreamEvent 流式增量事件（SSE data 行）。文本增量事件省略 Type，
+// 流末 done 事件带 Type="done" 与 Tools。字段名由本结构体 json tag 统一管理。
+type ModelCallStreamEvent struct {
+	Type             string         `json:"type,omitempty"`
+	Content          map[string]any `json:"content,omitempty"`
+	ReasoningContent string         `json:"reasoningContent,omitempty"`
+	TotalTokens      int64          `json:"totalTokens,omitempty"`
+	PromptTokens     int64          `json:"promptTokens,omitempty"`
+	CompletionTokens int64          `json:"completionTokens,omitempty"`
+	Tools            []ModelTool    `json:"tools,omitempty"`
+	Cost             float64        `json:"cost,omitempty" dc:"费用（元），done事件携带最终费用，增量事件不带"`
+}
+
+// ModelCallStreamReq 模型调用流式请求
+type ModelCallStreamReq struct {
+	g.Meta         `path:"/modelCallStream" method:"post" tags:"模型管理" summary:"模型调用流式" dc:"模型调用流式"`
+	ModelId        int64          `json:"modelId" v:"required#modelId不能为空" dc:"模型ID"`
+	BizName        string         `json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
+	SessionId      string         `json:"sessionId" v:"required#sessionId不能为空" dc:"会话ID"`
+	RequestParams  map[string]any `json:"requestParams" dc:"请求参数（模板字段）"`
+	BusinessParams map[string]any `json:"businessParams" dc:"业务参数（按业务字段名传，按 RequestBusinessFieldMapping 写入请求体）"`
+}
+
+type ModelMsg struct {
+	TaskID           int64          `json:"id" dc:"任务ID"`
+	ModelId          int64          `json:"modelId" dc:"生效模型ID（引用行=解析后的系统模型ID，计价按此）"`
+	MediaType        string         `json:"mediaType" dc:"输入媒体类型（shop词汇: text/audio/video）"`
+	TotalTokens      int64          `json:"totalTokens" dc:"总token"`
+	PromptTokens     int64          `json:"promptTokens" dc:"输入token"`
+	CompletionTokens int64          `json:"completionTokens" dc:"输出token"`
+	Content          map[string]any `json:"content" dc:"内容"`
+	Cost             float64        `json:"cost" dc:"费用（元）"`
+	Duration         int64          `json:"duration" dc:"时长（秒）"`
+	ErrorMsg         string         `json:"errorMsg" dc:"错误消息"`
+}
+
+//========================
+// Upload 文件上传定义
+//========================
+
+// UploadFileBytesReq 上传文件请求(字节流)
+type UploadFileBytesReq struct {
+	FileName     string `json:"fileName" dc:"文件名"`
+	FileBytes    []byte `json:"fileBytes" dc:"文件字节流"`
+	FileStoreURL string `json:"fileStoreURL" dc:"文件存储URL"`
+}
+
+type UploadFileBytesRes struct {
+	FileURL           string `json:"fileURL" dc:"上传地址"`
+	FileSize          int    `json:"fileSize" dc:"文件大小"`
+	FileName          string `json:"fileName" dc:"文件名称"`
+	FileFormat        string `json:"fileFormat" dc:"文件格式"`
+	FileAddressPrefix string `json:"fileAddressPrefix"`
+}
+
+// ===========================
+// Template 元数据模板定义
+// ===========================
+
+// UploadRule 上传文件规则
+type UploadRule struct {
+	Format   string `json:"format" dc:"格式"`
+	MaxSize  int    `json:"maxSize" dc:"最大大小"`
+	MaxCount int    `json:"maxCount" dc:"最大数量"`
+}
+
+// Constraint 字段约束
+type Constraint struct {
+	// 数字类型：int、float、double
+	NumberType string `json:"numberType" dc:"数字类型"`
+	Min        any    `json:"min" dc:"最小值"`
+	Max        any    `json:"max" dc:"最大值"`
+
+	// 字符串类型：string、text
+	MinLength int    `json:"minLength" dc:"最小长度"`
+	MaxLength int    `json:"maxLength" dc:"最大长度"`
+	Pattern   string `json:"pattern" dc:"正则"`
+
+	// 上传文件类型
+	UploadTotalMaxCount int          `json:"uploadTotalMaxCount" dc:"上传文件最大数量"`
+	UploadTotalMaxSize  int          `json:"uploadTotalMaxSize" dc:"上传文件最大大小"`
+	UploadRules         []UploadRule `json:"uploadRules" dc:"上传文件规则"`
+}
+
+// SelectOptionT 选择框选项
+type SelectOptionT struct {
+	Label string `json:"label" dc:"展示标签"`
+	Value any    `json:"value" dc:"选项值"`
+}
+
+// Template 元数据模板结构体
+// 用于描述单个字段的元数据，包括类型、值、默认值、校验规则等
+type Template struct {
+	Type            string              `json:"type" dc:"类型：string/boolean/number/object/array/null"`
+	Value           any                 `json:"value" dc:"值"`
+	DefaultValue    any                 `json:"defaultValue" dc:"默认值"`
+	Required        bool                `json:"required" dc:"是否必填"`
+	FieldType       string              `json:"fieldType" dc:"字段类型（输入框/选择框/多行文本/数字输入等）"`
+	Attrs           any                 `json:"attrs" dc:"子字段（object/array时使用）"`
+	IsContainer     bool                `json:"isContainer" dc:"是否为容器"`
+	LinkRules       []map[string]string `json:"linkRules" dc:"链接规则"`
+	Label           string              `json:"label" dc:"展示标签"`
+	Loop            bool                `json:"loop" dc:"是否为循环"`
+	IsForm          bool                `json:"isForm" dc:"是否为表单"`
+	Constraint      Constraint          `json:"constraint" dc:"字段约束"`
+	Options         []SelectOptionT     `json:"options" dc:"选项列表（字段类型为选择框时使用）"`
+	FieldConstraint string              `json:"fieldConstraint" dc:"字段额外约束"`
+	EnumValues      []any               `json:"enumValues" dc:"枚举值列表"`
+}

model/dto/model_dto.go

@@ -1,134 +0,0 @@
-package dto
-
-import (
-	"gitea.com/red-future/common/beans"
-	"github.com/gogf/gf/v2/frame/g"
-)
-
-// 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#modelName不能为空" dc:"模型名称（唯一标识）"`
-	ModelType            int    `p:"modelType" json:"modelType" v:"required#modelType不能为空" dc:"模型类型：1-文本生成 2-图像生成 3-语音 4-视频 5-多模态"`
-	BaseURL              string `p:"baseUrl" json:"baseUrl" v:"required#baseUrl不能为空" dc:"模型服务基础地址（如 http(s)://host:port）"`
-	HttpMethod           string `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST（默认POST）"`
-	HeadMsg              string `p:"headMsg" json:"headMsg" dc:"请求头绑定（支持多个，逗号分隔），示例：Authorization:Bearer xxx,Content-Type:application/json"`
-	IsPrivate            *int   `p:"isPrivate" json:"isPrivate" v:"in:0,1#私有化参数只能为0或1" dc:"是否私有化：0-私有（默认） 1-公共"`
-	Enabled              *int   `p:"enabled" json:"enabled" v:"in:0,1#启用参数只能为0或1" dc:"是否启用：0-禁用，1-启用（默认1）"`
-	IsChatModel          *int   `p:"isChatModel" json:"isChatModel" v:"in:0,1#对话模型参数只能为0或1" dc:"是否为对话模型：0-否，1-是（默认0）"`
-	IsOwner              *int   `p:"isOwner" json:"isOwner" v:"in:0,1#是否为所有者参数只能为0或1" dc:"是否为所有者：0-否，1-是（默认0）"`
-	ApiKey               string `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥，用于模型认证"`
-	Form                 any    `p:"form" json:"form" dc:"动态表单配置（JSON），用于前端渲染配置项"`
-	RequestMapping       any    `p:"requestMapping" json:"requestMapping" dc:"请求映射"`
-	ResponseMapping      any    `p:"responseMapping" json:"responseMapping" dc:"返回映射"`
-	ResponseBody         any    `p:"responseBody" json:"responseBody" dc:"返回主体"`
-	TokenMapping         string `p:"tokenMapping" json:"tokenMapping" dc:"token映射"`
-	MaxConcurrency       int    `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数（默认10）"`
-	QueueLimit           int    `p:"queueLimit" json:"queueLimit" dc:"排队队列上限（默认1000）"`
-	TimeoutSeconds       int    `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒，默认600）"`
-	ExpectedSeconds      int    `p:"expectedSeconds" json:"expectedSeconds" dc:"模型预计执行时间（秒，默认600）"`
-	RetryTimes           int    `p:"retryTimes" json:"retryTimes" dc:"失败重试次数（默认3）"`
-	RetryQueueMaxSeconds int    `p:"retryQueueMaxSeconds" json:"retryQueueMaxSeconds" dc:"失败重试最大排队时间（秒，默认600）"`
-	AutoCleanSeconds     int    `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"任务完成后自动清理时间（秒，默认86400）"`
-	Remark               string `p:"remark" json:"remark" dc:"备注说明"`
-}
-
-type CreateModelRes struct {
-	ID int64 `json:"id,string" dc:"配置ID"`
-}
-
-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（唯一标识）"`
-	ModelType            int    `p:"modelType" json:"modelType" dc:"模型类型ID列表（逗号分隔）（可选更新）"`
-	BaseURL              string `p:"baseUrl" json:"baseUrl" dc:"模型服务基础地址"`
-	HttpMethod           string `p:"httpMethod" json:"httpMethod" dc:"请求方式：GET/POST（可选更新）"`
-	HeadMsg              string `p:"headMsg" json:"headMsg" dc:"请求头绑定（可选更新）"`
-	ApiKey               string `p:"apiKey" json:"apiKey" dc:"调用凭证/密钥，用于模型认证（可选更新）"`
-	Form                 any    `p:"form" json:"form" dc:"动态表单配置（JSON）（可选更新）"`
-	RequestMapping       any    `p:"requestMapping" json:"requestMapping" dc:"请求参数映射（可选更新）"`
-	ResponseMapping      any    `p:"responseMapping" json:"responseMapping" dc:"返回参数映射（可选更新）"`
-	ResponseBody         any    `p:"responseBody" json:"responseBody" dc:"返回主体（可选更新）"`
-	TokenMapping         string `p:"tokenMapping" json:"tokenMapping" dc:"token映射（可选更新）"`
-	Enabled              *int   `p:"enabled" json:"enabled" dc:"是否启用：0-禁用，1-启用（可选更新）"`
-	IsPrivate            *int   `p:"isPrivate" json:"isPrivate" v:"in:0,1#私有化参数只能为0或1" dc:"是否私有化：0-私有（默认） 1-公共"`
-	IsChatModel          *int   `p:"isChatModel" json:"isChatModel" v:"in:0,1#对话模型参数只能为0或1" dc:"是否为对话模型：0-否，1-是（默认0）"`
-	IsOwner              *int   `p:"isOwner" json:"isOwner" v:"in:0,1#是否为所有者参数只能为0或1" dc:"是否为所有者：0-否，1-是（默认0）"`
-	MaxConcurrency       int    `p:"maxConcurrency" json:"maxConcurrency" dc:"最大并发数（可选更新）"`
-	QueueLimit           int    `p:"queueLimit" json:"queueLimit" dc:"排队队列上限（可选更新）"`
-	TimeoutSeconds       int    `p:"timeoutSeconds" json:"timeoutSeconds" dc:"请求超时时间（秒）（可选更新）"`
-	ExpectedSeconds      int    `p:"expectedSeconds" json:"expectedSeconds" dc:"模型预计执行时间（秒）（可选更新）"`
-	RetryTimes           int    `p:"retryTimes" json:"retryTimes" dc:"失败重试次数（可选更新）"`
-	RetryQueueMaxSeconds int    `p:"retryQueueMaxSeconds" json:"retryQueueMaxSeconds" dc:"失败重试最大排队时间（秒）（可选更新）"`
-	AutoCleanSeconds     int    `p:"autoCleanSeconds" json:"autoCleanSeconds" dc:"自动清理间隔（秒）（可选更新）"`
-	Remark               string `p:"remark" json:"remark" dc:"备注说明（可选更新）"`
-}
-
-// DeleteModelReq 删除模型配置
-type DeleteModelReq struct {
-	g.Meta `path:"/deleteModel" method:"delete" tags:"模型管理" summary:"删除模型配置" dc:"删除指定ID的模型配置"`
-	ID     string `p:"id" json:"id,string" v:"required#id不能为空" dc:"配置ID"`
-}
-
-// GetModelReq 获取模型配置详情
-type GetModelReq struct {
-	g.Meta  `path:"/getModel" method:"get" tags:"模型管理" summary:"获取模型配置" dc:"根据模型ID获取配置详情"`
-	ID      int64  `p:"id" json:"id,string" v:"required#id不能为空" dc:"配置ID"`
-	Creator string `p:"creator" json:"creator" dc:"创建人"`
-}
-
-type GetModelRes struct {
-	Model any `json:"model" dc:"模型配置详情"`
-}
-
-// ListModelReq 配置列表
-type ListModelReq struct {
-	g.Meta    `path:"/listModel" method:"get" tags:"模型管理" summary:"模型配置列表" dc:"分页获取模型配置列表"`
-	Page      *beans.Page `json:"page"`
-	ModelName string      `p:"modelName" json:"modelName" dc:"模型名称（模糊查询，可选）"`
-	ModelType int         `p:"modelType" json:"modelType" dc:"模型类型"`
-	Enabled   *int        `p:"enabled" json:"enabled" dc:"是否启用：0-禁用，1-启用"`
-	IsPrivate *int        `p:"isPrivate" json:"isPrivate" dc:"是否私有化 0-私有 1-公共"`
-	IsOwner   *int        `p:"isOwner" json:"isOwner" dc:"是否为所有者 0-否 1-是"`
-	Creator   string      `p:"creator" json:"creator" dc:"创建人"`
-}
-
-type ListModelRes struct {
-	List  any `json:"list" dc:"列表数据"`
-	Total int `json:"total" dc:"总数"`
-}
-
-// AutoTuneReq 动态调参（由上层定时任务每小时触发一次）
-type AutoTuneReq struct {
-	g.Meta        `path:"/autoTune" method:"post" tags:"模型管理" summary:"动态调参" dc:"按 model_name 维度统计指定时间窗口内执行耗时(P90)，动态生成运行时 max_concurrency/queue_limit（不超过配置上限），写入 Redis 供 Worker/CreateTask 使用；windowSeconds 不传默认 3600"`
-	WindowSeconds int `p:"windowSeconds" json:"windowSeconds" dc:"统计窗口秒数；不传/<=0 默认 3600（1小时）"`
-}
-
-type AutoTuneRes struct {
-	List any `json:"list" dc:"调参结果列表"`
-}
-
-type ModelTypeModelItem struct {
-	ID   int64  `json:"id" dc:"模型主键ID"`
-	Name string `json:"name" dc:"模型名称"`
-	Form any    `json:"form" dc:"动态表单配置（JSON数组），用于前端渲染"`
-}
-
-// ListModelTypeReq 模型类型列表（分页）
-type ListTypeReq struct {
-	g.Meta `path:"/listType" method:"get" tags:"模型类型列表" summary:"模型类型列表" dc:"分页获取模型类型列表"`
-}
-
-type TypeItem struct {
-	Type map[int]string `json:"type" dc:"模型类型ID到名称的映射"`
-}
-
-type UpdateChatModelReq struct {
-	g.Meta `path:"/updateChatModel" method:"post" tags:"模型管理" summary:"更新聊天模型" dc:"更新指定模型的聊天模型"`
-	Id     int64 `p:"id" json:"id" v:"required#model不能为空" dc:"模型id"`
-}
-
-type GetIsChatModelReq struct {
-	g.Meta `path:"/getIsChatModel" method:"get" tags:"模型管理" summary:"获取模型是否为聊天模型" dc:"根据模型ID获取是否为聊天模型"`
-}

model/dto/model_error_memory_dto.go

@@ -0,0 +1,34 @@
+package dto
+
+import (
+	"gitea.redpowerfuture.com/red-future/common/beans"
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// GetErrorMemoryListReq 错误重试记忆列表
+type GetErrorMemoryListReq struct {
+	g.Meta      `path:"/errorMemory/list" method:"get" tags:"错误记忆" summary:"错误重试记忆列表" dc:"查看错误→可重试结论记忆"`
+	*beans.Page `json:"page"`
+}
+
+type GetErrorMemoryListRes struct {
+	List  []ErrorMemoryItem `json:"list" dc:"记忆条目"`
+	Total int64             `json:"total" dc:"总数"`
+}
+
+type ErrorMemoryItem struct {
+	Id             int64  `json:"id"`
+	MemoryKey      string `json:"memoryKey"`
+	Upstream       string `json:"upstream"`
+	ErrorCode      string `json:"errorCode"`
+	MsgFingerprint string `json:"msgFingerprint"`
+	Retryable      bool   `json:"retryable"`
+	Reason         string `json:"reason"`
+	AnalyzedBy     string `json:"analyzedBy"`
+}
+
+// DeleteErrorMemoryReq 删除错误重试记忆
+type DeleteErrorMemoryReq struct {
+	g.Meta `path:"/errorMemory/delete" method:"post" tags:"错误记忆" summary:"删除错误重试记忆" dc:"手动清理永久记忆条目"`
+	Id     int64 `json:"id" v:"required#id不能为空" dc:"记忆ID"`
+}

model/dto/model_manage_dto.go

@@ -0,0 +1,159 @@
+package dto
+
+import (
+	"model-gateway/consts/model"
+	"model-gateway/consts/public"
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// CreateModelManageReq 添加模型配置
+type CreateModelManageReq struct {
+	g.Meta                       `path:"/createModelManage" method:"post" tags:"模型管理" summary:"创建模型配置" dc:"添加新的模型配置"`
+	ModelSupplier                model.SupplierType       `json:"modelSupplier" v:"required#模型供应商不能为空" dc:"模型供应商"`
+	ModelName                    string                   `json:"modelName" v:"required#模型名称不能为空" dc:"模型名称"`
+	ModelType                    model.ModelType          `json:"modelType" v:"required#模型类型不能为空" dc:"模型类型"`
+	BaseURL                      string                   `json:"baseUrl" v:"required#模型服务地址不能为空" dc:"模型服务地址"`
+	SystemModel                  *bool                    `json:"systemModel" dc:"系统模型"`
+	RefSystemModelId             int64                    `json:"refSystemModelId" dc:"引用的系统模型ID（引用创建时填，普通创建留空）"`
+	HttpMethod                   string                   `json:"httpMethod" dc:"请求方式：GET/POST" d:"POST"`
+	ChatModel                    *bool                    `json:"chatModel" dc:"对话模型"`
+	ResponseType                 model.ResponseType       `json:"responseType" v:"required#调用模式不能为空" dc:"调用模式：0-同步 1-异步 2-流式"`
+	ApiKey                       string                   `json:"apiKey" dc:"调用凭证/密钥"`
+	Enabled                      *bool                    `json:"enabled" dc:"启用"`
+	RequestHeadMapping           map[string]string        `json:"requestHeadMapping" dc:"请求头映射"`
+	RequestBodyMapping           map[string]any           `json:"requestBodyMapping" dc:"请求体映射"`
+	RequestBusinessFieldMapping  map[string]string        `json:"requestBusinessFieldMapping" dc:"业务字段映射"`
+	ResponseMapping              map[string]any           `json:"responseMapping" dc:"返回映射"`
+	ResponseBodyMapping          map[string]string        `json:"responseBodyMapping" dc:"返回体映射"`
+	ResponseBusinessFieldMapping map[string]string        `json:"responseBusinessFieldMapping" dc:"业务字段映射"`
+	MaxConcurrency               int                      `json:"maxConcurrency" dc:"最大并发数（默认10）"`
+	TokenMapping                 *entity.TokenMapping     `json:"tokenMapping" dc:"token映射"`
+	AsyncTaskMapping             *entity.AsyncTaskMapping `json:"asyncTaskMapping" dc:"异步任务映射"`
+	TokenPredictPrice            float64                  `json:"tokenPredictPrice" dc:"模型Token预估价格"`
+	TokenPredictPriceUnit        string                   `json:"tokenPredictPriceUnit" dc:"模型Token预估价格单位"`
+	MaxTokens                    int                      `json:"maxTokens" dc:"最大token数"`
+	MinDuration                  int                      `json:"minDuration" dc:"最小时长"`
+	MaxDuration                  int                      `json:"maxDuration" dc:"最大时长"`
+	LastFrame                    string                   `json:"lastFrame" dc:"视频的尾帧图像"`
+	ErrorMessageMapping          map[string]any           `json:"errorMessageMapping" dc:"错误消息映射（schema 树，解析模型错误用）"`
+}
+
+type CreateModelManageRes struct {
+	Id int64 `json:"id,string" dc:"配置ID"`
+}
+
+type UpdateModelManageReq struct {
+	g.Meta                       `path:"/updateModelManage" 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                    `json:"id" v:"required#id不能为空" dc:"配置ID"`
+	ModelSupplier                model.SupplierType       `json:"modelSupplier" dc:"模型供应商"`
+	ModelName                    string                   `json:"modelName" dc:"模型名称"`
+	ModelType                    model.ModelType          `json:"modelType" dc:"模型类型"`
+	BaseURL                      string                   `json:"baseUrl" dc:"模型服务地址"`
+	SystemModel                  *bool                    `json:"systemModel" dc:"系统模型"`
+	RefSystemModelId             int64                    `json:"refSystemModelId" dc:"引用的系统模型ID（引用行只改个人字段，本字段无效）"`
+	HttpMethod                   string                   `json:"httpMethod" dc:"请求方式：GET/POST"`
+	ChatModel                    *bool                    `json:"chatModel" dc:"对话模型"`
+	ResponseType                 model.ResponseType       `json:"responseType" dc:"调用模式：0-同步 1-异步 2-流式"`
+	ApiKey                       string                   `json:"apiKey" dc:"调用凭证/密钥"`
+	Enabled                      *bool                    `json:"enabled" dc:"启用"`
+	RequestHeadMapping           map[string]string        `json:"requestHeadMapping" dc:"请求头映射"`
+	RequestBodyMapping           map[string]any           `json:"requestBodyMapping" dc:"请求体映射"`
+	RequestBusinessFieldMapping  map[string]string        `json:"requestBusinessFieldMapping" dc:"业务字段映射"`
+	ResponseMapping              map[string]any           `json:"responseMapping" dc:"返回映射"`
+	ResponseBodyMapping          map[string]string        `json:"responseBodyMapping" dc:"返回主体映射"`
+	ResponseBusinessFieldMapping map[string]string        `json:"responseBusinessFieldMapping" dc:"业务字段映射"`
+	MaxConcurrency               int                      `json:"maxConcurrency" dc:"最大并发数（默认10）"`
+	TokenMapping                 *entity.TokenMapping     `json:"tokenMapping" dc:"token映射"`
+	AsyncTaskMapping             *entity.AsyncTaskMapping `json:"asyncTaskMapping" dc:"异步任务映射"`
+	TokenPredictPrice            float64                  `json:"tokenPredictPrice" dc:"模型Token预估价格"`
+	TokenPredictPriceUnit        string                   `json:"tokenPredictPriceUnit" dc:"模型Token预估价格单位"`
+	MaxTokens                    int                      `json:"maxTokens" dc:"最大token数"`
+	MinDuration                  int                      `json:"minDuration" dc:"最小时长"`
+	MaxDuration                  int                      `json:"maxDuration" dc:"最大时长"`
+	LastFrame                    string                   `json:"lastFrame" dc:"视频的尾帧图像"`
+	ErrorMessageMapping          map[string]any           `json:"errorMessageMapping" dc:"错误消息映射（schema 树，解析模型错误用）"`
+}
+
+type DeleteModelManageReq struct {
+	g.Meta `path:"/deleteModelManage" method:"delete" tags:"模型管理" summary:"删除模型配置" dc:"删除指定ID的模型配置"`
+	Id     int64 `p:"id" json:"id,string" v:"required#id不能为空" dc:"配置ID"`
+}
+
+type GetModelManage struct {
+	ChatModel *bool  `json:"chatModel" dc:"对话模型"`
+	Creator   string `json:"creator" dc:"创建人"`
+	ModelName string `json:"modelName" dc:"模型名称"`
+}
+
+type GetModelManageReq struct {
+	g.Meta `path:"/getModelManage" method:"get" tags:"模型管理" summary:"获取模型配置" dc:"获取指定ID的模型配置"`
+	Id     int64 `p:"id" json:"id,string" v:"required#id不能为空" dc:"配置ID"`
+}
+
+type GetModelManageRes struct {
+	ModelManage *entity.ModelManage `json:"modelManage"`
+}
+
+type GetChatModelReq struct {
+	g.Meta `path:"/getChatModel" method:"get" tags:"模型管理" summary:"获取聊天模型" dc:"获取聊天模型"`
+}
+
+type GetChatModelRes struct {
+	ModelManage *entity.ModelManage `json:"modelManage"`
+}
+
+// ListModelManageReq 配置列表
+type ListModelManageReq struct {
+	g.Meta      `path:"/listModelManage" method:"get" tags:"模型管理" summary:"模型配置列表" dc:"分页获取模型配置列表"`
+	*beans.Page `json:"page"`
+	Id          int64           `p:"id" json:"id,string" dc:"配置ID"`
+	ModelName   string          `p:"modelName" json:"modelName" dc:"模型名称（模糊查询，可选）"`
+	ModelType   model.ModelType `p:"modelType" json:"modelType" dc:"模型类型"`
+	IsSameType  bool            `p:"isSameType" json:"isSameType" dc:"是否相同类型"`
+	Creator     string          `json:"creator" dc:"创建人"`
+}
+
+type ListModelManageRes struct {
+	List  []*entity.ModelManage `json:"list" dc:"列表数据"`
+	Total int                   `json:"total" dc:"总数"`
+}
+
+type CheckChatModelReq struct {
+	g.Meta `path:"/checkChatModel" method:"get" tags:"模型管理" summary:"检查是否为聊天模型" dc:"检查是否为聊天模型"`
+}
+
+type CheckChatModelRes struct {
+	IsChatModel bool `json:"isChatModel" dc:"是否为聊天模型"`
+}
+
+// ModelTypeReq 模型类型列表（分页）
+type ModelTypeReq struct {
+	g.Meta `path:"/modelType" method:"get" tags:"模型管理" summary:"模型类型列表" dc:"分页获取模型类型列表"`
+}
+
+type ModelTypeRes struct {
+	List []*model.TypeTree `json:"list" dc:"模型类型ID到名称的映射"`
+}
+
+type ModelSupplierReq struct {
+	g.Meta `path:"/modelSupplier" method:"get" tags:"模型管理" summary:"获取运营商列表" dc:"获取运营商列表"`
+}
+
+type ModelSupplierRes struct {
+	List []*public.Option `json:"list" dc:"运营商名称到ID的映射"`
+}
+
+// BuildSchemaMappingReq 构建 Schema 映射请求
+type BuildSchemaMappingReq struct {
+	g.Meta    `path:"/buildSchemaMapping" method:"post" tags:"模型管理" summary:"自动构建 Schema 映射" dc:"根据模型类型和 Schema JSON，自动生成业务字段映射"`
+	ModelType int            `json:"modelType" v:"required#模型类型不能为空" dc:"模型类型编码"`
+	Schema    map[string]any `json:"schema" v:"required#Schema不能为空" dc:"模型的完整 Schema JSON"`
+}
+
+type BuildSchemaMappingRes struct {
+	SchemaMapping map[string]any `json:"schemaMapping" dc:"生成的 Schema 映射 JSON"`
+}
+

model/dto/model_session_dto.go

@@ -0,0 +1,32 @@
+package dto
+
+import "model-gateway/model/entity"
+
+type CallModelSessionReq struct {
+	Id            int64               `json:"id" v:"required#id不能为空" dc:"任务ID"`
+	ModelInfo     *entity.ModelManage `json:"modelInfo" dc:"模型信息"`
+	RequestParams map[string]any      `json:"requestParams" dc:"新请求参数"`
+}
+
+// CreateModelSessionReq 创建会话
+type CreateModelSessionReq struct {
+	ModelId             int64  `json:"modelId" dc:"模型ID"`
+	BizName             string `json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
+	SessionId           string `json:"sessionId" v:"required#sessionId不能为空" dc:"会话ID"`
+	RequestPath         string `json:"requestPath" dc:"请求参数保存路径"`
+	OriginalRequestPath string `json:"originalRequestPath" dc:"原始请求参数保存路径"`
+}
+
+// UpdateModelSessionReq 修改会话
+type UpdateModelSessionReq struct {
+	Id                   int64   `json:"id" v:"required#id不能为空" dc:"任务ID"`
+	RetryCount           int     `json:"retryCount" dc:"重试"`
+	ResponsePath         string  `json:"responsePath" dc:"响应结果保存路径"`
+	OriginalResponsePath string  `json:"originalResponsePath" dc:"原始响应结果保存路径"`
+	DurationSeconds      int64   `json:"durationSeconds" dc:"耗时（秒）"`
+	PromptTokens         int64   `json:"promptTokens" dc:"输入token"`
+	CompletionTokens     int64   `json:"completionTokens" dc:"输出token"`
+	TotalTokens          int64   `json:"totalTokens" dc:"总token"`
+	TotalCost            float64 `json:"totalCost" dc:"总费用（元）"`
+	ErrorMsg             string  `json:"errorMsg" dc:"错误消息"`
+}

model/dto/model_task_end_dto.go

@@ -0,0 +1,16 @@
+package dto
+
+type CreateModelTaskEndReq struct {
+	ModelId                int64          `json:"modelId" dc:"模型ID"`
+	BizName                string         `json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
+	MsgTopic               string         `json:"msgTopic" dc:"消息主题（可选，用于后续业务通知）"`
+	TaskId                 string         `json:"taskId" dc:"任务ID"`
+	ResponseParams         map[string]any `json:"responseParams" dc:"响应结果"`
+	OriginalResponseParams map[string]any `json:"originalResponseParams" dc:"原始响应结果"`
+	DurationSeconds        int64          `json:"durationSeconds" dc:"耗时（秒）"`
+	PromptTokens           int64          `json:"promptTokens" dc:"输入token"`
+	CompletionTokens       int64          `json:"completionTokens" dc:"输出token"`
+	TotalTokens            int64          `json:"totalTokens" dc:"总token"`
+	TotalCost              float64        `json:"totalCost" dc:"本次调用总费用（元），未配置计费规则为0"`
+	ErrorMsg               string         `json:"errorMsg" dc:"错误消息"`
+}

model/dto/model_task_start_dto.go

@@ -0,0 +1,46 @@
+package dto
+
+import (
+	"model-gateway/model/entity"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+type CallModelTaskStartReq struct {
+	Id            int64               `json:"id" v:"required#id不能为空" dc:"任务ID"`
+	ModelInfo     *entity.ModelManage `json:"modelInfo" dc:"模型信息"`
+	RequestParams map[string]any      `json:"requestParams" dc:"新请求参数"`
+}
+
+// CreateModelTaskStartReq 创建任务
+type CreateModelTaskStartReq struct {
+	ModelId             int64  `json:"modelId" dc:"模型ID"`
+	BizName             string `json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
+	MsgTopic            string `json:"msgTopic" dc:"消息主题（可选，用于后续业务通知）"`
+	RequestPath         string `json:"requestPath" dc:"请求参数保存路径"`
+	OriginalRequestPath string `json:"originalRequestPath" dc:"原始请求参数保存路径"`
+	MediaType           string `json:"mediaType" dc:"输入媒体类型快照（audio/video，空=无媒体引用；shop 计费词汇，创建任务时按请求体推导）"`
+}
+
+type CreateModelTaskStartRes struct {
+	Id int64 `json:"id" dc:"任务ID"`
+}
+
+// UpdateModelTaskStartReq 修改任务
+type UpdateModelTaskStartReq struct {
+	Id                     int64          `json:"id" v:"required#id不能为空" dc:"任务ID"`
+	RetryCount             int            `json:"retryCount" dc:"重试"`
+	ResponseParams         map[string]any `json:"responseParams" dc:"响应结果"`
+	OriginalResponseParams map[string]any `json:"originalResponseParams" dc:"原始响应结果"`
+	TaskId                 string         `json:"taskId" dc:"任务ID"`
+	DurationSeconds        int64          `json:"durationSeconds" dc:"耗时（秒）"`
+	ErrorMsg               string         `json:"errorMsg" dc:"错误消息"`
+}
+
+type DeleteModelTaskStartReq struct {
+	Id int64 `json:"id" v:"required#ids不能为空" dc:"任务ID"`
+}
+
+type GetModelTaskStartListReq struct {
+	Page *beans.Page `json:"page"`
+}

model/dto/stat_dto.go

@@ -1,20 +0,0 @@
-package dto
-
-import "github.com/gogf/gf/v2/frame/g"
-
-// ListModelStatReq 统计列表
-type ListModelStatReq struct {
-	g.Meta    `path:"/listModelStat" method:"get" tags:"统计" summary:"模型请求统计列表" dc:"按天统计模型请求次数，支持分页与条件筛选"`
-	PageNum   int         `p:"pageNum" json:"pageNum" dc:"页码（默认1）"`
-	PageSize  int         `p:"pageSize" json:"pageSize" dc:"每页条数（默认10）"`
-	StartDay  string      `p:"startDay" json:"startDay" dc:"开始日期（YYYY-MM-DD，可选）"`
-	EndDay    string      `p:"endDay" json:"endDay" dc:"结束日期（YYYY-MM-DD，可选）"`
-	TenantID  *int64      `p:"tenantId" json:"tenantId" dc:"租户ID（可选）"`
-	Creator   string      `p:"creator" json:"creator" dc:"创建人（可选，模糊匹配）"`
-	ModelName string      `p:"modelName" json:"modelName" dc:"模型名称（可选，模糊匹配）"`
-}
-
-type ListModelStatRes struct {
-	List  any   `json:"list" dc:"列表数据"`
-	Total int64 `json:"total" dc:"总数"`
-}

model/dto/task_dto.go

@@ -1,80 +0,0 @@
-package dto
-
-import "github.com/gogf/gf/v2/frame/g"
-
-// CreateTaskReq 创建异步任务
-type CreateTaskReq struct {
-	g.Meta         `path:"/createTask" method:"post" tags:"任务管理" summary:"创建异步任务" dc:"创建异步任务并返回任务ID；创建成功后会立即异步尝试执行当前任务，执行成功后按回调配置触发钩子"`
-	ModelName      string `p:"modelName" json:"modelName" v:"required#modelName不能为空" dc:"模型名称"`
-	BizName        string `p:"bizName" json:"bizName" dc:"业务名称（调用方模块/系统，用于统计）"`
-	CallbackUrl    string `p:"callbackUrl" json:"callbackUrl" dc:"回调地址（可选，用于后续业务通知）"`
-	InputRef       string `p:"inputRef" json:"inputRef" dc:"输入引用（如OSS/文件引用等）"`
-	RequestPayload any    `p:"requestPayload" json:"requestPayload" dc:"请求负载（透传给模型服务）"`
-	EpicycleId     int64  `json:"epicycleId" dc:"轮次ID"`
-}
-
-type CreateTaskRes struct {
-	TaskID string `json:"taskId" dc:"任务ID"`
-}
-
-// GetTaskResultReq 获取结果（只返回 oss 地址）
-type GetTaskResultReq struct {
-	g.Meta `path:"/getTaskResult" method:"get" tags:"任务管理" summary:"获取任务结果" dc:"根据任务ID获取结果（只返回OSS地址）"`
-	TaskID string `p:"taskId" json:"taskId" v:"required#taskId不能为空" dc:"任务ID"`
-}
-
-type GetTaskResultRes struct {
-	OssFile string `json:"ossFile" dc:"结果文件OSS地址"`
-	State   int    `json:"state" dc:"任务状态"`
-}
-
-// GetTaskBatchReq 批量查询任务（并对成功任务标记为已下载）
-type GetTaskBatchReq struct {
-	g.Meta  `path:"/getTaskBatch" method:"post" tags:"任务管理" summary:"批量查询任务" dc:"批量查询任务状态与OSS地址；对成功(state=2)的任务自动标记为已下载(state=4)，并写入保留到期时间"`
-	TaskIDs []string `p:"taskIds" json:"taskIds" v:"required#taskIds不能为空" dc:"任务ID列表"`
-}
-
-type GetTaskBatchItem struct {
-	TaskID  string `json:"taskId" dc:"任务ID"`
-	State   int    `json:"state" dc:"任务状态"`
-	OssFile string `json:"ossFile" dc:"结果文件OSS地址"`
-}
-
-type GetTaskBatchRes struct {
-	List []GetTaskBatchItem `json:"list" dc:"任务列表"`
-}
-
-// ListTaskReq 任务列表分页查询
-type ListTaskReq struct {
-	g.Meta    `path:"/listTask" method:"get" tags:"任务管理" summary:"任务列表" dc:"分页查询任务列表，支持按状态/模型名称/task_id过滤"`
-	PageNum   int    `p:"pageNum" json:"pageNum" dc:"页码（默认1）"`
-	PageSize  int    `p:"pageSize" json:"pageSize" dc:"每页条数（默认10）"`
-	ModelName string `p:"modelName" json:"modelName" dc:"模型名称（模糊匹配）"`
-	TaskID    string `p:"taskId" json:"taskId" dc:"任务ID（模糊匹配）"`
-	State     *int   `p:"state" json:"state" dc:"任务状态（0/1/2/3/4，可选）"`
-}
-
-type ListTaskRes struct {
-	List  any   `json:"list" dc:"列表数据"`
-	Total int64 `json:"total" dc:"总数"`
-}
-
-// RunWorkReq 手动触发 worker 执行一次（由上层定时任务调用）
-type RunWorkReq struct {
-	g.Meta     `path:"/runWork" method:"post" tags:"任务管理" summary:"执行一次Worker" dc:"手动触发一次Worker抢占并处理排队中的任务；适合处理 createTask 立即执行时未处理到的任务以及积压队列"`
-	BatchSize  int `p:"batchSize" json:"batchSize" dc:"本次抢占任务数量（默认10）"`
-	Goroutines int `p:"goroutines" json:"goroutines" dc:"本次并发数（默认1）"`
-}
-
-type RunWorkRes struct {
-	Claimed int `json:"claimed" dc:"本次抢占并处理的任务数"`
-}
-
-// CleanWorkReq 手动触发 cleaner 执行一次（由上层定时任务调用）
-type CleanWorkReq struct {
-	g.Meta `path:"/cleanWork" method:"post" tags:"任务管理" summary:"执行一次Cleaner" dc:"手动触发一次清理/重试（用于由上层定时任务控制）"`
-}
-
-type CleanWorkRes struct {
-	Ok bool `json:"ok" dc:"是否执行成功"`
-}

model/entity/asynch_model.go

@@ -1,88 +0,0 @@
-package entity
-
-import "gitea.com/red-future/common/beans"
-
-type asynchModelCol struct {
-	beans.SQLBaseCol
-	ModelName         string
-	ModelType         string
-	BaseURL           string
-	HttpMethod        string
-	HeadMsg           string
-	FormJSON          string
-	RequestMapping    string
-	ResponseMapping   string
-	ResponseBody      string
-	TokenMapping      string
-	Prompt            string
-	IsPrivate         string
-	IsChatModel       string
-	ApiKey            string
-	Enabled           string
-	MaxConcurrency    string
-	QueueLimit        string
-	TimeoutSeconds    string
-	ExpectedSeconds   string
-	RetryTimes        string
-	RetryQueueMaxSecs string
-	AutoCleanSeconds  string
-	Remark            string
-	IsOwner           string
-}
-
-var AsynchModelCol = asynchModelCol{
-	SQLBaseCol:        beans.DefSQLBaseCol,
-	ModelName:         "model_name",
-	ModelType:         "model_type",
-	BaseURL:           "base_url",
-	HttpMethod:        "http_method",
-	HeadMsg:           "head_msg",
-	FormJSON:          "form_json",
-	RequestMapping:    "request_mapping",
-	ResponseMapping:   "response_mapping",
-	ResponseBody:      "response_body",
-	TokenMapping:      "token_mapping",
-	Prompt:            "prompt",
-	IsPrivate:         "is_private",
-	IsChatModel:       "is_chat_model",
-	ApiKey:            "api_key",
-	Enabled:           "enabled",
-	MaxConcurrency:    "max_concurrency",
-	QueueLimit:        "queue_limit",
-	TimeoutSeconds:    "timeout_seconds",
-	ExpectedSeconds:   "expected_seconds",
-	RetryTimes:        "retry_times",
-	RetryQueueMaxSecs: "retry_queue_max_seconds",
-	AutoCleanSeconds:  "auto_clean_seconds",
-	Remark:            "remark",
-	IsOwner:           "is_owner",
-}
-
-// AsynchModel 异步模型配置
-type AsynchModel struct {
-	beans.SQLBaseDO      `orm:",inline"`
-	ModelName            string `orm:"model_name" json:"modelName"`
-	ModelType            int    `orm:"model_type" json:"modelType"`
-	BaseURL              string `orm:"base_url" json:"baseUrl"`
-	HttpMethod           string `orm:"http_method" json:"httpMethod"`
-	HeadMsg              string `orm:"head_msg" json:"headMsg"`
-	Form                 any    `orm:"form_json" json:"form"`
-	RequestMapping       any    `orm:"request_mapping" json:"requestMapping"`
-	ResponseMapping      any    `orm:"response_mapping" json:"responseMapping"`
-	ResponseBody         any    `orm:"response_body" json:"responseBody"`
-	TokenMapping         string `orm:"token_mapping" json:"tokenMapping"`
-	Prompt               string `orm:"prompt" json:"prompt"`
-	IsPrivate            *int   `orm:"is_private" json:"isPrivate"`
-	IsChatModel          *int   `orm:"is_chat_model" json:"isChatModel"`
-	ApiKey               string `orm:"api_key" json:"apiKey"`
-	Enabled              *int   `orm:"enabled" json:"enabled"`
-	MaxConcurrency       int    `orm:"max_concurrency" json:"maxConcurrency"`
-	QueueLimit           int    `orm:"queue_limit" json:"queueLimit"`
-	TimeoutSeconds       int    `orm:"timeout_seconds" json:"timeoutSeconds"`
-	ExpectedSeconds      int    `orm:"expected_seconds" json:"expectedSeconds"`
-	RetryTimes           int    `orm:"retry_times" json:"retryTimes"`
-	RetryQueueMaxSeconds int    `orm:"retry_queue_max_seconds" json:"retryQueueMaxSeconds"`
-	AutoCleanSeconds     int    `orm:"auto_clean_seconds" json:"autoCleanSeconds"`
-	Remark               string `orm:"remark" json:"remark"`
-	IsOwner              *int   `json:"isOwner" orm:"is_owner"` // 1=当前用户创建的，0=超级管理员的
-}

model/entity/asynch_model_type.go

@@ -1,26 +0,0 @@
-package entity
-
-import "gitea.com/red-future/common/beans"
-
-type asynchModelTypeCol struct {
-	beans.SQLBaseCol
-	TypeID   string
-	TypeName string
-	Remark   string
-}
-
-var AsynchModelTypeCol = asynchModelTypeCol{
-	SQLBaseCol: beans.DefSQLBaseCol,
-	TypeID:     "type_id",
-	TypeName:   "type_name",
-	Remark:     "remark",
-}
-
-// AsynchModelType 模型类型（图片/音频/视频等）
-type AsynchModelType struct {
-	beans.SQLBaseDO `orm:",inline"`
-	TypeID         int    `orm:"type_id" json:"typeId"`
-	TypeName       string `orm:"type_name" json:"type"`
-	Remark         string `orm:"remark" json:"remark"`
-}
-

model/entity/asynch_task.go

@@ -1,89 +0,0 @@
-package entity
-
-import (
-	"gitea.com/red-future/common/beans"
-	"github.com/gogf/gf/v2/os/gtime"
-)
-
-type asynchTaskCol struct {
-	beans.SQLBaseCol
-	ModelName       string
-	TaskID          string
-	BizName         string
-	CallbackURL     string
-	ModelKey        string
-	State           string
-	OssFile         string
-	FileType        string
-	FileSize        string
-	ErrorMsg        string
-	StartedAt       string
-	FinishedAt      string
-	DurationSeconds string
-	ExpireAt        string
-	RetryCount      string
-	EnqueueAt       string
-	Phase           string
-	TmpFile         string
-	InputRef        string
-	RequestPayload  string
-	TextResult      string
-	EpicycleId      string
-	ExpendTokens    string
-}
-
-var AsynchTaskCol = asynchTaskCol{
-	SQLBaseCol:      beans.DefSQLBaseCol,
-	ModelName:       "model_name",
-	TaskID:          "task_id",
-	BizName:         "biz_name",
-	CallbackURL:     "callback_url",
-	ModelKey:        "model_key",
-	State:           "state",
-	OssFile:         "oss_file",
-	FileType:        "file_type",
-	FileSize:        "file_size",
-	ErrorMsg:        "error_msg",
-	StartedAt:       "started_at",
-	FinishedAt:      "finished_at",
-	DurationSeconds: "duration_seconds",
-	ExpireAt:        "expire_at",
-	RetryCount:      "retry_count",
-	EnqueueAt:       "enqueue_at",
-	Phase:           "phase",
-	TmpFile:         "tmp_file",
-	InputRef:        "input_ref",
-	RequestPayload:  "request_payload",
-	TextResult:      "text_result",
-	EpicycleId:      "epicycle_id",
-	ExpendTokens:    "expend_tokens",
-}
-
-// AsynchTask 异步任务
-type AsynchTask struct {
-	beans.SQLBaseDO      `orm:",inline"`
-	ModelName            string      `orm:"model_name" json:"modelName"`
-	TaskID               string      `orm:"task_id" json:"taskId"`
-	BizName              string      `orm:"biz_name" json:"bizName"`
-	CallbackURL          string      `orm:"callback_url" json:"callbackUrl"`
-	ModelKey             string      `orm:"model_key" json:"modelKey"`
-	State                int         `orm:"state" json:"state"` // 0排队中/1执行中/2成功/3失败/4已下载
-	OssFile              string      `orm:"oss_file" json:"ossFile"`
-	FileType             string      `orm:"file_type" json:"fileType"`
-	FileSize             int64       `orm:"file_size" json:"fileSize"`
-	ErrorMsg             string      `orm:"error_msg" json:"errorMsg"`
-	StartedAt            *gtime.Time `orm:"started_at" json:"startedAt"`
-	FinishedAt           *gtime.Time `orm:"finished_at" json:"finishedAt"`
-	DurationSeconds      int64       `orm:"duration_seconds" json:"durationSeconds"`
-	ExpireAt             *gtime.Time `orm:"expire_at" json:"expireAt"` // 已下载(state=4)后的过期时间
-	RetryCount           int         `orm:"retry_count" json:"retryCount"`
-	EnqueueAt            *gtime.Time `orm:"enqueue_at" json:"enqueueAt"`
-	Phase                int         `orm:"phase" json:"phase"`      // 0模型阶段/1OSS阶段
-	TmpFile              string      `orm:"tmp_file" json:"tmpFile"` // 临时结果文件路径
-	InputRef             string      `orm:"input_ref" json:"inputRef"`
-	RequestPayload       any         `orm:"request_payload" json:"requestPayload"`
-	TextResult           string      `orm:"text_result" json:"text"`
-	EpicycleId           int64       `orm:"epicycle_id" json:"epicycleId"`     // 轮次ID（用于标识同一轮次的任务）
-	ExpendTokens         int64       `orm:"expend_tokens" json:"expendTokens"` // 消耗 token 数
-	RetryQueueMaxSeconds int         `orm:"retry_queue_max_seconds" json:"-"`
-}

model/entity/logs_model_op.go

@@ -1,57 +0,0 @@
-package entity
-
-import (
-	"gitea.com/red-future/common/beans"
-)
-
-type LogsModelPpCol struct {
-	beans.SQLBaseCol
-	IP              string
-	UserAgent       string
-	APIPath         string
-	HttpMethod      string
-	BizName         string
-	ModelName       string
-	TaskID          string
-	OpType          string
-	Success         string
-	ErrorMsg        string
-	CostMs          string
-	RequestPayload  string
-	ResponsePayload string
-}
-
-var LogsModelOpCol = LogsModelPpCol{
-	SQLBaseCol:      beans.DefSQLBaseCol,
-	IP:              "ip",
-	UserAgent:       "user_agent",
-	APIPath:         "api_path",
-	HttpMethod:      "http_method",
-	BizName:         "biz_name",
-	ModelName:       "model_name",
-	TaskID:          "task_id",
-	OpType:          "op_type",
-	Success:         "success",
-	ErrorMsg:        "error_msg",
-	CostMs:          "cost_ms",
-	RequestPayload:  "request_payload",
-	ResponsePayload: "response_payload",
-}
-
-// LogsModelOp 操作日志（创建任务等）
-type LogsModelOp struct {
-	beans.SQLBaseDO `orm:",inline"`
-	IP              string `orm:"ip" json:"ip"`
-	UserAgent       string `orm:"user_agent" json:"userAgent"`
-	APIPath         string `orm:"api_path" json:"apiPath"`
-	HttpMethod      string `orm:"http_method" json:"httpMethod"`
-	BizName         string `orm:"biz_name" json:"bizName"`
-	ModelName       string `orm:"model_name" json:"modelName"`
-	TaskID          string `orm:"task_id" json:"taskId"`
-	OpType          string `orm:"op_type" json:"opType"`
-	Success         int    `orm:"success" json:"success"`
-	ErrorMsg        string `orm:"error_msg" json:"errorMsg"`
-	CostMs          int64  `orm:"cost_ms" json:"costMs"`
-	RequestPayload  any    `orm:"request_payload" json:"requestPayload"`
-	ResponsePayload any    `orm:"response_payload" json:"responsePayload"`
-}

model/entity/logs_model_stat.go

@@ -1,38 +0,0 @@
-package entity
-
-import (
-	"github.com/gogf/gf/v2/os/gtime"
-)
-
-// LogsModelStatCol 字段常量
-type LogsModelStatCol struct {
-	Day          string
-	TenantId     string
-	Creator      string
-	ModelName    string
-	RequestCount string
-	CreatedAt    string
-	UpdatedAt    string
-}
-
-var LogsModelStatCols = LogsModelStatCol{
-	Day:          "day",
-	TenantId:     "tenant_id",
-	Creator:      "creator",
-	ModelName:    "model_name",
-	RequestCount: "request_count",
-	CreatedAt:    "created_at",
-	UpdatedAt:    "updated_at",
-}
-
-// LogsModelStat 按天统计：某天/租户/创建人/模型的请求次数
-// 注：这里不走通用 SQLBaseDO，采用联合唯一键（day,tenant_id,creator,model_name）做 UPSERT 原子累加。
-type LogsModelStat struct {
-	Day          *gtime.Time `orm:"day" json:"day"`                    // 日期（建议仅使用日期部分）
-	TenantId     int64       `orm:"tenant_id" json:"tenantId"`         // 租户ID
-	Creator      string      `orm:"creator" json:"creator"`            // 创建人/操作人
-	ModelName    string      `orm:"model_name" json:"modelName"`       // 模型名称
-	RequestCount int64       `orm:"request_count" json:"requestCount"` // 请求次数
-	CreatedAt    *gtime.Time `orm:"created_at" json:"createdAt"`       // 创建时间
-	UpdatedAt    *gtime.Time `orm:"updated_at" json:"updatedAt"`       // 更新时间
-}

model/entity/model_error_memory.go

@@ -0,0 +1,37 @@
+package entity
+
+import "gitea.redpowerfuture.com/red-future/common/beans"
+
+type modelErrorMemoryCol struct {
+	beans.SQLBaseCol
+	MemoryKey      string
+	Upstream       string
+	ErrorCode      string
+	MsgFingerprint string
+	Retryable      string
+	Reason         string
+	AnalyzedBy     string
+}
+
+var ModelErrorMemoryCol = modelErrorMemoryCol{
+	SQLBaseCol:     beans.DefSQLBaseCol,
+	MemoryKey:      "memory_key",
+	Upstream:       "upstream",
+	ErrorCode:      "error_code",
+	MsgFingerprint: "msg_fingerprint",
+	Retryable:      "retryable",
+	Reason:         "reason",
+	AnalyzedBy:     "analyzed_by",
+}
+
+// ModelErrorMemory 错误重试记忆（LLM 分析结论持久化，永久有效）
+type ModelErrorMemory struct {
+	beans.SQLBaseDO `orm:",inline"`
+	MemoryKey       string `orm:"memory_key" json:"memoryKey" dc:"记忆键=SHA-256(upstream|code|归一化消息)"`
+	Upstream        string `orm:"upstream" json:"upstream" dc:"失败上游BaseURL"`
+	ErrorCode       string `orm:"error_code" json:"errorCode" dc:"错误码"`
+	MsgFingerprint  string `orm:"msg_fingerprint" json:"msgFingerprint" dc:"归一化消息md5"`
+	Retryable       bool   `orm:"retryable" json:"retryable" dc:"是否可重试"`
+	Reason          string `orm:"reason" json:"reason" dc:"分析原因"`
+	AnalyzedBy      string `orm:"analyzed_by" json:"analyzedBy" dc:"分析模型名"`
+}

model/entity/model_manage.go

@@ -0,0 +1,116 @@
+package entity
+
+import (
+	"model-gateway/consts/model"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+type modelManageCol struct {
+	beans.SQLBaseCol
+	ModelSupplier                string
+	ModelName                    string
+	ModelType                    string
+	BaseURL                      string
+	SystemModel                  string
+	RefSystemModelId             string
+	HttpMethod                   string
+	ChatModel                    string
+	ResponseType                 string
+	ApiKey                       string
+	Enabled                      string
+	RequestHeadMapping           string
+	RequestBodyMapping           string
+	RequestBusinessFieldMapping  string
+	ResponseMapping              string
+	ResponseBodyMapping          string
+	ResponseBusinessFieldMapping string
+	MaxConcurrency               string
+	TokenPredictPrice            string
+	TokenPredictPriceUnit        string
+	MaxTokens                    string
+	MinDuration                  string
+	MaxDuration                  string
+	LastFrame                    string
+}
+
+var ModelManageCol = modelManageCol{
+	SQLBaseCol:                   beans.DefSQLBaseCol,
+	ModelSupplier:                "model_supplier",
+	ModelName:                    "model_name",
+	ModelType:                    "model_type",
+	BaseURL:                      "base_url",
+	SystemModel:                  "system_model",
+	RefSystemModelId:             "ref_system_model_id",
+	HttpMethod:                   "http_method",
+	ChatModel:                    "chat_model",
+	ResponseType:                 "response_type",
+	ApiKey:                       "api_key",
+	Enabled:                      "enabled",
+	RequestHeadMapping:           "request_head_mapping",
+	RequestBodyMapping:           "request_body_mapping",
+	RequestBusinessFieldMapping:  "request_business_field_mapping",
+	ResponseMapping:              "response_mapping",
+	ResponseBodyMapping:          "response_body_mapping",
+	ResponseBusinessFieldMapping: "response_business_field_mapping",
+	MaxConcurrency:               "max_concurrency",
+	TokenPredictPrice:            "token_predict_price",
+	TokenPredictPriceUnit:        "token_predict_price_unit",
+	MaxTokens:                    "max_tokens",
+	MinDuration:                  "min_duration",
+	MaxDuration:                  "max_duration",
+	LastFrame:                    "last_frame",
+}
+
+type ModelManage struct {
+	beans.SQLBaseDO              `orm:",inline"`
+	ModelSupplier                model.SupplierType `orm:"model_supplier" json:"modelSupplier" description:"模型供应商"`
+	ModelName                    string             `orm:"model_name" json:"modelName" description:"模型名称"`
+	ModelType                    model.ModelType    `orm:"model_type" json:"modelType" description:"模型类型"`
+	BaseURL                      string             `orm:"base_url" json:"baseUrl" description:"模型地址"`
+	SystemModel                  *bool              `orm:"system_model" json:"systemModel" description:"系统模型"`
+	RefSystemModelId             int64              `orm:"ref_system_model_id" json:"refSystemModelId" description:"引用的系统模型ID（NULL=非引用行）"`
+	HttpMethod                   string             `orm:"http_method" json:"httpMethod" description:"http方法"`
+	ChatModel                    *bool              `orm:"chat_model" json:"chatModel" description:"是否聊天模型"`
+	ResponseType                 model.ResponseType `orm:"response_type" json:"responseType" description:"返回类型：1同步，2异步，3流"`
+	ApiKey                       string             `orm:"api_key" json:"apiKey" description:"api key"`
+	Enabled                      *bool              `orm:"enabled" json:"enabled" description:"是否启用"`
+	RequestHeadMapping           map[string]string  `orm:"request_head_mapping" json:"requestHeadMapping" description:"请求头映射"`
+	RequestBodyMapping           map[string]any     `orm:"request_body_mapping" json:"requestBodyMapping" description:"请求体映射"`
+	RequestBusinessFieldMapping  map[string]string  `orm:"request_business_field_mapping" json:"requestBusinessFieldMapping" description:"请求业务字段映射"`
+	ResponseMapping              map[string]any     `orm:"response_mapping" json:"responseMapping" description:"响应映射"`
+	ResponseBodyMapping          map[string]string  `orm:"response_body_mapping" json:"responseBodyMapping" description:"响应主体映射"`
+	ResponseBusinessFieldMapping map[string]string  `orm:"response_business_field_mapping" json:"responseBusinessFieldMapping" description:"响应业务字段映射"`
+	MaxConcurrency               int                `orm:"max_concurrency" json:"maxConcurrency" description:"最大并发数"`
+	TokenMapping                 *TokenMapping      `orm:"token_mapping" json:"tokenMapping" description:"token映射"`
+	AsyncTaskMapping             *AsyncTaskMapping  `orm:"async_task_mapping" json:"asyncTaskMapping" description:"异步任务映射"`
+	TokenPredictPrice            float64            `orm:"token_predict_price" json:"tokenPredictPrice" description:"模型Token预估价格"`
+	TokenPredictPriceUnit        string             `orm:"token_predict_price_unit" json:"tokenPredictPriceUnit" description:"模型token预估价格单位（秒，百万Token，千Token，字数）"`
+	MaxTokens                    int                `orm:"max_tokens" json:"maxTokens" description:"最大token数"`
+	MinDuration                  int                `orm:"min_duration" json:"minDuration" description:"最小时长（秒）"`
+	MaxDuration                  int                `orm:"max_duration" json:"maxDuration" description:"最大时长（秒）"`
+	LastFrame                    string             `orm:"last_frame" json:"lastFrame" description:"视频的尾帧图像"`
+	ErrorMessageMapping          map[string]any     `orm:"error_message_mapping" json:"errorMessageMapping" description:"错误消息映射"`
+}
+
+type TokenMapping struct {
+	PromptTokens     string `json:"promptTokens" dc:"输入token"`
+	CompletionTokens string `json:"completionTokens" dc:"输出token"`
+	TotalTokens      string `json:"totalTokens" dc:"总token"`
+}
+
+type AsyncTaskMapping struct {
+	Url                string            `json:"url" dc:"url"`
+	HttpMethod         string            `json:"httpMethod" dc:"http方法" d:"POST"`
+	RequestBodyMapping map[string]any    `json:"requestBodyMapping" description:"请求体映射"`
+	RequestHeadMapping map[string]string `json:"requestHeadMapping" description:"请求头映射"`
+	ResponseMapping    map[string]any    `json:"responseMapping" description:"响应映射"`
+	TaskId             string            `json:"taskId" dc:"任务id"`
+	TaskStatus         string            `json:"taskStatus" dc:"任务状态"`
+	TaskStatusPending  string            `json:"taskStatusPending" dc:"任务状态-待处理"`
+	TaskStatusRunning  string            `json:"taskStatusRunning" dc:"任务状态-运行中"`
+	TaskStatusSuccess  string            `json:"taskStatusSuccess" dc:"任务状态-成功"`
+	TaskStatusFailed   string            `json:"taskStatusFailed" dc:"任务状态-失败"`
+	TaskStatusCancel   string            `json:"taskStatusCancel" dc:"任务状态-取消"`
+	TaskStatusUnknown  string            `json:"taskStatusUnknown" dc:"任务状态-未知"`
+}

model/entity/model_session.go

@@ -0,0 +1,60 @@
+package entity
+
+import (
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+type modelSessionCol struct {
+	beans.SQLBaseCol
+	ModelId              string
+	BizName              string
+	SessionId            string
+	RetryCount           string
+	RequestPath          string
+	ResponsePath         string
+	OriginalRequestPath  string
+	OriginalResponsePath string
+	DurationSeconds      string
+	PromptTokens         string
+	CompletionTokens     string
+	TotalTokens          string
+	TotalCost            string
+	ErrorMsg             string
+}
+
+var ModelSessionCol = modelSessionCol{
+	SQLBaseCol:           beans.DefSQLBaseCol,
+	ModelId:              "model_id",
+	BizName:              "biz_name",
+	SessionId:            "session_id",
+	RetryCount:           "retry_count",
+	RequestPath:          "request_path",
+	ResponsePath:         "response_path",
+	OriginalRequestPath:  "original_request_path",
+	OriginalResponsePath: "original_response_path",
+	DurationSeconds:      "duration_seconds",
+	PromptTokens:         "prompt_tokens",
+	CompletionTokens:     "completion_tokens",
+	TotalTokens:          "total_tokens",
+	TotalCost:            "total_cost",
+	ErrorMsg:             "error_msg",
+}
+
+// ModelSession 模型网关任务
+type ModelSession struct {
+	beans.SQLBaseDO      `orm:",inline"`
+	ModelId              int64   `orm:"model_id" json:"modelId" dc:"模型ID"`
+	BizName              string  `orm:"biz_name" json:"bizName" dc:"业务名称"`
+	SessionId            string  `orm:"session_id" json:"sessionId" dc:"会话ID"`
+	RetryCount           int     `orm:"retry_count" json:"retryCount" dc:"重试"`
+	RequestPath          string  `orm:"request_path" json:"requestPath" dc:"请求参数保存路径"`
+	ResponsePath         string  `orm:"response_path" json:"responsePath" dc:"响应结果保存路径"`
+	OriginalRequestPath  string  `orm:"original_request_path" json:"originalRequestPath" dc:"原始请求参数保存路径"`
+	OriginalResponsePath string  `orm:"original_response_path" json:"originalResponsePath" dc:"原始响应结果保存路径"`
+	DurationSeconds      int64   `orm:"duration_seconds" json:"durationSeconds" dc:"耗时（秒）"`
+	PromptTokens         int64   `orm:"prompt_tokens" json:"promptTokens" dc:"输入token"`
+	CompletionTokens     int64   `orm:"completion_tokens" json:"completionTokens" dc:"输出token"`
+	TotalTokens          int64   `orm:"total_tokens" json:"totalTokens" dc:"总token"`
+	TotalCost            float64 `orm:"total_cost" json:"totalCost" dc:"总费用（元）"`
+	ErrorMsg             string  `orm:"error_msg" json:"errorMsg" dc:"错误消息"`
+}

model/entity/model_task_end.go

@@ -0,0 +1,57 @@
+package entity
+
+import (
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+type modelTaskEndCol struct {
+	beans.SQLBaseCol
+	ModelId                string
+	BizName                string
+	MsgTopic               string
+	RetryCount             string
+	ResponseParams         string
+	OriginalResponseParams string
+	DurationSeconds        string
+	TaskId                 string
+	PromptTokens           string
+	CompletionTokens       string
+	TotalTokens            string
+	TotalCost              string
+	ErrorMsg               string
+}
+
+var ModelTaskEndCol = modelTaskEndCol{
+	SQLBaseCol:             beans.DefSQLBaseCol,
+	ModelId:                "model_id",
+	BizName:                "biz_name",
+	MsgTopic:               "msg_topic",
+	RetryCount:             "retry_count",
+	ResponseParams:         "response_params",
+	OriginalResponseParams: "original_response_params",
+	DurationSeconds:        "duration_seconds",
+	TaskId:                 "task_id",
+	PromptTokens:           "prompt_tokens",
+	CompletionTokens:       "completion_tokens",
+	TotalTokens:            "total_tokens",
+	TotalCost:              "total_cost",
+	ErrorMsg:               "error_msg",
+}
+
+// ModelTaskEnd 模型网关任务
+type ModelTaskEnd struct {
+	beans.SQLBaseDO        `orm:",inline"`
+	ModelId                int64   `orm:"model_id" json:"modelId" dc:"模型ID"`
+	BizName                string  `orm:"biz_name" json:"bizName" dc:"业务名称"`
+	MsgTopic               string  `orm:"msg_topic" json:"msgTopic" dc:"消息主题"`
+	RetryCount             int     `orm:"retry_count" json:"retryCount" dc:"重试"`
+	ResponseParams         string  `orm:"response_params" json:"responseParams" dc:"响应结果"`
+	OriginalResponseParams string  `orm:"original_response_params" json:"originalResponseParams" dc:"原始响应结果"`
+	DurationSeconds        int64   `orm:"duration_seconds" json:"durationSeconds" dc:"耗时（秒）"`
+	TaskId                 string  `orm:"task_id" json:"taskId" dc:"任务ID"`
+	PromptTokens           int64   `orm:"prompt_tokens" json:"promptTokens" dc:"输入token"`
+	CompletionTokens       int64   `orm:"completion_tokens" json:"completionTokens" dc:"输出token"`
+	TotalTokens            int64   `orm:"total_tokens" json:"totalTokens" dc:"总token"`
+	TotalCost              float64 `orm:"total_cost" json:"totalCost" dc:"本次调用总费用（元），未配置计费规则为0"`
+	ErrorMsg               string  `orm:"error_msg" json:"errorMsg" dc:"错误消息"`
+}

model/entity/model_task_start.go

@@ -0,0 +1,54 @@
+package entity
+
+import (
+	"gitea.redpowerfuture.com/red-future/common/beans"
+)
+
+type modelTaskStartCol struct {
+	beans.SQLBaseCol
+	ModelId                string
+	BizName                string
+	MsgTopic               string
+	RetryCount             string
+	RequestPath            string
+	OriginalRequestPath    string
+	ResponseParams         string
+	OriginalResponseParams string
+	DurationSeconds        string
+	TaskId                 string
+	MediaType              string
+	ErrorMsg               string
+}
+
+var ModelTaskStartCol = modelTaskStartCol{
+	SQLBaseCol:             beans.DefSQLBaseCol,
+	ModelId:                "model_id",
+	BizName:                "biz_name",
+	MsgTopic:               "msg_topic",
+	RetryCount:             "retry_count",
+	RequestPath:            "request_path",
+	OriginalRequestPath:    "original_request_path",
+	ResponseParams:         "response_params",
+	OriginalResponseParams: "original_response_params",
+	DurationSeconds:        "duration_seconds",
+	TaskId:                 "task_id",
+	MediaType:              "media_type",
+	ErrorMsg:               "error_msg",
+}
+
+// ModelTaskStart 模型网关任务
+type ModelTaskStart struct {
+	beans.SQLBaseDO        `orm:",inline"`
+	ModelId                int64          `orm:"model_id" json:"modelId" dc:"模型ID"`
+	BizName                string         `orm:"biz_name" json:"bizName" dc:"业务名称"`
+	MsgTopic               string         `orm:"msg_topic" json:"msgTopic" dc:"消息主题"`
+	RetryCount             int            `orm:"retry_count" json:"retryCount" dc:"重试"`
+	RequestPath            string         `orm:"request_path" json:"requestPath" dc:"请求参数保存路径"`
+	OriginalRequestPath    string         `orm:"original_request_path" json:"originalRequestPath" dc:"原始请求参数保存路径"`
+	ResponseParams         map[string]any `orm:"response_params" json:"responseParams" dc:"响应结果"`
+	OriginalResponseParams map[string]any `orm:"original_response_params" json:"originalResponseParams" dc:"原始响应结果"`
+	DurationSeconds        int64          `orm:"duration_seconds" json:"durationSeconds" dc:"耗时（秒）"`
+	TaskId                 string         `orm:"task_id" json:"taskId" dc:"任务ID"`
+	MediaType              string         `orm:"media_type" json:"mediaType" dc:"输入媒体类型快照（audio/video，空=无媒体引用；shop 计费词汇，创建任务时按请求体推导）"`
+	ErrorMsg               string         `orm:"error_msg" json:"errorMsg" dc:"错误消息"`
+}

service/auto_tune.go

@@ -1,194 +0,0 @@
-package service
-
-import (
-	"context"
-	"fmt"
-	"math"
-
-	"model-gateway/consts/public"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/db/gfdb"
-	"github.com/gogf/gf/v2/frame/g"
-)
-
-// AutoTuneResult 单次调参结果（按 model_name）
-type AutoTuneResult struct {
-	ModelName string  `json:"modelName"`      // 模型名称（asynch_models.model_name）
-	Samples   int     `json:"samples"`        // 统计样本数（窗口内 state=2/3 且 started_at/finished_at 非空的任务数量）
-	P90Exec   float64 `json:"p90ExecSeconds"` // 执行耗时 P90（秒），口径：finished_at - started_at
-
-	CapMaxConcurrency int `json:"capMaxConcurrency"` // 配置上限：asynch_models.max_concurrency（cap，不会被动态调参覆盖）
-	OldMaxConcurrency int `json:"oldMaxConcurrency"` // 调参前运行时值（Redis），若无则等于 cap
-	NewMaxConcurrency int `json:"newMaxConcurrency"` // 本次计算出的运行时值（将写入 Redis），受 ±50% 约束且不超过 cap
-
-	CapQueueLimit int `json:"capQueueLimit"` // 配置上限：asynch_models.queue_limit（cap，不会被动态调参覆盖）
-	OldQueueLimit int `json:"oldQueueLimit"` // 调参前运行时值（Redis），若无则等于 cap
-	NewQueueLimit int `json:"newQueueLimit"` // 本次计算出的运行时值（将写入 Redis），受 ±50% 约束且不超过 cap
-
-	ExpectedSeconds int `json:"expectedSeconds"` // 模型预计执行时间（秒）：asynch_models.expected_seconds（用于 queue_limit 计算绑定）
-}
-
-// AutoTune 由上层定时任务通过接口触发：
-// - 统计指定时间窗口内该模型任务的执行耗时（finished_at - started_at，取 P90）
-// - 基于吞吐与 P90 执行耗时估算 max_concurrency 的运行时值（不超过 cap）
-// - queue_limit 与 expected_seconds 绑定（允许排队时间 = expected_seconds * 2），生成运行时值（不超过 cap）
-// - 单次调整幅度限制 ±50%，写入 Redis（带 TTL）
-func AutoTune(ctx context.Context, windowSeconds int) ([]AutoTuneResult, error) {
-	if windowSeconds <= 0 {
-		windowSeconds = 3600
-	}
-	// 1) 读取模型配置（cap），按 model_name 聚合去重（如果表里有多租户重复数据，取较大上限）
-	var modelRows []*entity.AsynchModel
-	if err := gfdb.DB(ctx).Model(ctx, public.TableNameModel).
-		Where("deleted_at IS NULL").
-		Where(entity.AsynchModelCol.Enabled, 1).
-		Scan(&modelRows); err != nil {
-		return nil, err
-	}
-	modelMap := make(map[string]*entity.AsynchModel)
-	for _, m := range modelRows {
-		if m == nil || m.ModelName == "" {
-			continue
-		}
-		cur := modelMap[m.ModelName]
-		if cur == nil {
-			modelMap[m.ModelName] = m
-			continue
-		}
-		// 取更大的 cap
-		if m.MaxConcurrency > cur.MaxConcurrency {
-			cur.MaxConcurrency = m.MaxConcurrency
-		}
-		if m.QueueLimit > cur.QueueLimit {
-			cur.QueueLimit = m.QueueLimit
-		}
-		if m.ExpectedSeconds > cur.ExpectedSeconds {
-			cur.ExpectedSeconds = m.ExpectedSeconds
-		}
-	}
-	if len(modelMap) == 0 {
-		return []AutoTuneResult{}, nil
-	}
-
-	// 2) 统计指定窗口：按 model_name 计算 cnt 和 P90 执行耗时
-	type statRow struct {
-		ModelName string
-		Cnt       int
-		P90Exec   float64
-	}
-	var stats []statRow
-	sql := fmt.Sprintf(`
-SELECT model_name,
-       COUNT(1) AS cnt,
-       COALESCE(percentile_cont(0.9) WITHIN GROUP (ORDER BY EXTRACT(EPOCH FROM (finished_at - started_at))), 0) AS p90_exec
-  FROM %s
- WHERE deleted_at IS NULL
-   AND state IN (2,3)
-   AND started_at IS NOT NULL
-   AND finished_at IS NOT NULL
-   AND finished_at >= (NOW() - (? || ' seconds')::interval)
- GROUP BY model_name`, public.TableNameTask)
-	r, err := gfdb.DB(ctx).GetAll(ctx, sql, windowSeconds)
-	if err != nil {
-		return nil, err
-	}
-	_ = r.Structs(&stats)
-	statMap := make(map[string]statRow, len(stats))
-	for _, s := range stats {
-		statMap[s.ModelName] = s
-	}
-
-	// 3) 调参计算
-	const utilization = 0.8
-	const maxChangeRatio = 0.5 // ±50%
-	const queueFactor = 2.0    // 与 expected_seconds 绑定：W_target = expected_seconds * 2
-
-	out := make([]AutoTuneResult, 0, len(modelMap))
-	for modelName, m := range modelMap {
-		s := statMap[modelName]
-		capMax := m.MaxConcurrency
-		capQueue := m.QueueLimit
-		oldMax := GetRuntimeMaxConcurrency(ctx, modelName, capMax)
-		oldQueue := GetRuntimeQueueLimit(ctx, modelName, capQueue)
-
-		// 默认：无样本则不调整
-		if s.Cnt <= 0 || s.P90Exec <= 0 {
-			out = append(out, AutoTuneResult{
-				ModelName:         modelName,
-				Samples:           s.Cnt,
-				P90Exec:           s.P90Exec,
-				CapMaxConcurrency: capMax,
-				OldMaxConcurrency: oldMax,
-				NewMaxConcurrency: oldMax,
-				CapQueueLimit:     capQueue,
-				OldQueueLimit:     oldQueue,
-				NewQueueLimit:     oldQueue,
-				ExpectedSeconds:   m.ExpectedSeconds,
-			})
-			continue
-		}
-
-		// arrival_rate ≈ 完成数/3600
-		arrivalRate := float64(s.Cnt) / 3600.0
-
-		// desiredMax = ceil(arrivalRate * p90 / utilization)
-		desiredMax := int(math.Ceil(arrivalRate * s.P90Exec / utilization))
-		if desiredMax < 1 {
-			desiredMax = 1
-		}
-		// 单次变化幅度限制
-		minMax := int(math.Floor(float64(oldMax) * (1 - maxChangeRatio)))
-		maxMax := int(math.Ceil(float64(oldMax) * (1 + maxChangeRatio)))
-		if minMax < 1 {
-			minMax = 1
-		}
-		newMax := clampInt(desiredMax, minMax, maxMax)
-		if capMax > 0 {
-			newMax = clampInt(newMax, 1, capMax)
-		}
-		setRuntimeInt(ctx, runtimeMaxConcurrencyKey(modelName), newMax)
-
-		// queue_limit：W_target = expected_seconds * queueFactor
-		exp := m.ExpectedSeconds
-		if exp <= 0 {
-			exp = 60
-		}
-		wTarget := float64(exp) * queueFactor
-		desiredQueue := int(math.Ceil(arrivalRate*wTarget)) + newMax
-		if desiredQueue < newMax {
-			desiredQueue = newMax
-		}
-
-		newQueue := oldQueue
-		if capQueue > 0 {
-			minQ := int(math.Floor(float64(oldQueue) * (1 - maxChangeRatio)))
-			maxQ := int(math.Ceil(float64(oldQueue) * (1 + maxChangeRatio)))
-			if minQ < newMax {
-				minQ = newMax
-			}
-			if maxQ < minQ {
-				maxQ = minQ
-			}
-			newQueue = clampInt(desiredQueue, minQ, maxQ)
-			newQueue = clampInt(newQueue, newMax, capQueue)
-			setRuntimeInt(ctx, runtimeQueueLimitKey(modelName), newQueue)
-		}
-
-		out = append(out, AutoTuneResult{
-			ModelName:         modelName,
-			Samples:           s.Cnt,
-			P90Exec:           s.P90Exec,
-			CapMaxConcurrency: capMax,
-			OldMaxConcurrency: oldMax,
-			NewMaxConcurrency: newMax,
-			CapQueueLimit:     capQueue,
-			OldQueueLimit:     oldQueue,
-			NewQueueLimit:     newQueue,
-			ExpectedSeconds:   m.ExpectedSeconds,
-		})
-	}
-
-	g.Log().Infof(ctx, "[auto_tune] done models=%d windowSeconds=%d", len(out), windowSeconds)
-	return out, nil
-}

service/callback.go

@@ -1,67 +0,0 @@
-package service
-
-import (
-	"context"
-	"encoding/json"
-
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/http"
-	"github.com/gogf/gf/v2/frame/g"
-)
-
-// triggerCallback 任务成功后的回调：
-// - JSON body 参数：task_id/state/oss_file/file_type/text（可选）
-func triggerCallback(ctx context.Context, t *entity.AsynchTask) {
-	callbackURL := t.BizName + t.CallbackURL
-	headers := forwardHeaders(ctx)
-	var req struct{}
-	payload := map[string]interface{}{
-		"task_id":   t.TaskID,
-		"state":     t.State,
-		"oss_file":  t.OssFile,
-		"file_type": t.FileType,
-		"text":      t.TextResult,
-		"error_msg": t.ErrorMsg,
-	}
-	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)
-		return
-	}
-	g.Log().Infof(ctx, "[回调] 开始发送 taskId=%s 回调地址=%s 请求头数量=%d 消息体大小=%d字节",
-		t.TaskID, callbackURL, len(headers), len(jsonData))
-
-	err = http.Post(ctx, callbackURL, headers, &req, jsonData)
-	if err != nil {
-		g.Log().Warningf(ctx, "[回调] 发送失败 taskId=%s 回调地址=%s 错误=%v", t.TaskID, callbackURL, err)
-		return
-	}
-	g.Log().Infof(ctx, "[回调] 发送成功 taskId=%s 回调地址=%s 消息体大小=%d字节", t.TaskID, callbackURL, len(jsonData))
-}
-
-// triggerPromptsCallback 任务成功后的提示词回调
-// - JSON body 参数：epicycleId（轮次id）/textResult（模型回答消息）
-func triggerPromptsCallback(ctx context.Context, t *entity.AsynchTask, epicycleId int64) {
-	callbackURL := "prompts-core/session/sessionCallback"
-	headers := forwardHeaders(ctx)
-	var req struct{}
-	payload := map[string]interface{}{
-		"epicycleId": epicycleId,
-		"text":       t.TextResult,
-	}
-	jsonData, err := json.Marshal(payload)
-	if err != nil {
-		g.Log().Warningf(ctx, "[提示词回调] JSON序列化失败 epicycleId=%d 错误=%v", epicycleId, err)
-		return
-	}
-	g.Log().Infof(ctx, "[提示词回调] 开始发送 epicycleId=%d 回调地址=%s 请求头数量=%d 消息体大小=%d字节",
-		t.EpicycleId, callbackURL, len(headers), len(jsonData))
-
-	err = http.Post(ctx, callbackURL, headers, &req, jsonData)
-	if err != nil {
-		g.Log().Warningf(ctx, "[提示词回调] 发送失败 epicycleId=%d 回调地址=%s 错误=%v", t.EpicycleId, callbackURL, err)
-		return
-	}
-	g.Log().Infof(ctx, "[提示词回调] 发送成功 epicycleId=%d 回调地址=%s 消息体大小=%d字节", t.EpicycleId, callbackURL, len(jsonData))
-}

service/cleaner.go

@@ -1,92 +0,0 @@
-package service
-
-import (
-	"context"
-	"time"
-
-	"model-gateway/dao"
-
-	"github.com/gogf/gf/v2/frame/g"
-)
-
-var Cleaner = &cleaner{}
-
-type cleaner struct{}
-
-// RunOnce 由上层定时任务触发：执行一次清理/重试
-func (c *cleaner) RunOnce(ctx context.Context) {
-	// 1) 清理已下载(state=4)且过期的任务（硬删除 + OSS）
-	expired, err := dao.Task.ListExpiredDownloadedGlobal(ctx, 200)
-	if err != nil {
-		g.Log().Errorf(ctx, "[cleaner] list expired(downloaded) error: %v", err)
-	} else {
-		for _, t := range expired {
-			deleteTmpResult(t.TmpFile)
-			_ = dao.Task.HardDeleteByIDGlobal(ctx, t.Id)
-		}
-		g.Log().Infof(ctx, "[cleaner] expired(downloaded) cleaned, count=%d", len(expired))
-	}
-
-	// 2) 超时任务标失败
-	list, err := dao.Task.ListTimeoutTasksGlobal(ctx, 200)
-	if err != nil {
-		g.Log().Errorf(ctx, "[cleaner] list timeout error: %v", err)
-	} else {
-		for _, t := range list {
-			_ = dao.Task.UpdateFailedGlobal(ctx, t.Id, "任务超时自动失败")
-			ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-		}
-		g.Log().Infof(ctx, "[cleaner] timeout cleaned, count=%d", len(list))
-	}
-
-	// 3) 失败(state=3)的任务按模型配置 retry_times 重新入队（放到队尾）
-	retryable, err := dao.Task.ListFailedRetryableGlobal(ctx, 200)
-	if err != nil {
-		g.Log().Errorf(ctx, "[cleaner] list failed retryable error: %v", err)
-	} else {
-		for _, t := range retryable {
-			// 失败任务重新入队（state=3 -> 0）前，先严格占用 queue_limit slot；占用失败则留在失败态，下一轮再尝试
-			// 获取模型配置以得到 queue_limit / expected_seconds
-			m, err := dao.Model.GetByModelNameForTenant(ctx, t.TenantId, t.ModelName)
-			if err != nil || m == nil {
-				continue
-			}
-			limit := GetRuntimeQueueLimit(ctx, t.ModelName, m.QueueLimit)
-			if limit > 0 {
-				ok, _ := AcquireQueueSlot(ctx, t.ModelName, t.TaskID, limit, m.ExpectedSeconds)
-				if !ok {
-					continue
-				}
-			}
-			// retry_queue_max_seconds 控制失败重试的排队策略：
-			// - =0：失败重试插队到队首
-			// - >0：当任务从创建到现在的排队时长 >= maxSeconds，则插队到队首；否则仍放到队尾
-			now := time.Now()
-			enqueueAt := now
-			maxSeconds := t.RetryQueueMaxSeconds
-			if maxSeconds == 0 {
-				enqueueAt = now.Add(-100 * 365 * 24 * time.Hour)
-			} else if maxSeconds > 0 && t.CreatedAt != nil {
-				if now.Sub(t.CreatedAt.Time) >= time.Duration(maxSeconds)*time.Second {
-					enqueueAt = now.Add(-100 * 365 * 24 * time.Hour)
-				}
-			}
-			_ = dao.Task.RequeueForRetryGlobal(ctx, t.Id, enqueueAt)
-		}
-		g.Log().Infof(ctx, "[cleaner] failed retryable cleaned, count=%d", len(retryable))
-	}
-
-	// 4) 超过重试次数仍失败(state=3)的任务：硬删除
-	exhausted, err := dao.Task.ListFailedExhaustedGlobal(ctx, 200)
-	if err != nil {
-		g.Log().Errorf(ctx, "[cleaner] list failed exhausted error: %v", err)
-	} else {
-		for _, t := range exhausted {
-			deleteTmpResult(t.TmpFile)
-			// 重试耗尽硬删除：释放闸门占位（兜底，若此前已释放则幂等）
-			ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-			_ = dao.Task.HardDeleteByIDGlobal(ctx, t.Id)
-		}
-		g.Log().Infof(ctx, "[cleaner] failed exhausted cleaned, count=%d", len(exhausted))
-	}
-}

service/error_analysis.go

@@ -0,0 +1,153 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// parseAnalysisResponse 解析分析模型输出的判定 JSON。
+// 容错：剥 ```json 代码块/首尾空白/多余文字，取首个 {...}。
+func parseAnalysisResponse(content string) (retryable bool, reason string, err error) {
+	s := strings.TrimSpace(content)
+	s = strings.TrimPrefix(s, "```json")
+	s = strings.TrimSuffix(s, "```")
+	s = strings.TrimSpace(s)
+	start, end := strings.IndexByte(s, '{'), strings.LastIndexByte(s, '}')
+	if start < 0 || end <= start {
+		return false, "", fmt.Errorf("分析响应中未找到JSON对象: %q", content)
+	}
+	var obj struct {
+		Retryable bool   `json:"retryable"`
+		Reason    string `json:"reason"`
+	}
+	if err = json.Unmarshal([]byte(s[start:end+1]), &obj); err != nil {
+		return false, "", fmt.Errorf("分析响应JSON解析失败: %v", err)
+	}
+	return obj.Retryable, obj.Reason, nil
+}
+
+const (
+	analysisTimeout   = 15 * time.Second
+	analysisMaxBody   = 2000
+	analysisMaxTokens = 256
+)
+
+const analysisSystemPrompt = `你是 AI 模型网关的错误分析器。上游 AI 模型调用返回了一个错误，你需要判断该错误是否"值得指数退避后重试"。
+
+## 值得重试（retryable: true）
+- 限流：429、rate limit、请求过密、并发超限
+- 服务端瞬时故障：5xx、InternalServiceError、服务过载、上游临时不可用
+- 超时/取消/连接：Timeout、RequestCanceled、Error while connecting、连接抖动
+- 媒体源暂不可用（视频/音频生成类上游常见）：Invalid video_url、Invalid audio track、Error while downloading、download failed —— 通常是源尚未就绪或下载瞬断，重试可成功，不要误判为永久参数错误
+- 资源暂时不足：quota 暂时受限
+
+## 不值得重试（retryable: false）
+- 请求/参数错误：400、invalid_argument、格式错误（注意 Invalid video_url / Invalid audio track 属上类的媒体源错误，不归此类）
+- 鉴权失败：401、403、invalid_api_key、签名错误
+- 模型不存在：404、model_not_found
+- 余额不足：insufficient_quota
+- 内容违规：内容安全拦截
+- 明确的永久性配置错误
+
+## 输出
+只输出一个 JSON 对象，不要任何多余文字、解释或代码块标记：
+{"retryable": true 或 false, "reason": "不超过20字的简要原因"}`
+
+// truncateStr 按字节截断到 max（中文可能截半个字符，仅用于分析输入，可接受）
+func truncateStr(s string, max int) string {
+	if len(s) <= max {
+		return s
+	}
+	return s[:max]
+}
+
+// buildAnalysisBody 构造分析请求体（OpenAI 兼容 messages 格式），纯函数便于单测。
+func buildAnalysisBody(ctx context.Context, modelName, code, msg, body string) map[string]any {
+	user := fmt.Sprintf("错误码: %s\n错误消息: %s\n错误响应体: %s", code, msg, truncateStr(body, analysisMaxBody))
+	// 打印错误信息
+	g.Log().Debugf(ctx, "分析请求体: %s", user)
+
+	return map[string]any{
+		"model": modelName,
+		"messages": []map[string]string{
+			{"role": "system", "content": analysisSystemPrompt},
+			{"role": "user", "content": user},
+		},
+		"max_tokens":  analysisMaxTokens,
+		"temperature": 0,
+	}
+}
+
+// resolveAnalysisModel 选择分析模型：失败模型自身是对话模型则复用，否则取当前用户对话模型。
+// 取不到返回 ok=false，调用方 fail-closed 不重试。
+func resolveAnalysisModel(ctx context.Context, modelInfo *entity.ModelManage) (model *entity.ModelManage, ok bool) {
+	if modelInfo != nil && modelInfo.ChatModel != nil && *modelInfo.ChatModel {
+		return modelInfo, true
+	}
+	chat, err := ModelManage.GetChatModel(ctx, &dto.GetChatModelReq{})
+	if err == nil && chat != nil && chat.ModelManage != nil {
+		return chat.ModelManage, true
+	}
+	return nil, false
+}
+
+// callAnalysisLLM 调分析模型（对话模型）判定错误是否可重试。
+// 独立短超时 http.Client；非 200 / 解析失败 / 超时 → 返回 err，调用方 fail-closed。
+func callAnalysisLLM(ctx context.Context, model *entity.ModelManage, code, msg, body string) (retryable bool, reason string, err error) {
+	reqBody, err := json.Marshal(buildAnalysisBody(ctx, model.ModelName, code, msg, body))
+	if err != nil {
+		return false, "", fmt.Errorf("marshal分析请求失败: %w", err)
+	}
+	httpMethod := model.HttpMethod
+	if httpMethod == "" {
+		httpMethod = http.MethodPost
+	}
+	httpReq, err := http.NewRequestWithContext(ctx, httpMethod, strings.TrimRight(model.BaseURL, "/"), bytes.NewBuffer(reqBody))
+	if err != nil {
+		return false, "", fmt.Errorf("创建分析请求失败: %w", err)
+	}
+	for k, v := range model.RequestHeadMapping {
+		httpReq.Header.Set(k, v)
+	}
+	httpReq.Header.Set("Authorization", "Bearer "+model.ApiKey)
+	httpReq.Header.Set("Content-Type", "application/json")
+
+	client := &http.Client{Timeout: analysisTimeout}
+	resp, err := client.Do(httpReq)
+	if err != nil {
+		return false, "", fmt.Errorf("分析请求失败: %w", err)
+	}
+	defer resp.Body.Close()
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return false, "", fmt.Errorf("读取分析响应失败: %w", err)
+	}
+	if resp.StatusCode != http.StatusOK {
+		return false, "", fmt.Errorf("分析接口非200: status=%d body=%s", resp.StatusCode, truncateStr(string(respBody), 500))
+	}
+	var apiResp struct {
+		Choices []struct {
+			Message struct {
+				Content string `json:"content"`
+			} `json:"message"`
+		} `json:"choices"`
+	}
+	if err = json.Unmarshal(respBody, &apiResp); err != nil {
+		return false, "", fmt.Errorf("解析分析响应失败: %w", err)
+	}
+	if len(apiResp.Choices) == 0 {
+		return false, "", fmt.Errorf("分析响应无choices")
+	}
+	return parseAnalysisResponse(apiResp.Choices[0].Message.Content)
+}

service/error_memory.go

@@ -0,0 +1,43 @@
+package service
+
+import (
+	"crypto/md5"
+	"crypto/sha256"
+	"encoding/hex"
+	"regexp"
+	"strings"
+)
+
+var (
+	reUUID      = regexp.MustCompile(`[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}`)
+	reISOTime   = regexp.MustCompile(`\d{4}-\d{2}-\d{2}[T ]\d{2}:\d{2}:\d{2}(\.\d+)?(Z|[+-]\d{2}:?\d{2})?`)
+	reUnixMs    = regexp.MustCompile(`\b1[4-9]\d{12}\b`)                     // unix 毫秒级时间戳
+	reRequestID = regexp.MustCompile(`\b(req[-_]?|request[-_]?|rid[-_:]?)[-_:]?[0-9a-zA-Z-]{4,}\b`)
+	reLongNum   = regexp.MustCompile(`\b\d{4,}\b`)                           // 连续≥4位数字
+)
+
+// normalizeErrorMsg 归一化错误消息：剔除易变片段（UUID/时间戳/请求ID/连续数字），
+// 使同因不同实例的错误命中同一记忆键。
+func normalizeErrorMsg(msg string) string {
+	m := msg
+	m = reUUID.ReplaceAllString(m, "{uuid}")
+	m = reISOTime.ReplaceAllString(m, "{time}")
+	m = reUnixMs.ReplaceAllString(m, "{ts}")
+	m = reRequestID.ReplaceAllString(m, "{reqid}")
+	m = reLongNum.ReplaceAllString(m, "{num}")
+	return strings.TrimSpace(m)
+}
+
+// buildMemoryKey 构造记忆键 = SHA-256(upstream|error_code|归一化消息)。
+// 含失败上游维度：不同上游的同类错误互不串扰。
+func buildMemoryKey(upstream, code, msg string) string {
+	raw := strings.Join([]string{upstream, code, normalizeErrorMsg(msg)}, "|")
+	sum := sha256.Sum256([]byte(raw))
+	return hex.EncodeToString(sum[:])
+}
+
+// msgFingerprint 归一化消息的 md5（观测/展示用）。
+func msgFingerprint(msg string) string {
+	sum := md5.Sum([]byte(normalizeErrorMsg(msg)))
+	return hex.EncodeToString(sum[:])
+}

service/file_detect.go

@@ -1,47 +0,0 @@
-package service
-
-import (
-	"net/http"
-	"strings"
-)
-
-// DetectFileType 根据返回的二进制内容推断 contentType + 扩展名（尽量稳定）
-func DetectFileType(data []byte) (contentType string, ext string) {
-	if len(data) == 0 {
-		return "application/octet-stream", ""
-	}
-	ct := http.DetectContentType(data)
-	// http.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 "video/mp4":
-		return ct, ".mp4"
-	case "image/png":
-		return ct, ".png"
-	case "image/jpeg":
-		return ct, ".jpg"
-	case "application/pdf":
-		return ct, ".pdf"
-	case "text/plain":
-		return ct, ".txt"
-	case "application/json":
-		return ct, ".json"
-	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, ""
-	}
-}

service/headers.go

@@ -1,53 +0,0 @@
-package service
-
-import (
-	"context"
-
-	"gitea.com/red-future/common/utils"
-	"github.com/gogf/gf/v2/frame/g"
-)
-
-// asyncCtx 固化异步执行所需的 token/user，避免请求结束后丢失（仅在“同请求内起 goroutine”有用）。
-// 本项目当前是“落库 + 后台 worker”模式，因此还会把必要信息持久化到任务表的 request_payload 中。
-func asyncCtx(ctx context.Context) context.Context {
-	asyncCtx := context.WithoutCancel(ctx)
-	if r := g.RequestFromCtx(ctx); r != nil {
-		if token := r.Header.Get("Authorization"); token != "" {
-			asyncCtx = context.WithValue(asyncCtx, "token", token)
-		}
-		if userInfo := r.Header.Get("X-User-Info"); userInfo != "" {
-			asyncCtx = context.WithValue(asyncCtx, "xUserInfo", userInfo)
-		}
-	}
-	if user, err := utils.GetUserInfo(ctx); err == nil && user != nil {
-		asyncCtx = context.WithValue(asyncCtx, "user", user)
-	}
-	return asyncCtx
-}
-
-// forwardHeaders 透传调用链路中必须的头信息（优先使用 ctx 里固化的 token / xUserInfo）。
-func forwardHeaders(ctx context.Context) map[string]string {
-	headers := make(map[string]string)
-
-	if token, ok := ctx.Value("token").(string); ok && token != "" {
-		headers["Authorization"] = token
-	}
-	if x, ok := ctx.Value("xUserInfo").(string); ok && x != "" {
-		headers["X-User-Info"] = x
-	}
-
-	// 兜底：从请求头拿
-	if r := g.RequestFromCtx(ctx); r != nil {
-		if headers["Authorization"] == "" {
-			if token := r.Header.Get("Authorization"); token != "" {
-				headers["Authorization"] = token
-			}
-		}
-		if headers["X-User-Info"] == "" {
-			if userInfo := r.Header.Get("X-User-Info"); userInfo != "" {
-				headers["X-User-Info"] = userInfo
-			}
-		}
-	}
-	return headers
-}

service/httpclient/httpclient.go

@@ -0,0 +1,167 @@
+// Package httpclient 模型网关的传输层：模型 HTTP 请求（含瞬时网络错误重试）与 SSE 流式解析。
+// 纯基础设施，不依赖 session/task/call 等业务逻辑；业务代码只通过三个导出函数使用。
+package httpclient
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	commonHttp "gitea.redpowerfuture.com/red-future/common/http"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/net/gclient"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// modelCallHeaderTimeout 模型响应头等待超时。
+// commonHttp 底层 gclient 默认 ResponseHeaderTimeout 只有 30s，模型生成首字节
+// （尤其非流式、大 max_tokens）经常超过 30s，导致 http2: timeout awaiting response
+// headers。模型调用必须用独立 client 并把该超时调大，与模型配置的超时保持一致。
+const modelCallHeaderTimeout = 30 * time.Minute
+
+// modelHTTPClient 构建模型调用专用 HTTP client：
+// 克隆 commonHttp 客户端（保留 ContentJson、header 注入等行为），但把
+// ResponseHeaderTimeout 从默认 30s 调大到 modelCallHeaderTimeout。
+func modelHTTPClient() *gclient.Client {
+	client := commonHttp.Httpclient.Clone()
+	if tr, ok := client.Transport.(*http.Transport); ok {
+		tr = tr.Clone() // 独立拷贝，避免改动全局共享 transport
+		tr.ResponseHeaderTimeout = modelCallHeaderTimeout
+		client.Transport = tr
+	}
+	return client
+}
+
+// modelNetRetryTimes 模型请求瞬时网络错误最大重试次数（不含首次）；modelNetRetryBackoff 为退避基数。
+// 模型域名 DNS 解析失败（Docker 内 127.0.0.11 偶发 no such host）是瞬时错误，短退避重试即可恢复。
+// 重试在 HTTP 层完成，覆盖同步/异步/流式全部调用路径；流式场景发生在写 SSE 响应头之前，重试安全。
+const (
+	modelNetRetryTimes   = 3
+	modelNetRetryBackoff = 500 * time.Millisecond
+)
+
+// isTransientNetError 判定是否可重试的瞬时网络错误。仅命中 DNS 解析失败（no such host）：
+// 模型域名解析抖动可重试恢复；连接拒绝/超时等其他网络错误可能反映真实配置问题，不纳入，避免掩盖错误。
+func isTransientNetError(err error) bool {
+	if err == nil {
+		return false
+	}
+	return strings.Contains(err.Error(), "no such host") || strings.Contains(err.Error(), "timeout")
+}
+
+// modelDoRaw 模型 HTTP 请求（等价 commonHttp.doRequestRaw，但使用调大超时的 client）。
+// DNS 解析失败等瞬时网络错误在请求层短退避重试（modelNetRetryTimes 次）；
+// 其余错误（含上游业务错误码）原样返回，由上层按错误码决定是否重试。
+func modelDoRaw(ctx context.Context, method string, url string, headers map[string]string, data ...any) (*gclient.Response, error) {
+	client := modelHTTPClient()
+
+	if (method == http.MethodPost || method == http.MethodPut || method == http.MethodDelete) && len(data) > 0 {
+		client = client.ContentJson()
+	}
+
+	if len(headers) > 0 {
+		client.SetHeaderMap(headers)
+	} else if r := g.RequestFromCtx(ctx); r != nil {
+		client.SetHeader("Authorization", r.Request.Header.Get("Authorization"))
+	}
+
+	doOnce := func() (*gclient.Response, error) {
+		if method == http.MethodGet && len(data) > 0 && len(data)%2 == 0 {
+			queryParams := make(map[string]string)
+			for i := 0; i < len(data); i += 2 {
+				if key, ok := data[i].(string); ok && i+1 < len(data) {
+					queryParams[key] = gconv.String(data[i+1])
+				}
+			}
+			return client.DoRequest(ctx, method, url, queryParams)
+		}
+		if len(data) == 1 {
+			return client.DoRequest(ctx, method, url, data[0])
+		}
+		return client.DoRequest(ctx, method, url, data...)
+	}
+
+	var response *gclient.Response
+	var err error
+	for attempt := 0; ; attempt++ {
+		response, err = doOnce()
+		if err == nil || !isTransientNetError(err) {
+			return response, err
+		}
+		if attempt >= modelNetRetryTimes {
+			break
+		}
+		wait := time.Duration(1<<attempt) * modelNetRetryBackoff
+		g.Log().Warningf(ctx, "[HttpModel] 模型请求瞬时网络错误，第 %d/%d 次重试(等待 %v): %v", attempt+1, modelNetRetryTimes, wait, err)
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-time.After(wait):
+		}
+	}
+	return response, err
+}
+
+// ModelHttpNormalRequest 同步/异步 普通HTTP全量请求
+func ModelHttpNormalRequest(ctx context.Context, url string, headers map[string]string, httpMethod string, body map[string]any) (res []byte, err error) {
+	response, err := modelDoRaw(ctx, httpMethod, url, headers, body)
+	if err != nil {
+		g.Log().Errorf(ctx, "[HttpModel] 模型请求失败 [Error]: %v", err)
+		return nil, fmt.Errorf("模型请求失败: %w", err)
+	}
+	defer response.Close()
+	return response.ReadAll(), nil
+}
+
+// ModelHttpStreamRequest 通用流式请求
+// stream=true 时设置 SSE 头并验证 Flusher；stream=false 时只返回 Reader，不设置响应头
+func ModelHttpStreamRequest(ctx context.Context, w http.ResponseWriter, url string, headers map[string]string, httpMethod string, body map[string]any) (io.Reader, error) {
+	// 1) 先发起上游请求（此时还没写任何 SSE 头，失败可以正常返回 error）
+	response, err := modelDoRaw(ctx, httpMethod, url, headers, body)
+	if err != nil {
+		g.Log().Errorf(ctx, "[HttpModel] 模型流式请求失败 [Error]: %v", err)
+		return nil, fmt.Errorf("模型流式请求失败: %w", err)
+	}
+
+	// 检查 HTTP 状态码
+	if response.StatusCode < 200 || response.StatusCode >= 300 {
+		bodyBytes, _ := io.ReadAll(response.Body)
+		response.Close()
+		return nil, fmt.Errorf("[HTTP][Stream] 状态码异常: %d, body=%s", response.StatusCode, string(bodyBytes))
+	}
+
+	if w != nil {
+		// 2) 上游连接成功，再设置 SSE 头
+		h := w.Header()
+		h.Set("Content-Type", "text/event-stream; charset=utf-8")
+		h.Set("Cache-Control", "no-cache")
+		h.Set("Connection", "keep-alive")
+		h.Set("X-Accel-Buffering", "no")
+
+		if _, ok := w.(http.Flusher); !ok {
+			response.Close()
+			return nil, errors.New("response writer not support flush")
+		}
+	}
+
+	// 下层统一托管关闭：用包装器保证流最终关闭
+	return &autoCloseReader{r: response.Body}, nil
+}
+
+// autoCloseReader 包装 io.ReadCloser，读取结束/销毁时自动 Close
+type autoCloseReader struct {
+	r io.ReadCloser
+}
+
+func (a *autoCloseReader) Read(p []byte) (int, error) {
+	n, err := a.r.Read(p)
+	// 读取完毕 / 读出错，主动关闭流
+	if err != nil {
+		_ = a.r.Close()
+	}
+	return n, err
+}

service/httpclient/sse.go

@@ -0,0 +1,85 @@
+package httpclient
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"io"
+	"strings"
+
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// SSE 常量
+const (
+	ssePrefixData    = "data:"
+	ssePrefixEvent   = "event:"
+	ssePrefixComment = ":"
+	sseStreamDone    = "[DONE]"
+
+	scanBufInitSize = 64 * 1024   // 64KB
+	scanMaxLineSize = 1024 * 1024 // 单行最大 1MB
+)
+
+// ParseSSEStream 标准 SSE 流式解析，逐分片回调，支持多行data、上下文取消
+func ParseSSEStream(ctx context.Context, respBody io.Reader, onChunk func(ctx context.Context, chunk map[string]any) error) {
+	scanner := bufio.NewScanner(respBody)
+	scanner.Buffer(make([]byte, 0, scanBufInitSize), scanMaxLineSize)
+
+	var dataBuilder strings.Builder
+
+	for scanner.Scan() {
+		// 监听上下文取消，及时终止
+		select {
+		case <-ctx.Done():
+			g.Log().Infof(ctx, "[SSE] 上下文取消，终止流读取: %v", ctx.Err())
+			return
+		default:
+		}
+
+		line := scanner.Text()
+
+		// 跳过注释、事件行
+		if strings.HasPrefix(line, ssePrefixComment) || strings.HasPrefix(line, ssePrefixEvent) {
+			continue
+		}
+
+		lineTrim := strings.TrimSpace(line)
+		// 空行 = 一个SSE事件结束
+		if lineTrim == "" {
+			if dataBuilder.Len() == 0 {
+				continue
+			}
+			dataStr := dataBuilder.String()
+			dataBuilder.Reset()
+
+			if dataStr == sseStreamDone {
+				continue
+			}
+
+			var chunk map[string]any
+			if err := json.Unmarshal([]byte(dataStr), &chunk); err != nil {
+				g.Log().Debugf(ctx, "[SSE] JSON解析失败: %s, err: %v", dataStr, err)
+				continue
+			}
+
+			if onChunk != nil {
+				onChunk(ctx, chunk)
+			}
+			continue
+		}
+
+		// 拼接多行 data 数据
+		if strings.HasPrefix(line, ssePrefixData) {
+			raw := strings.TrimPrefix(line, ssePrefixData)
+			dataBuilder.WriteString(strings.TrimSpace(raw))
+		}
+	}
+
+	// 捕获读取异常
+	if err := scanner.Err(); err != nil {
+		g.Log().Errorf(ctx, "[SSE] 流读取异常: %v", err)
+		return
+	}
+	g.Log().Infof(ctx, "[SSE] 流式读取正常结束")
+}

service/model_call_service.go

@@ -0,0 +1,364 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"model-gateway/consts/model"
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+	modelUtils "model-gateway/service/utils"
+	"net/http"
+	"time"
+
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/os/glog"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelCall = &modelCallService{}
+
+type modelCallService struct{}
+
+func (s *modelCallService) ModelCall(ctx context.Context, req *dto.ModelCallReq) (res *dto.ModelCallRes, err error) {
+	// 1) 检查模型配置
+	var modelInfo *entity.ModelManage
+	modelInfo, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{
+		Id: req.ModelId,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("获取模型配置失败: %v", err)
+	}
+	if modelInfo == nil {
+		return nil, fmt.Errorf("模型不存在")
+	}
+	// 引用行 → 解析为系统模型配置+本人 apiKey；系统模型已删除等解析失败 → 阻塞调用
+	modelInfo, err = modelUtils.ResolveModelConfig(ctx, modelInfo)
+	if err != nil {
+		return nil, err
+	}
+	if modelInfo.Enabled != nil && !*modelInfo.Enabled {
+		return nil, fmt.Errorf("模型不存在或未启用")
+	}
+
+	now := time.Now()
+	userInfo, err := utils.GetUserInfo(ctx)
+	if err != nil {
+		return
+	}
+	if !g.IsEmpty(modelInfo.RefSystemModelId) {
+		// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
+		if err = modelBillable(ctx, modelInfo.Id); err != nil {
+			return nil, err
+		}
+	}
+	err = queue(ctx, modelInfo.ModelName, userInfo.TenantId, gconv.Int64(modelInfo.MaxConcurrency), func(ctx context.Context) (err error) {
+		if *modelInfo.ResponseType == *model.ResponseTypeSync.Code() || *modelInfo.ResponseType == *model.ResponseTypeStream.Code() {
+			var newRequestParams map[string]any
+			var id int64
+			id, newRequestParams, err = s.saveModelRequestParams(ctx, now, modelInfo, req)
+			if err != nil {
+				return err
+			}
+			if g.IsEmpty(id) || g.IsEmpty(newRequestParams) {
+				return fmt.Errorf("保存模型请求参数失败")
+			}
+			if *modelInfo.ResponseType == *model.ResponseTypeSync.Code() {
+				res, err = ModelSession.CreateSession(ctx, &dto.CallModelSessionReq{
+					Id:            id,
+					ModelInfo:     modelInfo,
+					RequestParams: newRequestParams,
+				})
+			} else {
+				res, err = ModelSession.CreateSessionStreamOnce(ctx, &dto.CallModelSessionReq{
+					Id:            id,
+					ModelInfo:     modelInfo,
+					RequestParams: newRequestParams,
+				})
+			}
+		}
+		if *modelInfo.ResponseType == *model.ResponseTypeAsync.Code() {
+			if g.IsEmpty(req.MsgTopic) {
+				return fmt.Errorf("请指定消息主题")
+			}
+			var newRequestParams map[string]any
+			var id int64
+			id, newRequestParams, err = s.saveModelRequestParams(ctx, now, modelInfo, req)
+			if err != nil {
+				return err
+			}
+			if g.IsEmpty(id) || g.IsEmpty(newRequestParams) {
+				return fmt.Errorf("保存模型请求参数失败")
+			}
+			res, err = ModelTaskStart.CreateTask(ctx, &dto.CallModelTaskStartReq{
+				Id:            id,
+				ModelInfo:     modelInfo,
+				RequestParams: newRequestParams,
+			})
+		}
+		return
+	})
+	return
+}
+
+func (s *modelCallService) ModelCallStream(ctx context.Context, w http.ResponseWriter, req *dto.ModelCallStreamReq) (err error) {
+	// 1) 检查模型配置
+	var modelInfo *entity.ModelManage
+	modelInfo, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{
+		Id: req.ModelId,
+	})
+	if err != nil {
+		return fmt.Errorf("获取模型配置失败: %v", err)
+	}
+	if modelInfo == nil {
+		return fmt.Errorf("模型不存在")
+	}
+	// 引用行 → 解析为系统模型配置+本人 apiKey；系统模型已删除等解析失败 → 阻塞调用
+	modelInfo, err = modelUtils.ResolveModelConfig(ctx, modelInfo)
+	if err != nil {
+		return err
+	}
+	if modelInfo.Enabled != nil && !*modelInfo.Enabled {
+		return fmt.Errorf("模型不存在或未启用")
+	}
+	if *modelInfo.ResponseType == *model.ResponseTypeStream.Code() {
+		now := time.Now()
+		userInfo, err := utils.GetUserInfo(ctx)
+		if err != nil {
+			return err
+		}
+		if !g.IsEmpty(modelInfo.RefSystemModelId) {
+			// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
+			if err = modelBillable(ctx, modelInfo.Id); err != nil {
+				return err
+			}
+		}
+		err = queue(ctx, modelInfo.ModelName, userInfo.TenantId, gconv.Int64(modelInfo.MaxConcurrency), func(ctx context.Context) (err error) {
+			var newRequestParams map[string]any
+			var id int64
+			id, newRequestParams, err = s.saveModelRequestParams(ctx, now, modelInfo, &dto.ModelCallReq{
+				ModelId:        req.ModelId,
+				RequestParams:  req.RequestParams,
+				BusinessParams: req.BusinessParams,
+				SessionId:      req.SessionId,
+				BizName:        req.BizName,
+			})
+			if err != nil {
+				return err
+			}
+			if g.IsEmpty(id) || g.IsEmpty(newRequestParams) {
+				return fmt.Errorf("保存模型请求参数失败")
+			}
+			_, err = ModelSession.CreateSessionStream(ctx, w, &dto.CallModelSessionReq{
+				Id:            id,
+				ModelInfo:     modelInfo,
+				RequestParams: newRequestParams,
+			})
+			return
+		})
+	} else {
+		return fmt.Errorf("模型响应类型错误")
+	}
+	return err
+}
+
+// saveModelRequestParams 保存模型请求参数
+func (s *modelCallService) saveModelRequestParams(ctx context.Context, now time.Time, modelInfo *entity.ModelManage, req *dto.ModelCallReq) (id int64, newRequestParams map[string]any, err error) {
+
+	// 统一走模板校验+构建：requestParams 只装模板字段，businessParams 只装业务字段
+	out, err := buildChatRequestParams(modelInfo, req.RequestParams, req.BusinessParams)
+	if err != nil {
+		return 0, nil, err
+	}
+
+	// 1) 上传模型原始请求参数文件（requestParams + businessParams 合并，保证审计完整）
+	originalParams := make(map[string]any, len(req.RequestParams)+len(req.BusinessParams))
+	for k, v := range req.RequestParams {
+		originalParams[k] = v
+	}
+	for k, v := range req.BusinessParams {
+		originalParams[k] = v
+	}
+	uploadOriginalReq, err := Upload(ctx, &dto.UploadFileBytesReq{
+		FileBytes: gconv.Bytes(gconv.String(originalParams)),
+		FileName:  fmt.Sprintf("modelRequestParams:%v.json", now.UnixMilli()),
+	})
+	if err != nil {
+		return 0, nil, fmt.Errorf("上传模型原始请求参数文件失败: %v", err)
+	}
+
+	// 2) 上传模型解析成功的请求参数文件
+	uploadNewReq, err := Upload(ctx, &dto.UploadFileBytesReq{
+		FileBytes: gconv.Bytes(gconv.String(out)),
+		FileName:  fmt.Sprintf("modelNewRequestParams:%v.json", now.UnixMilli()),
+	})
+	if err != nil {
+		return 0, nil, fmt.Errorf("上传模型解析请求参数文件失败：%v", err)
+	}
+
+	// 3) 保存模型请求信息（快照媒体类型=shop 计费词汇 audio/video，空=无媒体引用，任务完成时直接用于算费；模型计费配置任务完成时按 modelId 现查）
+	if *modelInfo.ResponseType == *model.ResponseTypeAsync.Code() {
+		id, err = dao.ModelTaskStart.Insert(ctx, &dto.CreateModelTaskStartReq{
+			ModelId:             req.ModelId,
+			BizName:             req.BizName,
+			MsgTopic:            req.MsgTopic,
+			RequestPath:         uploadNewReq.FileURL,
+			OriginalRequestPath: uploadOriginalReq.FileURL,
+			MediaType:           modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, out),
+		})
+		if err != nil {
+			return 0, nil, fmt.Errorf("保存模型请求信息失败: %v", err)
+		}
+	} else {
+		id, err = dao.ModelSession.Insert(ctx, &dto.CreateModelSessionReq{
+			ModelId:             req.ModelId,
+			BizName:             req.BizName,
+			SessionId:           req.SessionId,
+			RequestPath:         uploadNewReq.FileURL,
+			OriginalRequestPath: uploadOriginalReq.FileURL,
+		})
+		if err != nil {
+			return 0, nil, fmt.Errorf("保存模型请求信息失败: %v", err)
+		}
+	}
+	return id, out, nil
+}
+
+func queue(ctx context.Context, modelName string, tenantId uint64, maxCon int64, f func(ctx context.Context) (err error)) (err error) {
+	const (
+		keyExpireSec = 600              // 名额Key兜底过期时间 10min（进程崩溃后自愈）
+		refreshStep  = keyExpireSec / 3 // 执行期间续期间隔
+		waitInterval = 10 * time.Second // 超限轮询等待间隔
+	)
+	// Redis 操作统一使用独立上下文，避免外部 ctx canceled
+	redisCtx := context.WithoutCancel(ctx)
+	concurrencyKey := fmt.Sprintf("model:concurrency:%d:%s", tenantId, modelName)
+
+	// 1) 原子占用并发名额：utils.SemaphoreAcquire 在 WATCH 事务内完成 判满→INCR→首设EXPIRE→超限不写，
+	//    不再需要旧 reserveSlot 的「分布式锁 + Incr + 回滚」组合（组合已事务化，外层锁冗余）。
+	//    超限（false）按 waitInterval 轮询重试；max<=0 视为不限制（SemaphoreAcquire 内部直接放行）。
+	for {
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+		ok, e := utils.SemaphoreAcquire(redisCtx, concurrencyKey, int(maxCon), keyExpireSec)
+		if e != nil {
+			return e
+		}
+		if ok {
+			// 展示当前并发数（占用后 GET，与旧 reserveSlot 的 Incr 后计数值语义一致）
+			if v, e := g.Redis().Get(redisCtx, concurrencyKey); e == nil {
+				glog.Infof(ctx, "并发数: %s %d/%d", concurrencyKey, v.Int64(), maxCon)
+			}
+			break
+		}
+		glog.Infof(ctx, "并发超限等待: %s max=%d", concurrencyKey, maxCon)
+		time.Sleep(waitInterval)
+	}
+
+	// 2) 执行业务期间周期续期名额Key：SemaphoreAcquire 仅在首次占用（计数从 0 起）时设 TTL，
+	//    长耗时调用靠本循环持续保活——Key 过期后计数归零会突破 max 并发上限造成超发。
+	stop := make(chan struct{})
+	go refreshTTL(redisCtx, concurrencyKey, keyExpireSec, refreshStep, stop)
+	// 3) 无论业务正常返回还是 panic，都停掉续期并释放名额（幂等，计数归零自动删除 key）
+	defer func() {
+		close(stop)
+		_ = utils.SemaphoreRelease(redisCtx, concurrencyKey)
+	}()
+
+	// 4) 执行业务
+	return f(ctx)
+}
+
+// refreshTTL 周期给名额 Key 续期，直到 stop 关闭；防止长耗时执行期间 Key 提前过期。
+func refreshTTL(redisCtx context.Context, concurrencyKey string, keyExpireSec, step int64, stop <-chan struct{}) {
+	interval := step
+	if interval < 1 {
+		interval = 1
+	}
+	ticker := time.NewTicker(time.Duration(interval) * time.Second)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ticker.C:
+			if _, err := g.Redis().Expire(redisCtx, concurrencyKey, keyExpireSec); err != nil {
+				glog.Errorf(context.TODO(), "redis refresh concurrency ttl err: %v", err)
+			}
+		case <-stop:
+			return
+		}
+	}
+}
+
+// buildChatRequestParams 按模型配置的请求模板 + 业务字段映射构建请求体（ModelCall 请求路径共用）：
+//  1. requestParams 只装模板字段，按 requestBodyMapping 模板校验（CheckParams）+ 构建（ParseConfigTemplate）；
+//     未配置映射的字段（如未配置映射的 messages/tools）会被模板拒绝，明确报错
+//  2. requestParams 为空时按配置模板构建请求结构（模板 defaultValue 生效），
+//     避免"结构由模板声明、值全走业务字段"的场景因请求体为空而构建失败
+//  3. businessParams 只装业务字段，按业务字段名（RequestBusinessFieldMapping 的 key）传值，
+//     TakeBusinessFields 解析为写入路径，构建完成后由 WriteBusinessFields 按路径写入最终请求体
+func buildChatRequestParams(modelInfo *entity.ModelManage, requestParams, businessParams map[string]any) (map[string]any, error) {
+
+	// requestParams 可能混有扁平路径 key（messages.enumValues...）与已是对象/数组的值（stream）。
+	// IsFlatMap 遇 map/slice 值即整体返回 false 会跳过 unflatten；sjson.Set 能处理任意值类型作为子树，
+	// 带点 key 按路径展开、无点 key 直接赋值，故始终 unflatten
+	var err error
+	requestParams, err = utils.UnFlatBySjson(requestParams)
+	if err != nil {
+		return nil, err
+	}
+
+	rest := make(map[string]any, len(requestParams))
+	for k, v := range requestParams {
+		rest[k] = v
+	}
+	// 请求结构源：模板字段兜底（模板 value/defaultValue 生效）+ requestParams 覆盖同名；
+	// 保证模板声明的结构字段（如 stream_options 对象）即使 requestParams 未传也进请求体
+	// 用户已传字段按模板类型元数据递归合并（补 type 包装、补默认字段），数组字段仅在用户提供时才合并
+	src := rest
+	for k, v := range modelInfo.RequestBodyMapping {
+		tmplMap, _ := v.(map[string]any)
+		if tmplMap != nil {
+			if t, _ := tmplMap["type"].(string); t == "array" {
+				if _, has := src[k]; !has {
+					continue
+				}
+				src[k] = modelUtils.MergeNode(src[k], v)
+				continue
+			}
+		}
+		if userVal, has := src[k]; has {
+			src[k] = modelUtils.MergeNode(userVal, v)
+		} else {
+			src[k] = modelUtils.DeepCopyNode(v)
+		}
+	}
+	if len(requestParams) > 0 {
+		// requestParams 非空才按模板严格校验模板字段（空值回填 default）；
+		// 为空时跳过，避免业务字段未写入就误报必填缺失
+		if err := modelUtils.CheckParams(src, modelInfo.RequestBodyMapping); err != nil {
+			return nil, err
+		}
+	}
+
+	out := modelUtils.ParseConfigTemplate(src)
+	if !g.IsEmpty(businessParams) {
+		// 业务字段：businessParams 按业务字段名传值，解析为映射路径后写入
+		bizValues, err := modelUtils.TakeBusinessFields(businessParams, modelInfo.RequestBusinessFieldMapping)
+		if err != nil {
+			return nil, err
+		}
+		// 业务字段按映射路径写入最终请求体（写 out 而非 rest：rest 只是模板字段容器，out 才是下发模型的请求体）
+		if err = modelUtils.WriteBusinessFields(out, bizValues); err != nil {
+			return nil, err
+		}
+		// 合并后按模板约束整体校验：必填/长度/范围
+		if err = modelUtils.CheckBody(out, modelInfo.RequestBodyMapping); err != nil {
+			return nil, err
+		}
+	}
+	// 按模板声明的 type 归一字段值类型（模板字段 value / 业务字段写入值都可能与声明类型不符）
+	out = modelUtils.CoerceBodyTypes(out, modelInfo.RequestBodyMapping)
+	return out, nil
+}

service/model_error_memory_service.go

@@ -0,0 +1,107 @@
+package service
+
+import (
+	"context"
+
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"github.com/gogf/gf/v2/frame/g"
+	"golang.org/x/sync/singleflight"
+)
+
+var analysisGroup singleflight.Group
+
+// shouldRetryWithMemory 统一重试判定：查持久记忆，未命中则调分析模型并落库。
+// 记忆/分析/DB 任一环节失败均 fail-closed 不重试。
+func shouldRetryWithMemory(ctx context.Context, modelInfo *entity.ModelManage, code, msg, rawBody string) (retry bool) {
+	if modelInfo == nil || (code == "" && msg == "") {
+		return false
+	}
+	key := buildMemoryKey(modelInfo.BaseURL, code, msg)
+	if row, err := dao.ModelErrorMemory.GetByKey(ctx, key); err != nil {
+		g.Log().Errorf(ctx, "查询错误重试记忆失败: %v", err)
+		return false
+	} else if row != nil {
+		return row.Retryable
+	}
+	retry, _ = analyzeOnce(key, func() (bool, string) {
+		model, ok := resolveAnalysisModel(ctx, modelInfo)
+		if !ok {
+			g.Log().Warningf(ctx, "无可用分析模型(对话模型)，错误不重试: code=%s", code)
+			return false, ""
+		}
+		r, reason, err := callAnalysisLLM(ctx, model, code, msg, rawBody)
+		if err != nil {
+			g.Log().Warningf(ctx, "错误分析失败，fail-closed不重试: %v", err)
+			return false, ""
+		}
+		row := &entity.ModelErrorMemory{
+			MemoryKey:      key,
+			Upstream:       modelInfo.BaseURL,
+			ErrorCode:      code,
+			MsgFingerprint: msgFingerprint(msg),
+			Retryable:      r,
+			Reason:         reason,
+			AnalyzedBy:     model.ModelName,
+		}
+		if err := dao.ModelErrorMemory.Upsert(ctx, row); err != nil {
+			g.Log().Errorf(ctx, "错误重试记忆落库失败: %v", err)
+		}
+		return r, reason
+	})
+	return retry
+}
+
+// analyzeOnce 按记忆键合并并发分析请求（singleflight）。
+// 注：fn 失败时结果在本次突发内共享（后续新错误会重新分析）。
+func analyzeOnce(key string, fn func() (bool, string)) (bool, string) {
+	v, err, _ := analysisGroup.Do(key, func() (any, error) {
+		retry, reason := fn()
+		return []any{retry, reason}, nil
+	})
+	if err != nil {
+		return false, ""
+	}
+	vals := v.([]any)
+	return vals[0].(bool), vals[1].(string)
+}
+
+var ModelErrorMemory = &modelErrorMemoryService{}
+
+type modelErrorMemoryService struct{}
+
+// List 错误重试记忆列表
+func (s *modelErrorMemoryService) List(ctx context.Context, req *dto.GetErrorMemoryListReq) (res *dto.GetErrorMemoryListRes, err error) {
+	page, size := 1, 20
+	if req.Page != nil && req.Page.PageNum > 0 {
+		page = int(req.Page.PageNum)
+	}
+	if req.Page != nil && req.Page.PageSize > 0 {
+		size = int(req.Page.PageSize)
+	}
+	list, total, err := dao.ModelErrorMemory.List(ctx, page, size)
+	if err != nil {
+		return nil, err
+	}
+	res = &dto.GetErrorMemoryListRes{Total: total, List: make([]dto.ErrorMemoryItem, 0, len(list))}
+	for _, m := range list {
+		res.List = append(res.List, dto.ErrorMemoryItem{
+			Id:             m.Id,
+			MemoryKey:      m.MemoryKey,
+			Upstream:       m.Upstream,
+			ErrorCode:      m.ErrorCode,
+			MsgFingerprint: m.MsgFingerprint,
+			Retryable:      m.Retryable,
+			Reason:         m.Reason,
+			AnalyzedBy:     m.AnalyzedBy,
+		})
+	}
+	return res, nil
+}
+
+// Delete 删除错误重试记忆（手动纠错永久记忆）
+func (s *modelErrorMemoryService) Delete(ctx context.Context, req *dto.DeleteErrorMemoryReq) (err error) {
+	return dao.ModelErrorMemory.Delete(ctx, req.Id)
+}

service/model_http_service.go

@@ -0,0 +1,35 @@
+package service
+
+import (
+	"context"
+	"model-gateway/model/dto"
+
+	commonHttp "gitea.redpowerfuture.com/red-future/common/http"
+	"gitea.redpowerfuture.com/red-future/common/oss"
+	"gitea.redpowerfuture.com/red-future/common/utils"
+)
+
+// IsSuperAdmin 调用admin-go服务检查是否是管理员
+func IsSuperAdmin(ctx context.Context) (res bool, err error) {
+	headers := utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true})
+	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
+	}
+	return r["isSuperAdmin"], err
+}
+
+// Upload 上传文件到 OSS。统一走 common/oss（multipart field=file、X-User-Info 三态注入与旧 setCtxHeader 等价）。
+func Upload(ctx context.Context, req *dto.UploadFileBytesReq) (*dto.UploadFileBytesRes, error) {
+	res, err := oss.UploadFileBytes(ctx, req.FileName, req.FileBytes)
+	if err != nil {
+		return nil, err
+	}
+	return &dto.UploadFileBytesRes{
+		FileURL:           res.FileURL,
+		FileSize:          res.FileSize,
+		FileName:          res.FileName,
+		FileFormat:        res.FileFormat,
+		FileAddressPrefix: res.FileAddressPrefix,
+	}, nil
+}

service/model_invoker.go

@@ -1,417 +0,0 @@
-package service
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"net/http"
-	"net/url"
-	"strings"
-	"time"
-
-	"model-gateway/model/entity"
-
-	"github.com/gogf/gf/v2/container/gvar"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/tidwall/gjson"
-	"github.com/tidwall/sjson"
-)
-
-// parseHeadMsgHeaders 支持多个 header 绑定，逗号分隔：
-// 示例：
-// - X-API-Key:qwen3-tts-key,operation:true,count:123
-// - X-API-Key:"qwen3-tts-key",operation:"true"
-//
-// 说明：
-// - HTTP Header 最终都是字符串，这里做的是“值的字符串化表达”。
-// - 若 value 用双引号包裹，会去掉外层引号再注入，便于在配置中区分字符串/布尔/数字等表达（以及避免值中包含特殊字符时歧义）。
-func parseHeadMsgHeaders(headMsg string) map[string]string {
-	headMsg = strings.TrimSpace(headMsg)
-	if headMsg == "" {
-		return nil
-	}
-	out := map[string]string{}
-	parts := strings.Split(headMsg, ",")
-	for _, p := range parts {
-		p = strings.TrimSpace(p)
-		if p == "" {
-			continue
-		}
-		// HeaderName:HeaderValue（推荐） / HeaderName=HeaderValue（兼容）
-		if strings.Contains(p, ":") {
-			kv := strings.SplitN(p, ":", 2)
-			k := strings.TrimSpace(kv[0])
-			v := strings.TrimSpace(kv[1])
-			v = strings.Trim(v, "\"")
-			if k != "" && v != "" {
-				out[k] = v
-			}
-			continue
-		}
-		if strings.Contains(p, "=") {
-			kv := strings.SplitN(p, "=", 2)
-			k := strings.TrimSpace(kv[0])
-			v := strings.TrimSpace(kv[1])
-			v = strings.Trim(v, "\"")
-			if k != "" && v != "" {
-				out[k] = v
-			}
-			continue
-		}
-	}
-	if len(out) == 0 {
-		return nil
-	}
-	return out
-}
-
-func payloadToQuery(payload any) (url.Values, error) {
-	if payload == nil {
-		return url.Values{}, nil
-	}
-	// 统一转成 map[string]any
-	b, err := json.Marshal(payload)
-	if err != nil {
-		return nil, err
-	}
-	m := map[string]any{}
-	if err := json.Unmarshal(b, &m); err != nil {
-		return nil, err
-	}
-	q := url.Values{}
-	for k, v := range m {
-		if v == nil {
-			continue
-		}
-		// 复杂类型直接 json 字符串化
-		switch vv := v.(type) {
-		case string:
-			q.Set(k, vv)
-		case float64, bool, int, int64, uint64:
-			q.Set(k, fmt.Sprintf("%v", vv))
-		default:
-			bs, _ := json.Marshal(v)
-			q.Set(k, string(bs))
-		}
-	}
-	return q, nil
-}
-
-// 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, m *entity.AsynchModel, payload any, modelKey string) ([]byte, error) {
-	if m == nil || m.BaseURL == "" {
-		return nil, fmt.Errorf("模型配置不完整")
-	}
-
-	// ============ 新增：请求参数映射 ============
-	mappedPayload, err := mapRequestPayload(m.RequestMapping, payload)
-	if err != nil {
-		return nil, fmt.Errorf("请求参数映射失败: %w", err)
-	}
-
-	url := strings.TrimRight(m.BaseURL, "/")
-	timeout := time.Duration(m.TimeoutSeconds) * time.Second
-	if timeout <= 0 {
-		timeout = 60 * time.Second
-	}
-	client := &http.Client{Timeout: timeout}
-
-	method := strings.ToUpper(strings.TrimSpace(m.HttpMethod))
-	if method == "" {
-		method = http.MethodPost
-	}
-
-	var (
-		req *http.Request
-	)
-	switch method {
-	case http.MethodGet:
-		q, err := payloadToQuery(mappedPayload) // 使用映射后的payload
-		if err != nil {
-			return nil, err
-		}
-		if len(q) > 0 {
-			if strings.Contains(url, "?") {
-				url = url + "&" + q.Encode()
-			} else {
-				url = url + "?" + q.Encode()
-			}
-		}
-		req, err = http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
-	default:
-		bodyBytes, err := json.Marshal(mappedPayload) // 使用映射后的payload
-		if err != nil {
-			return nil, err
-		}
-		req, err = http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(bodyBytes))
-	}
-	if err != nil {
-		return nil, err
-	}
-
-	// 先注入模型配置 head_msg（静态头部，适合公共模型固定 API Key）
-	for hk, hv := range parseHeadMsgHeaders(m.HeadMsg) {
-		req.Header.Set(hk, hv)
-	}
-
-	// 最后注入动态 modelKey（允许覆盖/补充静态 head_msg），适合按请求动态传密钥。
-	for hk, hv := range parseHeadMsgHeaders(modelKey) {
-		req.Header.Set(hk, hv)
-	}
-
-	if method != http.MethodGet {
-		req.Header.Set("Content-Type", "application/json")
-	}
-
-	resp, err := client.Do(req)
-	if err != nil {
-		return nil, err
-	}
-	defer resp.Body.Close()
-
-	b, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return nil, err
-	}
-	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
-		msg := string(b)
-		if len(msg) > 2000 {
-			msg = msg[:2000]
-		}
-		return nil, fmt.Errorf("模型服务返回非2xx: %d, body=%s", resp.StatusCode, msg)
-	}
-
-	// ============ 新增：响应参数映射 ============
-	mappedResponse, err := mapResponsePayload(m.ResponseMapping, b)
-	if err != nil {
-		// 响应映射失败不阻塞，返回原始数据
-		g.Log().Warningf(ctx, "响应参数映射失败: %v，返回原始数据", err)
-		return b, nil
-	}
-	// =========================================
-
-	return mappedResponse, nil
-}
-
-// ============================================
-// 映射相关函数
-// ============================================
-
-// mapRequestPayload 将标准请求映射为模型特定格式
-func mapRequestPayload(mappingAny any, payload any) (any, error) {
-	// 1. 解析请求映射配置（值是any类型，支持bool、number等）
-	mapping, err := parseRequestMapping(mappingAny)
-	if err != nil {
-		return nil, err
-	}
-
-	// 如果没有映射配置，直接返回原始payload
-	if len(mapping) == 0 {
-		return payload, nil
-	}
-
-	// 2. 将payload转为map
-	var payloadMap map[string]any
-	switch v := payload.(type) {
-	case map[string]any:
-		payloadMap = v
-	case []map[string]any:
-		// 如果传进来的是纯messages数组，包装成标准格式
-		payloadMap = map[string]any{
-			"messages": v,
-		}
-	default:
-		// 通过JSON转换
-		jsonBytes, err := json.Marshal(payload)
-		if err != nil {
-			return nil, fmt.Errorf("序列化payload失败: %w", err)
-		}
-		if err := json.Unmarshal(jsonBytes, &payloadMap); err != nil {
-			return nil, fmt.Errorf("反序列化payload失败: %w", err)
-		}
-	}
-
-	// 3. 用数据库固定参数覆盖/补充
-	for key, value := range mapping {
-		if existingValue, exists := payloadMap[key]; !exists || isEmptyValue(existingValue) {
-			payloadMap[key] = value
-		}
-	}
-
-	return payloadMap, nil
-}
-
-// mapResponsePayload 将模型响应映射为标准格式
-func mapResponsePayload(mappingAny any, responseBytes []byte) ([]byte, error) {
-	mapping, err := parseResponseMapping(mappingAny)
-	if err != nil {
-		return nil, err
-	}
-	if len(mapping) == 0 {
-		return responseBytes, nil
-	}
-
-	responseStr := string(responseBytes)
-	resultStr := `{}`
-
-	for standardField, modelPath := range mapping {
-		value := gjson.Get(responseStr, modelPath)
-		if !value.Exists() {
-			continue
-		}
-
-		resultStr, err = sjson.SetRaw(resultStr, standardField, value.Raw)
-		if err != nil {
-			return nil, fmt.Errorf("提取字段 %s <- %s 失败: %w", standardField, modelPath, err)
-		}
-	}
-
-	return []byte(resultStr), nil
-}
-
-func parseRequestMapping(mappingAny any) (map[string]any, error) {
-	if mappingAny == nil {
-		return nil, nil
-	}
-
-	result := make(map[string]any)
-
-	switch v := mappingAny.(type) {
-	case *gvar.Var:
-		if v == nil || v.IsNil() || v.IsEmpty() {
-			return nil, nil
-		}
-		// 尝试转成 map
-		if m := v.Map(); m != nil {
-			for k, val := range m {
-				result[k] = val
-			}
-			return result, nil
-		}
-		// 尝试转成 string
-		if s := v.String(); s != "" && s != "{}" && s != "null" {
-			if err := json.Unmarshal([]byte(s), &result); err != nil {
-				return nil, fmt.Errorf("解析请求映射字符串失败: %w", err)
-			}
-			return result, nil
-		}
-		return nil, nil
-	// =======================================================
-
-	case map[string]interface{}:
-		result = v
-
-	case string:
-		if v == "" || v == "{}" || v == "null" {
-			return nil, nil
-		}
-		if err := json.Unmarshal([]byte(v), &result); err != nil {
-			return nil, fmt.Errorf("解析请求映射字符串失败: %w", err)
-		}
-
-	case []byte:
-		if len(v) == 0 {
-			return nil, nil
-		}
-		if err := json.Unmarshal(v, &result); err != nil {
-			return nil, fmt.Errorf("解析请求映射字节失败: %w", err)
-		}
-
-	default:
-		jsonBytes, err := json.Marshal(mappingAny)
-		if err != nil {
-			return nil, fmt.Errorf("序列化映射配置失败: %w", err)
-		}
-		if err := json.Unmarshal(jsonBytes, &result); err != nil {
-			return nil, fmt.Errorf("解析映射配置失败: %w", err)
-		}
-	}
-
-	return result, nil
-}
-
-// parseResponseMapping 解析响应映射配置
-// 返回值类型为 map[string]string，值都是JSON路径字符串
-func parseResponseMapping(mappingAny any) (map[string]string, error) {
-	if mappingAny == nil {
-		return nil, nil
-	}
-
-	mapping := make(map[string]string)
-
-	switch v := mappingAny.(type) {
-	case *gvar.Var:
-		if v == nil || v.IsNil() || v.IsEmpty() {
-			return nil, nil
-		}
-		if m := v.Map(); m != nil {
-			for k, val := range m {
-				if strVal, ok := val.(string); ok {
-					mapping[k] = strVal
-				}
-			}
-			return mapping, nil
-		}
-		if s := v.String(); s != "" && s != "{}" && s != "null" {
-			if err := json.Unmarshal([]byte(s), &mapping); err != nil {
-				return nil, fmt.Errorf("解析响应映射字符串失败: %w", err)
-			}
-			return mapping, nil
-		}
-		return nil, nil
-	case string:
-		if v == "" || v == "{}" || v == "null" {
-			return nil, nil
-		}
-		if err := json.Unmarshal([]byte(v), &mapping); err != nil {
-			return nil, fmt.Errorf("解析响应映射字符串失败: %w", err)
-		}
-
-	case map[string]interface{}:
-		// 数据库JSONB直接返回的map
-		for k, val := range v {
-			if strVal, ok := val.(string); ok {
-				mapping[k] = strVal
-			}
-		}
-
-	case []byte:
-		if len(v) == 0 {
-			return nil, nil
-		}
-		if err := json.Unmarshal(v, &mapping); err != nil {
-			return nil, fmt.Errorf("解析响应映射字节失败: %w", err)
-		}
-
-	default:
-		jsonBytes, err := json.Marshal(mappingAny)
-		if err != nil {
-			return nil, fmt.Errorf("序列化响应映射配置失败: %w", err)
-		}
-		if err := json.Unmarshal(jsonBytes, &mapping); err != nil {
-			return nil, fmt.Errorf("解析响应映射配置失败: %w", err)
-		}
-	}
-
-	return mapping, nil
-}
-
-// isEmptyValue 判断值是否为空
-func isEmptyValue(v any) bool {
-	if v == nil {
-		return true
-	}
-	switch val := v.(type) {
-	case string:
-		return val == ""
-	case []any:
-		return len(val) == 0
-	case map[string]any:
-		return len(val) == 0
-	default:
-		return false
-	}
-}

service/model_manage_service.go

@@ -0,0 +1,386 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"model-gateway/consts/model"
+	"model-gateway/consts/public"
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+	modelUtils "model-gateway/service/utils"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	"github.com/gogf/gf/v2/database/gdb"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelManage = &modelManageService{}
+
+type modelManageService struct{}
+
+// Create 创建模型
+func (s *modelManageService) Create(ctx context.Context, req *dto.CreateModelManageReq) (res *dto.CreateModelManageRes, err error) {
+	err = gfdb.DB(ctx, public.DbNameModelGateway).Transaction(ctx, func(ctx context.Context, tx gdb.TX) (err error) {
+		// 1）检查是否是超管
+		var isSuperAdmin bool
+		isSuperAdmin, err = IsSuperAdmin(ctx)
+		if err != nil {
+			return
+		}
+		req.SystemModel = &isSuperAdmin
+		// 1）如果设为会话模型，先把该用户旧会话模型取消
+		err = s.CancelChatModel(ctx, req.ModelType, req.ChatModel, isSuperAdmin)
+		if err != nil {
+			return
+		}
+		// 2）模型名称唯一性：同一用户下不允许同名模型
+		var user *beans.User
+		user, err = utils.GetUserInfo(ctx)
+		if err != nil {
+			return
+		}
+		if !g.IsEmpty(req.ModelName) {
+			var exist *entity.ModelManage
+			exist, err = dao.ModelManage.GetByCreatorAndName(ctx, user.UserName, req.ModelName)
+			if err != nil {
+				return
+			}
+			if exist != nil {
+				return fmt.Errorf("模型名称 [%s] 已存在，同一用户下不能重复添加同名模型", req.ModelName)
+			}
+		}
+		// 3）插入数据
+		id, err := dao.ModelManage.Insert(ctx, req)
+		if err != nil {
+			return
+		}
+		res = &dto.CreateModelManageRes{Id: id}
+		return
+	})
+	return
+}
+
+// Update 更新模型配置
+// 引用行：只改 apiKey/enabled/chatModel，配置锁定跟随系统模型（其余字段忽略）；
+// 系统模型：仅创建者（超管）可改配置，改名时同步引用行 model_name；
+// 非超管编辑系统模型 → 建引用行（不拷贝配置，apiKey 必填）；用户自有模型：创建者可改全配置。非创建者操作他人行 → 无权限。
+func (s *modelManageService) Update(ctx context.Context, req *dto.UpdateModelManageReq) (res *dto.GetModelManageRes, err error) {
+	err = gfdb.DB(ctx, public.DbNameModelGateway).Transaction(ctx, func(ctx context.Context, tx gdb.TX) (err error) {
+		var get *entity.ModelManage
+		get, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{
+			Id: req.Id,
+		})
+		if err != nil {
+			return
+		}
+		if get == nil {
+			return fmt.Errorf("模型不存在")
+		}
+		var user *beans.User
+		user, err = utils.GetUserInfo(ctx)
+		if err != nil {
+			return
+		}
+
+		// 引用行：只改 apiKey/enabled/chatModel，配置锁定跟随系统模型（其余字段忽略；isSuperAdmin=false=个人会话模型开关）
+		if get.RefSystemModelId > 0 {
+			if get.Creator != user.UserName {
+				return fmt.Errorf("无权限操作")
+			}
+			if err = s.CancelChatModel(ctx, get.ModelType, req.ChatModel, false); err != nil {
+				return err
+			}
+			_, err = dao.ModelManage.Update(ctx, &dto.UpdateModelManageReq{
+				Id:        req.Id,
+				ApiKey:    req.ApiKey,
+				Enabled:   req.Enabled,
+				ChatModel: req.ChatModel,
+			})
+			return
+		}
+
+		// 同名唯一性：同一用户下不允许同名模型（排除自身，允许不改名编辑）
+		if !g.IsEmpty(req.ModelName) {
+			var exist *entity.ModelManage
+			exist, err = dao.ModelManage.GetByCreatorAndName(ctx, user.UserName, req.ModelName)
+			if err != nil {
+				return
+			}
+			if exist != nil && exist.Id != req.Id {
+				res, err = s.Get(ctx, &dto.GetModelManageReq{Id: exist.Id})
+				return
+				//return fmt.Errorf("模型名称 [%s] 已存在，同一用户下不能重复添加同名模型", req.ModelName)
+			}
+		}
+
+		// 1）检查是否是超管
+		var isSuperAdmin bool
+		isSuperAdmin, err = IsSuperAdmin(ctx)
+		if err != nil {
+			return
+		}
+		if err = s.CancelChatModel(ctx, get.ModelType, req.ChatModel, isSuperAdmin); err != nil {
+			return err
+		}
+		if isSuperAdmin {
+			if get.Creator != user.UserName {
+				return fmt.Errorf("无权限操作")
+			}
+			// 3）系统模型改名 → 同步引用行 model_name（保列表 DISTINCT ON 去重正确）
+			if get.SystemModel != nil && *get.SystemModel && req.ModelName != "" && req.ModelName != get.ModelName {
+				if _, err = dao.ModelManage.UpdateReferencesName(ctx, get.Id, req.ModelName); err != nil {
+					return err
+				}
+			}
+		} else {
+			if get.SystemModel != nil && *get.SystemModel {
+				// 用户引用系统模型：apiKey 必填（与旧拷贝分支一致，空 key 引用行无意义）
+				if g.IsEmpty(req.ApiKey) {
+					return fmt.Errorf("模型apiKey不能为空")
+				}
+				var exist *entity.ModelManage
+				exist, err = dao.ModelManage.GetByCreatorAndName(ctx, user.UserName, get.ModelName)
+				if err != nil {
+					return
+				}
+				if exist != nil {
+					return fmt.Errorf("模型名称 [%s] 已存在，同一用户下不能重复添加同名模型", get.ModelName)
+				}
+				// 4）插引用行（配置列零值即可，解析层只读系统行配置）
+				enabled := true
+				if req.Enabled != nil {
+					enabled = *req.Enabled
+				}
+				var id int64
+				id, err = dao.ModelManage.Insert(ctx, &dto.CreateModelManageReq{
+					ModelSupplier:    get.ModelSupplier,
+					ModelName:        get.ModelName,
+					ModelType:        get.ModelType,
+					SystemModel:      gconv.PtrBool(false),
+					ChatModel:        req.ChatModel,
+					ApiKey:           req.ApiKey,
+					Enabled:          gconv.PtrBool(enabled),
+					RefSystemModelId: get.Id,
+				})
+				if err != nil {
+					return err
+				}
+				res, err = s.Get(ctx, &dto.GetModelManageReq{
+					Id: id,
+				})
+				return
+			}
+			if get.Creator != user.UserName {
+				return fmt.Errorf("无权限操作")
+			}
+		}
+		// 4）更新数据
+		_, err = dao.ModelManage.Update(ctx, req)
+		return
+	})
+	return
+}
+
+func (s *modelManageService) CancelChatModel(ctx context.Context, modelType model.ModelType, chatModel *bool, isSuperAdmin bool) (err error) {
+	if !g.IsEmpty(chatModel) && *chatModel {
+		if !g.IsEmpty(modelType) && *modelType == *model.ModelTypeInference.Code {
+			if isSuperAdmin {
+				return fmt.Errorf("超级管理员不能设置会话模型")
+			}
+			// 2）获取该用户信息
+			var user *beans.User
+			user, err = utils.GetUserInfo(ctx)
+			if err != nil {
+				return
+			}
+			// 3）取消该用户之前的会话模型
+			var get *entity.ModelManage
+			get, err = dao.ModelManage.Get(ctx, &dto.GetModelManage{
+				Creator:   user.UserName,
+				ChatModel: chatModel,
+			})
+			if err != nil {
+				return
+			}
+			if g.IsEmpty(get) {
+				return
+			}
+			_, err = dao.ModelManage.Update(ctx, &dto.UpdateModelManageReq{
+				Id:        get.Id,
+				ChatModel: gconv.PtrBool(false),
+			})
+			if err != nil {
+				return
+			}
+		} else {
+			return fmt.Errorf("只有推理模型可以设置成会话模型")
+		}
+	}
+	return
+}
+
+// Delete 删除模型：系统模型被引用 → 拒绝；引用行/自有行仅创建者可删
+func (s *modelManageService) Delete(ctx context.Context, req *dto.DeleteModelManageReq) error {
+	user, err := utils.GetUserInfo(ctx)
+	if err != nil {
+		return err
+	}
+	get, err := dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{Id: req.Id})
+	if err != nil {
+		return err
+	}
+	if get == nil {
+		return nil
+	}
+	// 系统模型被引用 → 拒绝删除（防悬挂）
+	if get.SystemModel != nil && *get.SystemModel {
+		n, err := dao.ModelManage.CountReferences(ctx, get.Id)
+		if err != nil {
+			return err
+		}
+		if n > 0 {
+			return fmt.Errorf("系统模型已被 %d 个用户引用，不能删除", n)
+		}
+		_, err = dao.ModelManage.Delete(ctx, req)
+		return err
+	}
+	// 引用行/自有行：仅创建者可删（删引用行即解除引用）
+	if get.Creator != user.UserName {
+		return fmt.Errorf("无权限操作")
+	}
+	_, err = dao.ModelManage.Delete(ctx, req)
+	return err
+}
+
+func (s *modelManageService) Get(ctx context.Context, req *dto.GetModelManageReq) (res *dto.GetModelManageRes, err error) {
+	get, err := dao.ModelManage.GetNotTenantId(ctx, req)
+	if err != nil {
+		return
+	}
+	if get == nil {
+		return new(dto.GetModelManageRes), nil
+	}
+	// 引用行 → 合入系统模型配置返回（前端展示/编辑需要完整配置；保留引用行自身 id 供更新/删除）
+	if get.RefSystemModelId > 0 {
+		sys, e := dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{Id: get.RefSystemModelId})
+		if e != nil {
+			return nil, e
+		}
+		if sys != nil {
+			get = modelUtils.MergeReferenceConfigForQuery(get, sys)
+		}
+	}
+	// 系统模型 apiKey 对非创建者脱敏（引用行/自有行仅本人可见）
+	user, err := utils.GetUserInfo(ctx)
+	if err != nil {
+		return
+	}
+	if get.SystemModel != nil && *get.SystemModel && get.Creator != user.UserName {
+		get.ApiKey = ""
+	}
+	res = new(dto.GetModelManageRes)
+	err = gconv.Struct(get, &res.ModelManage)
+	return
+}
+
+func (s *modelManageService) GetChatModel(ctx context.Context, req *dto.GetChatModelReq) (res *dto.GetChatModelRes, err error) {
+	user, err := utils.GetUserInfo(ctx)
+	if err != nil {
+		return
+	}
+	get, err := dao.ModelManage.Get(ctx, &dto.GetModelManage{
+		Creator:   user.UserName,
+		ChatModel: gconv.PtrBool(true),
+	})
+	if err != nil {
+		return
+	}
+	res = &dto.GetChatModelRes{
+		ModelManage: get,
+	}
+	return
+}
+
+// List 获取模型列表
+func (s *modelManageService) List(ctx context.Context, req *dto.ListModelManageReq) (res *dto.ListModelManageRes, err error) {
+	if req.IsSameType && !g.IsEmpty(req.Id) {
+		var get *entity.ModelManage
+		get, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{
+			Id: req.Id,
+		})
+		if err != nil {
+			return
+		}
+		req.ModelType = get.ModelType
+	}
+	var user *beans.User
+	user, err = utils.GetUserInfo(ctx)
+	if err != nil {
+		return
+	}
+	req.Creator = user.UserName
+	list, total, err := dao.ModelManage.ListNotTenantId(ctx, req)
+	if err != nil {
+		return
+	}
+	// 引用行 → 合入系统模型配置返回（与 Get 展示一致；保留引用行自身 Id/SystemModel 供更新/脱敏判断）
+	sysCache := make(map[int64]*entity.ModelManage)
+	for _, row := range list {
+		if row.RefSystemModelId <= 0 {
+			continue
+		}
+		sys, ok := sysCache[row.RefSystemModelId]
+		if !ok {
+			sys, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{Id: row.RefSystemModelId})
+			if err != nil {
+				return
+			}
+			if sys == nil {
+				continue
+			}
+			sysCache[row.RefSystemModelId] = sys
+		}
+		*row = *modelUtils.MergeReferenceConfigForQuery(row, sys)
+	}
+	// 系统模型 apiKey 对非创建者脱敏（引用行/自有行仅本人可见）
+	for _, row := range list {
+		if row.SystemModel != nil && *row.SystemModel && row.Creator != user.UserName {
+			row.ApiKey = ""
+		}
+	}
+	res = &dto.ListModelManageRes{
+		Total: total,
+	}
+	err = gconv.Struct(list, &res.List)
+	return
+}
+
+func (s *modelManageService) CheckChatModel(ctx context.Context, req *dto.CheckChatModelReq) (res *dto.CheckChatModelRes, err error) {
+	get, err := s.GetChatModel(ctx, &dto.GetChatModelReq{})
+	if err != nil {
+		return
+	}
+	res = &dto.CheckChatModelRes{
+		IsChatModel: !g.IsEmpty(get),
+	}
+	return
+}
+
+// GetModelType 获取模型类型
+func (s *modelManageService) GetModelType(ctx context.Context, req *dto.ModelTypeReq) (res *dto.ModelTypeRes, err error) {
+	res = &dto.ModelTypeRes{
+		List: model.GetTypeTreeList(),
+	}
+	return res, nil
+}
+
+// GetModelSupplier 获取运营商列表
+func (s *modelManageService) GetModelSupplier(ctx context.Context, req *dto.ModelSupplierReq) (res *dto.ModelSupplierRes, err error) {
+	return &dto.ModelSupplierRes{
+		List: model.GetSupplierOptionList(),
+	}, nil
+}

service/model_service.go

@@ -1,255 +0,0 @@
-package service
-
-import (
-	"context"
-	"errors"
-	"model-gateway/dao"
-	"model-gateway/model/dto"
-	"model-gateway/model/entity"
-
-	"gitea.com/red-future/common/beans"
-	"gitea.com/red-future/common/db/gfdb"
-	"gitea.com/red-future/common/http"
-	"gitea.com/red-future/common/utils"
-	"github.com/gogf/gf/v2/database/gdb"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-var Model = &modelService{}
-
-type modelService struct{}
-
-// IsSuperAdmin 调用admin-go服务检查是否是超级管理员
-func (s *modelService) IsSuperAdmin(ctx context.Context) (res bool, err error) {
-	headers := forwardHeaders(ctx)
-	var r = make(map[string]bool)
-	if err = http.Get(ctx, "admin-go/api/v1/system/user/checkIsSuperAdmin", headers, &r); err != nil {
-		return false, err
-	}
-	return r["isSuperAdmin"], err
-}
-
-func (s *modelService) Create(ctx context.Context, req *dto.CreateModelReq) (res *dto.CreateModelRes, err error) {
-	// 获取当前会话模型
-	if !g.IsEmpty(req.IsChatModel) && *req.IsChatModel == 1 {
-		var model *entity.AsynchModel
-		model, err = dao.Model.GetByIsChatModel(ctx)
-		if err != nil {
-			return nil, err
-		}
-		// 如果有会话模型，那就改变为 0
-		if model != nil {
-			_, err = dao.Model.Update(ctx, &dto.UpdateModelReq{
-				ID:          model.Id,
-				IsChatModel: gconv.PtrInt(0),
-			})
-			if err != nil {
-				return nil, err
-			}
-		}
-	}
-
-	req.IsOwner = gconv.PtrInt(1)
-	admin, err := s.IsSuperAdmin(ctx)
-	if err != nil {
-		return
-	}
-	if admin {
-		req.IsOwner = gconv.PtrInt(0)
-	}
-	id, err := dao.Model.Insert(ctx, req)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.CreateModelRes{ID: id}, nil
-}
-
-func (s *modelService) Update(ctx context.Context, req *dto.UpdateModelReq) error {
-	//根据当前 isChatModel 来判断是否更新模型
-	if req.IsChatModel == gconv.PtrInt(1) {
-		//判断当前用户是否有会话模型
-		model, err := dao.Model.GetByIsChatModel(ctx)
-		if err != nil {
-			return err
-		}
-		if model != nil {
-			return errors.New("用户已存在会话模型，不能创建")
-		}
-	}
-
-	req.IsOwner = gconv.PtrInt(1)
-	admin, err := s.IsSuperAdmin(ctx)
-	if err != nil {
-		return err
-	}
-	if admin {
-		req.IsOwner = gconv.PtrInt(0)
-		_, err = dao.Model.Update(ctx, req)
-		if err != nil {
-			return err
-		}
-		return nil
-	}
-
-	var user *beans.User
-	user, err = utils.GetUserInfo(ctx)
-	if err != nil {
-		return err
-	}
-	// 判断当前传过来的模型id的模型是否是超级管理员的。如果是超管的进行创建，否则更新
-	var count int
-	count, err = dao.Model.Count(ctx, &dto.GetModelReq{
-		ID:      req.ID,
-		Creator: user.UserName,
-	})
-	if err != nil {
-		return err
-	}
-	if count == 0 {
-		insertDto := new(dto.CreateModelReq)
-		err = gconv.Struct(req, insertDto)
-		if err != nil {
-			return err
-		}
-		_, err = dao.Model.Insert(ctx, insertDto)
-		return err
-	}
-	_, err = dao.Model.Update(ctx, req)
-	return err
-}
-
-func (s *modelService) Delete(ctx context.Context, id string) error {
-	_, err := dao.Model.DeleteByID(ctx, id)
-	return err
-}
-
-func (s *modelService) Get(ctx context.Context, id int64) (*entity.AsynchModel, error) {
-	model, err := dao.Model.Get(ctx, id)
-	if err != nil {
-		return nil, err
-	}
-	model.Form = ParseJSONField(model.Form)
-	model.RequestMapping = ParseJSONField(model.RequestMapping)
-	model.ResponseMapping = ParseJSONField(model.ResponseMapping)
-	model.ResponseBody = ParseJSONField(model.ResponseBody)
-	return model, nil
-}
-
-func (s *modelService) List(ctx context.Context, req *dto.ListModelReq) (list []*entity.AsynchModel, total int, err error) {
-	var models []*entity.AsynchModel
-
-	req.IsOwner = gconv.PtrInt(1)
-	admin, err := s.IsSuperAdmin(ctx)
-	if err != nil {
-		return
-	}
-	if admin {
-		req.IsOwner = gconv.PtrInt(0)
-	}
-
-	var user *beans.User
-	user, err = utils.GetUserInfo(ctx)
-	if err != nil {
-		return nil, 0, err
-	}
-	req.Creator = user.UserName
-
-	models, total, err = dao.Model.GetByCreatorAndPlatform(ctx, req)
-	if err != nil {
-		return
-	}
-
-	// 处理列表中每条记录的 JSONB 字段
-	for _, m := range models {
-		m.Form = ParseJSONField(m.Form)
-		m.RequestMapping = ParseJSONField(m.RequestMapping)
-		m.ResponseMapping = ParseJSONField(m.ResponseMapping)
-		m.ResponseBody = ParseJSONField(m.ResponseBody)
-	}
-	return models, total, nil
-}
-
-// GetModelTypesFromConfig 从配置文件读取模型类型
-func GetModelTypesFromConfig(ctx context.Context) map[int]string {
-	typeMap := make(map[int]string)
-
-	// 读取配置
-	configMap := g.Cfg().MustGet(ctx, "modelType.types").Map()
-	for k, v := range configMap {
-		typeID := gconv.Int(k)
-		typeName := gconv.String(v)
-		if typeID > 0 && typeName != "" {
-			typeMap[typeID] = typeName
-		}
-	}
-	// 如果配置为空，使用默认值
-	if len(typeMap) == 0 {
-		typeMap = map[int]string{
-			1: "推理模型",
-			2: "图片模型",
-			3: "音频模型",
-			4: "向量化模型",
-			5: "全模态模型",
-		}
-	}
-	return typeMap
-}
-
-func (s *modelService) UpdateChatModel(ctx context.Context, req *dto.UpdateChatModelReq) error {
-	// 校验新会话模型是否存在
-	newModel, err := dao.Model.Get(ctx, req.Id)
-	if err != nil {
-		return err
-	}
-	if newModel == nil {
-		return errors.New("新会话模型不存在")
-	}
-
-	// 获取当前用户会话模型
-	currentModel, err := dao.Model.GetByIsChatModel(ctx)
-	if err != nil {
-		return err
-	}
-	err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) error {
-		if !g.IsEmpty(currentModel) {
-			if currentModel.ModelType != 1 {
-				return errors.New("当前模型为非推理模型,不能设置为会话模型")
-			}
-
-			// 如果点击的就是当前会话模型（已经是1），取消它（设为0）
-			if currentModel.Id != req.Id {
-				_, err = dao.Model.Update(ctx, &dto.UpdateModelReq{
-					ID:          currentModel.Id,
-					IsChatModel: gconv.PtrInt(0),
-				})
-				if err != nil {
-					return err
-				}
-			}
-		}
-
-		// 设置当前为会话模型（设为1）
-		_, err = dao.Model.Update(ctx, &dto.UpdateModelReq{
-			ID:          req.Id,
-			IsChatModel: gconv.PtrInt(1),
-		})
-		return err
-	})
-	return err
-}
-
-func (s *modelService) GetIsChatModel(ctx context.Context) (*entity.AsynchModel, error) {
-	model, err := dao.Model.GetByIsChatModel(ctx)
-	if err != nil {
-		return nil, err
-	}
-	if model == nil {
-		return nil, nil
-	}
-	model.Form = ParseJSONField(model.Form)
-	model.RequestMapping = ParseJSONField(model.RequestMapping)
-	model.ResponseMapping = ParseJSONField(model.ResponseMapping)
-	model.ResponseBody = ParseJSONField(model.ResponseBody)
-	return model, nil
-}

service/model_task_end_service.go

@@ -0,0 +1,376 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"model-gateway/consts/public"
+	"model-gateway/dao"
+	"model-gateway/model/domain"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+	"model-gateway/service/httpclient"
+	modelUtils "model-gateway/service/utils"
+	"regexp"
+	"time"
+
+	"gitea.redpowerfuture.com/red-future/common/beans"
+	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
+	"gitea.redpowerfuture.com/red-future/common/oss"
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	gmq "github.com/bjang03/gmq/core/gmq"
+	"github.com/bjang03/gmq/mq"
+	"github.com/bjang03/gmq/types"
+	"github.com/gogf/gf/v2/container/gvar"
+	"github.com/gogf/gf/v2/database/gdb"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelTaskEndService = &modelTaskEndService{}
+
+type modelTaskEndService struct{}
+
+// GetTaskStartList 获取待执行任务
+func (s *modelTaskEndService) GetTaskStartList(ctx context.Context) (err error) {
+	workerNum := g.Cfg().MustGet(ctx, "pool.workerNum", modelUtils.DefaultWorkerNum).Int64()
+
+	redisKey := "model_video_task:"
+	var (
+		pageNum = gconv.Int64(1)
+		remain  = workerNum
+	)
+	// 字段列表
+	cols := []string{
+		entity.ModelTaskStartCol.Id,
+		entity.ModelTaskStartCol.TaskId,
+		entity.ModelTaskStartCol.ModelId,
+		entity.ModelTaskStartCol.BizName,
+		entity.ModelTaskStartCol.Creator,
+		entity.ModelTaskStartCol.TenantId,
+		entity.ModelTaskStartCol.MsgTopic,
+		entity.ModelTaskStartCol.MediaType,
+	}
+
+	for remain > 0 {
+		req := &dto.GetModelTaskStartListReq{
+			Page: &beans.Page{
+				PageNum:  pageNum,
+				PageSize: remain, // 每页只查当前需要的数量
+			},
+		}
+
+		var list []entity.ModelTaskStart
+		list, err = dao.ModelTaskStart.ListByLimitNotTenantId(ctx, req, cols...)
+		if err != nil {
+			return fmt.Errorf("查询任务失败: %w", err)
+		}
+		if len(list) == 0 {
+			break
+		}
+
+		// 3. 组装锁key，批量查询Redis（性能最优）
+		taskMap := make(map[string]*entity.ModelTaskStart, len(list))
+		lockKeys := make([]string, 0, len(list))
+		for _, item := range list {
+			key := redisKey + gconv.String(item.Id)
+			taskMap[key] = &item
+			lockKeys = append(lockKeys, key)
+		}
+
+		var mGetRes map[string]*gvar.Var
+		mGetRes, err = g.Redis().MGet(ctx, lockKeys...)
+		if err != nil {
+			return fmt.Errorf("批量查询锁状态失败: %w", err)
+		}
+
+		// 4. 提交异步处理：锁在 goroutine 内抢（WithLock 单次尝试），此处 MGet 只做快速预筛
+		for _, key := range lockKeys {
+			val := gconv.String(mGetRes[key])
+			// 已被其他实例抢占，跳过（MGet 只做快速预筛；真正互斥靠 goroutine 内的原子抢锁）
+			if val != "" {
+				continue
+			}
+			err = s.handleSingleTask(ctx, taskMap[key], key)
+			if err != nil {
+				g.Log().Errorf(ctx, "提交任务失败: %v", err)
+			}
+			remain-- // 占用一个槽位
+			if remain <= 0 {
+				break // 槽位已满，终止遍历
+			}
+		}
+		pageNum++ // 页码动态累加，不再写死2
+	}
+
+	return nil
+}
+
+// taskLockTTL 任务锁 TTL（秒）。utils.WithLock 自动续期锁住整个任务处理，TTL 仅作崩溃兜底：
+// worker 崩溃后续期停止，TTL 过期后其他 worker 重新抢占。
+const taskLockTTL = 1200
+
+var urlParamReg = regexp.MustCompile(`\{.+?\}`)
+
+// handleSingleTask 提交异步处理：锁在 goroutine 内抢（utils.WithLock 自动续期 + 单次尝试）。
+// 自动续期：锁持满整个任务处理，任务 >20min 不提前过期，避免其它 worker 重新抢到导致重复处理；
+// 单次尝试：锁被其它 worker 持有（任务已被别人处理）时立刻跳过——等待会拿着过期 item 在行删除后
+// 重复扣费/重复回调。Submit 失败（池关闭）goroutine 不运行、从没抢锁，无锁泄漏路径。
+func (s *modelTaskEndService) handleSingleTask(ctx context.Context, item *entity.ModelTaskStart, lockKey string) error {
+	return modelUtils.Submit(ctx, func(ctx context.Context) {
+		asyncCtx := context.WithoutCancel(ctx)
+		// OSS 桶名依赖 ctx 中的用户（GetBucketName → tenantid-{tenantId}），
+		// 响应临时路径转存 OSS 需要用户信息，故在任务体最前面注入
+		asyncCtx = context.WithValue(asyncCtx, "user", &beans.User{
+			UserName: item.Creator,
+			TenantId: item.TenantId,
+		})
+		ok, err := utils.WithLock(asyncCtx, lockKey, taskLockTTL, func(ctx context.Context) error {
+			return s.processClaimedTask(ctx, item)
+		}, 1)
+		if err != nil || !ok {
+			// 锁被其它实例持有或抢锁失败：跳过，任务行保留由持有方处理，下轮扫描不再命中
+			g.Log().Warningf(asyncCtx, "任务锁未抢占，跳过 taskId=%d: %v", item.Id, err)
+		}
+	})
+}
+
+// processClaimedTask 抢到任务锁后的完整处理：轮询模型结果 → 终态落库 + 发布。
+// 终态结果统一承载：成功/错误/解析失败任何路径都写进 docMsg.ErrorMsg 后走 finalize 落库+发布，
+// 避免早期直接 return 把任务丢弃——任务行不删、无结果落库、调用方永远收不到通知，只会被其他 worker 反复重捡。
+func (s *modelTaskEndService) processClaimedTask(asyncCtx context.Context, item *entity.ModelTaskStart) error {
+	startTime := time.Now()
+	docMsg := new(dto.ModelMsg)
+	docMsg.TaskID = item.Id
+	var respObj map[string]any
+
+	// 终态处理：删任务行 → 插结果行（含 ErrorMsg）→ NATS 发布结果给调用方
+	finalize := func() {
+		err := gfdb.DB(asyncCtx, public.DbNameModelGateway).Transaction(asyncCtx, func(asyncCtx context.Context, tx gdb.TX) (err error) {
+			// 删除视频任务
+			_, err = dao.ModelTaskStart.Delete(asyncCtx, &dto.DeleteModelTaskStartReq{
+				Id: item.Id,
+			})
+			if err != nil {
+				return err
+			}
+
+			// 保存视频任务结果
+			_, err = dao.ModelTaskEnd.Insert(asyncCtx, &dto.CreateModelTaskEndReq{
+				ModelId:                item.ModelId,
+				BizName:                item.BizName,
+				MsgTopic:               item.MsgTopic,
+				TaskId:                 item.TaskId,
+				ResponseParams:         docMsg.Content,
+				OriginalResponseParams: respObj,
+				DurationSeconds:        int64(time.Since(startTime).Seconds()),
+				PromptTokens:           docMsg.PromptTokens,
+				CompletionTokens:       docMsg.CompletionTokens,
+				TotalTokens:            docMsg.TotalTokens,
+				TotalCost:              docMsg.Cost,
+				ErrorMsg:               docMsg.ErrorMsg,
+			})
+			if err != nil {
+				return err
+			}
+			return
+		})
+		if err != nil {
+			g.Log().Errorf(asyncCtx, "保存视频任务结果失败: %v", err)
+		}
+		// 发布消息
+		if err = TaskMsgPublish(asyncCtx, item.MsgTopic, docMsg); err != nil {
+			g.Log().Errorf(asyncCtx, "模型消息发布失败: %v", err)
+		}
+	}
+
+	// 按 modelId 现查模型配置（异步映射/token 映射/计费规则不随任务快照，任务完成时取当前配置）
+	modelInfo, err := dao.ModelManage.GetNotTenantId(asyncCtx, &dto.GetModelManageReq{Id: item.ModelId})
+	if err != nil {
+		g.Log().Errorf(asyncCtx, "查询模型配置失败: modelId=%d err=%v", item.ModelId, err)
+		docMsg.ErrorMsg = fmt.Sprintf("查询模型配置失败: %v", err)
+		finalize()
+		return nil
+	}
+	if modelInfo == nil {
+		g.Log().Errorf(asyncCtx, "模型配置不存在: modelId=%d", item.ModelId)
+		docMsg.ErrorMsg = fmt.Sprintf("模型配置不存在: modelId=%d", item.ModelId)
+		finalize()
+		return nil
+	}
+
+	// 引用行 → 解析为系统模型配置+本人 apiKey（轮询/计价均用系统模型）
+	modelInfo, err = modelUtils.ResolveModelConfig(asyncCtx, modelInfo)
+	if err != nil {
+		g.Log().Errorf(asyncCtx, "模型配置解析失败: modelId=%d err=%v", item.ModelId, err)
+		docMsg.ErrorMsg = fmt.Sprintf("模型配置解析失败: %v", err)
+		finalize()
+		return nil
+	}
+
+	// 连续轮询失败上限：瞬时抖动（HTTP 错/空响应/解析失败）先有限重试，超限按终态错误落库
+	const maxPollErrRetries = 3
+	pollErrCnt := 0
+LOOP:
+	// 替换URL占位符
+	url := urlParamReg.ReplaceAllString(modelInfo.AsyncTaskMapping.Url, item.TaskId)
+	// 组装查询请求体：POST 查询接口需要 body（从 RequestBodyMapping 出发，替换 {…} 占位符为任务 ID）
+	reqBody := buildAsyncTaskBody(modelInfo.AsyncTaskMapping.RequestBodyMapping, item.TaskId)
+	// 发起HTTP请求
+	modelRespBody, err := httpclient.ModelHttpNormalRequest(
+		asyncCtx,
+		url,
+		modelInfo.AsyncTaskMapping.RequestHeadMapping,
+		modelInfo.AsyncTaskMapping.HttpMethod, reqBody,
+	)
+	if err != nil {
+		g.Log().Errorf(asyncCtx, "模型请求失败: %v", err)
+		if pollErrCnt < maxPollErrRetries {
+			pollErrCnt++
+			time.Sleep(10 * time.Second)
+			goto LOOP
+		}
+		docMsg.ErrorMsg = fmt.Sprintf("模型请求失败: %v", err)
+		finalize()
+		return nil
+	}
+	if modelRespBody == nil {
+		g.Log().Errorf(asyncCtx, "模型返回参数为空")
+		if pollErrCnt < maxPollErrRetries {
+			pollErrCnt++
+			time.Sleep(10 * time.Second)
+			goto LOOP
+		}
+		docMsg.ErrorMsg = "模型返回参数为空"
+		finalize()
+		return nil
+	}
+	pollErrCnt = 0 // 请求成功一次即重置连续失败计数
+
+	// 异常响应识别：按模型 ErrorMessageMapping 解析，无错误返回空串
+	if _, docMsg.ErrorMsg, err = parseModelError(modelRespBody, modelInfo.ErrorMessageMapping); err != nil {
+		g.Log().Errorf(asyncCtx, "模型返回参数解析失败：%v", err)
+		if pollErrCnt < maxPollErrRetries {
+			pollErrCnt++
+			time.Sleep(10 * time.Second)
+			goto LOOP
+		}
+		docMsg.ErrorMsg = fmt.Sprintf("模型返回参数解析失败: %v", err)
+		finalize()
+		return nil
+	}
+	// 统一字段路径（GetByPath）读取基于该对象
+	if err = json.Unmarshal(modelRespBody, &respObj); err != nil {
+		g.Log().Errorf(asyncCtx, "模型返回参数解析失败：%v", err)
+		if pollErrCnt < maxPollErrRetries {
+			pollErrCnt++
+			time.Sleep(10 * time.Second)
+			goto LOOP
+		}
+		docMsg.ErrorMsg = fmt.Sprintf("模型返回参数解析失败: %v", err)
+		finalize()
+		return nil
+	}
+
+	// 无错误时才组装成功内容（错误响应按终态处理，跳过成功解析/轮询）
+	if docMsg.ErrorMsg == "" {
+		// 组装业务返回内容
+		respBodyMap := modelUtils.CleanMapFieldPath(modelInfo.ResponseBodyMapping)
+		content := make(map[string]any, len(respBodyMap))
+		for bizKey, jsonPath := range respBodyMap {
+			content[bizKey] = oss.TempURLToOSS(asyncCtx, modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(jsonPath)))
+		}
+		docMsg.Content = content
+
+		// 解析 ResponseBusinessFieldMapping 字段
+		businessField := make(map[string]any, len(modelInfo.ResponseBusinessFieldMapping))
+		for key, value := range modelInfo.ResponseBusinessFieldMapping {
+			businessField[key] = modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(value))
+		}
+		businessFieldRes := new(domain.VideoFieldsRes)
+		err = gconv.Struct(businessField, businessFieldRes)
+		if err != nil {
+			docMsg.ErrorMsg = fmt.Sprintf("解析 ResponseBusinessFieldMapping 字段失败: %v", err)
+			finalize()
+			return nil
+		}
+		docMsg.Duration = businessFieldRes.Duration
+
+		// 解析Token
+		totalTokPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.TotalTokens)
+		promptTokPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.PromptTokens)
+		compTokPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.CompletionTokens)
+
+		docMsg.TotalTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, totalTokPath))
+		docMsg.PromptTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, promptTokPath))
+		docMsg.CompletionTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, compTokPath))
+
+		// 判断任务状态，轮询等待
+		statusPath := modelUtils.CleanFieldPath(modelInfo.AsyncTaskMapping.TaskStatus)
+		status := gconv.String(modelUtils.GetByPathValue(respObj, statusPath))
+		if status == modelInfo.AsyncTaskMapping.TaskStatusPending || status == modelInfo.AsyncTaskMapping.TaskStatusRunning {
+			time.Sleep(10 * time.Second)
+			goto LOOP
+		}
+
+		// 调 shop-user-trade 按用量算费（媒体类型取任务创建时的快照；subject=解析后的系统模型 id）
+		docMsg.ModelId = modelInfo.Id // 引用行=系统模型 id，供 per_token 结算按系统模型计价
+		docMsg.MediaType = item.MediaType
+		docMsg.Cost = calcModelCost(asyncCtx, modelInfo.Id,
+			buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, item.MediaType, docMsg.Duration))
+	}
+	// 成功或已识别出错误的终态统一落库+发布（内容组装完成/错误消息已写入 docMsg）
+	finalize()
+	return nil
+}
+
+// buildAsyncTaskBody 组装异步任务查询请求体：从 AsyncTaskMapping.RequestBodyMapping 出发，
+// 把 {…} 占位符（如 {taskId}）替换为实际任务 ID，兼容 POST 查询接口需要请求体的场景。
+// 未配置映射时返回 nil（GET 查询/无需 body 的场景）。
+func buildAsyncTaskBody(mapping map[string]any, taskID string) map[string]any {
+	if len(mapping) == 0 {
+		return nil
+	}
+	out := make(map[string]any, len(mapping))
+	for k, v := range mapping {
+		out[k] = replaceTaskPlaceholder(v, taskID)
+	}
+	return out
+}
+
+// replaceTaskPlaceholder 递归替换结构体中的 {…} 占位符为任务 ID
+func replaceTaskPlaceholder(v any, taskID string) any {
+	switch val := v.(type) {
+	case string:
+		return urlParamReg.ReplaceAllString(val, taskID)
+	case map[string]any:
+		m := make(map[string]any, len(val))
+		for k, x := range val {
+			m[k] = replaceTaskPlaceholder(x, taskID)
+		}
+		return m
+	case []any:
+		arr := make([]any, len(val))
+		for i, x := range val {
+			arr[i] = replaceTaskPlaceholder(x, taskID)
+		}
+		return arr
+	default:
+		return v
+	}
+}
+
+func TaskMsgPublish(ctx context.Context, topic string, data *dto.ModelMsg) (err error) {
+	err = gmq.GetGmq(public.GmqMsgPluginsName).GmqPublish(ctx, &mq.NatsPubMessage{
+		PubMessage: types.PubMessage{
+			Topic: topic,
+			Data:  data,
+		},
+		Durable: true,
+	})
+	if err != nil {
+		g.Log().Errorf(ctx, "[TaskMsgPublish] 发布消息失败 [Error]: %v", err)
+		return fmt.Errorf("发布消息失败")
+	}
+	return
+}

service/model_task_start_service.go

@@ -0,0 +1,90 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/service/httpclient"
+	modelUtils "model-gateway/service/utils"
+	"time"
+
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelTaskStart = &modelTaskStartService{}
+
+type modelTaskStartService struct{}
+
+// CreateTask 创建任务
+func (s *modelTaskStartService) CreateTask(ctx context.Context, req *dto.CallModelTaskStartReq) (res *dto.ModelCallRes, err error) {
+	startTime := time.Now()
+	attempt := 0
+
+	id := req.Id
+	modelInfo := req.ModelInfo
+	newRequestParams := req.RequestParams
+
+LOOP:
+	// 6) 模型请求
+	modelRespBody, err := httpclient.ModelHttpNormalRequest(ctx, modelInfo.BaseURL, modelInfo.RequestHeadMapping, modelInfo.HttpMethod, newRequestParams)
+	if err != nil {
+		if attempt < modelCallMaxRetries && shouldRetryWithMemory(ctx, modelInfo, "", err.Error(), "") {
+			attempt++
+			if waitErr := retryWait(ctx, attempt); waitErr != nil {
+				return nil, waitErr
+			}
+			goto LOOP
+		}
+		return nil, fmt.Errorf("模型请求失败: %v", err)
+	}
+	if modelRespBody == nil {
+		return nil, fmt.Errorf("模型返回参数是空")
+	}
+	// 7) 更新视频任务信息（统一字段路径 GetByPath 基于该对象读取）
+	var respObj map[string]any
+	if err = json.Unmarshal(modelRespBody, &respObj); err != nil {
+		return nil, fmt.Errorf("模型返回参数解析失败：%v", err)
+	}
+	updateModelReq := dto.UpdateModelTaskStartReq{
+		Id:                     id,
+		OriginalResponseParams: respObj,
+	}
+	docMsg := new(dto.ModelCallRes)
+	docMsg.TaskId = id
+	// 按模型 ErrorMessageMapping 解析错误响应，无错误返回空串
+	var errCode string
+	if errCode, docMsg.ErrorMsg, err = parseModelError(modelRespBody, modelInfo.ErrorMessageMapping); err != nil {
+		return nil, fmt.Errorf("模型返回参数解析失败：%v", err)
+	}
+	if docMsg.ErrorMsg != "" {
+		if attempt < modelCallMaxRetries && shouldRetryWithMemory(ctx, modelInfo, errCode, docMsg.ErrorMsg, string(modelRespBody)) {
+			attempt++
+			if waitErr := retryWait(ctx, attempt); waitErr != nil {
+				return nil, waitErr
+			}
+			goto LOOP
+		}
+		updateModelReq.ErrorMsg = docMsg.ErrorMsg
+	}
+	if docMsg.ErrorMsg == "" {
+		taskIDPath := modelUtils.CleanFieldPath(modelInfo.AsyncTaskMapping.TaskId)
+		docMsg.Content = map[string]any{
+			"respBody": modelUtils.GetByPathValue(respObj, taskIDPath),
+		}
+	}
+	if !g.IsEmpty(docMsg.Content) {
+		updateModelReq.ResponseParams = docMsg.Content
+		updateModelReq.TaskId = gconv.String(docMsg.Content["respBody"])
+	}
+	updateModelReq.DurationSeconds = int64(time.Since(startTime).Seconds())
+	// 8) 更新模型视频任务信息
+	_, err = dao.ModelTaskStart.Update(ctx, &updateModelReq)
+	if err != nil {
+		return nil, fmt.Errorf("更新模型视频任务信息失败: %v", err)
+	}
+
+	return docMsg, nil
+}

service/parse_error.go

@@ -0,0 +1,194 @@
+package service
+
+import (
+	"encoding/json"
+	"reflect"
+
+	modelUtils "model-gateway/service/utils"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// parseModelError 按模型配置的 ErrorMessageMapping 解析错误响应，返回错误码与错误消息（无错误均返回空串）。
+// ErrorMessageMapping 为 schema 树（与请求模板同格式：type/value/label/isForm/required/fieldType/defaultValue/attrs），
+// 解析时剔除包装字段取 code/message 的字段路径（复用 normalizeSchemaPath 归一 attrs/数组段），
+// 经 GetByPath 从响应体提取。成功判定：
+//   - code 提取值为空（nil/空串/数字零）→ 成功
+//   - code 节点配置了 defaultValue 且提取值等于它 → 成功
+//   - 否则 → 错误（返回 code + message）
+//
+// 未配置 ErrorMessageMapping → 不识别错误，一律返回成功（纯配置驱动）。
+// 解析失败返回 err，由调用方决定重试/终态。
+func parseModelError(body []byte, mapping map[string]any) (code, msg string, err error) {
+	if len(mapping) == 0 {
+		return "", "", nil
+	}
+	var respObj map[string]any
+	if err = json.Unmarshal(body, &respObj); err != nil {
+		return "", "", err
+	}
+	codePath, msgPath, hasCodeDefault, codeDefault := collectErrorMapping(mapping)
+	codeVal := modelUtils.GetByPathValue(respObj, codePath)
+	msgVal := modelUtils.GetByPathValue(respObj, msgPath)
+	if codePath != "" {
+		if isEmptyErrorCode(codeVal) {
+			return "", "", nil
+		}
+		if hasCodeDefault && sameErrorValue(codeVal, codeDefault) {
+			return "", "", nil
+		}
+		return gconv.String(codeVal), gconv.String(msgVal), nil
+	}
+	// 未配置 code 路径：以 message 是否为空判定错误
+	if !isEmptyErrorCode(msgVal) {
+		return "", gconv.String(msgVal), nil
+	}
+	return "", "", nil
+}
+
+// collectErrorMapping 遍历 ErrorMessageMapping schema 树，提取 code/message 的字段路径与 code 节点的 defaultValue。
+// 支持 schema 节点（含 type/attrs 等包装）与纯字符串路径两种形态；数组段经 normalizeSchemaPath 归一到 [*]。
+// 返回路径为已归一字段路径；未配置返回空串。
+func collectErrorMapping(mapping map[string]any) (codePath, msgPath string, hasCodeDefault bool, codeDefault any) {
+	var walk func(node map[string]any, prefix string)
+	walk = func(node map[string]any, prefix string) {
+		for key, val := range node {
+			if isErrorMetaKey(key) {
+				continue
+			}
+			m, ok := val.(map[string]any)
+			if !ok {
+				// 纯值形态：字段值直接是字段路径字符串
+				if s, ok := val.(string); ok {
+					path := normalizeSchemaPath(joinErrorFieldPath(prefix, s))
+					if key == "code" && codePath == "" {
+						codePath = path
+					} else if key == "message" && msgPath == "" {
+						msgPath = path
+					}
+				}
+				continue
+			}
+			nodeType, _ := m["type"].(string)
+			hasDef := false
+			var def any
+			if d, has := m["defaultValue"]; has && d != nil {
+				hasDef, def = true, d
+			}
+			switch {
+			case nodeType == "object":
+				if attrs, ok := m["attrs"].(map[string]any); ok {
+					walk(attrs, joinErrorFieldPath(prefix, key))
+				} else if attrs, ok := m["attrs"].([]any); ok && len(attrs) > 0 {
+					if elm, ok := attrs[0].(map[string]any); ok {
+						walkErrorArrayElement(elm, joinErrorArrayPath(prefix, key), walk)
+					}
+				} else {
+					walk(m, joinErrorFieldPath(prefix, key))
+				}
+			case nodeType == "array":
+				walkErrorArrayContainer(m, joinErrorArrayPath(prefix, key), walk)
+			case nodeType == "":
+				// 无 type 键：纯容器（字段直接作为键），递归下钻
+				walk(m, joinErrorFieldPath(prefix, key))
+			default:
+				// 标量叶子字段
+				path := normalizeSchemaPath(joinErrorFieldPath(prefix, key))
+				if key == "code" && codePath == "" {
+					codePath, hasCodeDefault, codeDefault = path, hasDef, def
+				} else if key == "message" && msgPath == "" {
+					msgPath = path
+				}
+			}
+		}
+	}
+	walk(mapping, "")
+	return
+}
+
+// walkErrorArrayContainer 数组节点：子字段容器依次尝试 attrs([]any)/enumValues/value([]any)
+func walkErrorArrayContainer(node map[string]any, prefix string, walk func(map[string]any, string)) {
+	for _, container := range []string{"attrs", "enumValues", "value"} {
+		items, ok := node[container].([]any)
+		if !ok || len(items) == 0 {
+			continue
+		}
+		if elm, ok := items[0].(map[string]any); ok {
+			walkErrorArrayElement(elm, prefix, walk)
+		}
+		return
+	}
+}
+
+// walkErrorArrayElement 数组元素：字段在其 attrs 下（对象元素）或直接作为键（纯元素）
+func walkErrorArrayElement(elm map[string]any, prefix string, walk func(map[string]any, string)) {
+	if attrs, ok := elm["attrs"].(map[string]any); ok {
+		walk(attrs, prefix)
+		return
+	}
+	walk(elm, prefix)
+}
+
+// isErrorMetaKey 判断是否为 schema 节点元数据字段（非业务字段，剔除）
+func isErrorMetaKey(key string) bool {
+	switch key {
+	case "type", "value", "label", "isForm", "required", "fieldType", "defaultValue", "description":
+		return true
+	}
+	return false
+}
+
+// joinErrorFieldPath 拼接字段路径（数组段 [0] 由 normalizeSchemaPath 归一为 [*]）
+func joinErrorFieldPath(prefix, key string) string {
+	if prefix == "" {
+		return key
+	}
+	return prefix + "." + key
+}
+
+// joinErrorArrayPath 数组字段路径：元素下标 [0] 由 normalizeSchemaPath 归一为 [*]
+func joinErrorArrayPath(prefix, key string) string {
+	if prefix == "" {
+		return key + "[0]"
+	}
+	return prefix + "." + key + "[0]"
+}
+
+// isEmptyErrorCode 判断错误码是否为空（空=无错误）：nil / 空串 / 布尔 false / 数字零值
+func isEmptyErrorCode(v any) bool {
+	if v == nil {
+		return true
+	}
+	switch t := v.(type) {
+	case string:
+		return t == ""
+	case bool:
+		return !t
+	}
+	if isNumericType(v) {
+		return gconv.Float64(v) == 0
+	}
+	return false
+}
+
+// sameErrorValue 判断提取值是否等于配置的 defaultValue（数值/字符串跨类型兼容）
+func sameErrorValue(a, b any) bool {
+	if isNumericType(a) && isNumericType(b) {
+		return gconv.Float64(a) == gconv.Float64(b)
+	}
+	return gconv.String(a) == gconv.String(b)
+}
+
+// isNumericType 判断是否为数值类型
+func isNumericType(v any) bool {
+	if v == nil {
+		return false
+	}
+	switch reflect.TypeOf(v).Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64,
+		reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64,
+		reflect.Float32, reflect.Float64:
+		return true
+	}
+	return false
+}

service/payload.go

@@ -1,25 +0,0 @@
-package service
-
-import "github.com/gogf/gf/v2/util/gconv"
-
-// parseStoredPayload 解析入库的 request_payload，拆出模型调用 payload 与透传 headers
-// 入库格式：{"payload": <any>, "headers": {"Authorization": "...", "X-User-Info":"..."}}
-func parseStoredPayload(v any) (payload any, headers map[string]string) {
-	if v == nil {
-		return nil, nil
-	}
-	m := gconv.Map(v)
-	if len(m) == 0 {
-		return v, nil
-	}
-	if h, ok := m["headers"]; ok {
-		headers = gconv.MapStrStr(h)
-	}
-	if p, ok := m["payload"]; ok {
-		payload = p
-	} else {
-		payload = v
-	}
-	return
-}
-

service/pricing_client.go

@@ -0,0 +1,110 @@
+package service
+
+import (
+	"context"
+	"fmt"
+
+	commonHttp "gitea.redpowerfuture.com/red-future/common/http"
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	"github.com/gogf/gf/v2/frame/g"
+)
+
+// ====================== shop-user-trade 计价对接（独立 module，本地 JSON 对齐） ======================
+
+// shopPricingConfig shop-user-trade config/get 响应（仅取启用标记与最低余额门限；费率规则在 shop-user-trade 侧，model-gateway 不消费）
+type shopPricingConfig struct {
+	Enabled    int     `json:"enabled"`
+	MinBalance float64 `json:"minBalance" dc:"门禁：调用前可用余额须>=该值（元），0=不校验"`
+}
+
+// shopWalletAccount shop-user-trade wallet/account/get 响应
+type shopWalletAccount struct {
+	ID       int64   `json:"id"`
+	UserID   int64   `json:"userId"`
+	Balance  float64 `json:"balance" dc:"可用余额（元），负值=欠费"`
+	Currency string  `json:"currency"`
+	Status   int     `json:"status" dc:"状态：1启用 0禁用 -1冻结"`
+}
+
+// shopCalcFeeRes shop-user-trade /calc 响应
+type shopCalcFeeRes struct {
+	Cost float64 `json:"cost"`
+}
+
+// pricingURL 组装 shop-user-trade 计价接口地址（common http.RouteRegister 推导前缀，同 ai-agent billing.go）
+func pricingURL(sub string) string {
+	return "shop-user-trade/pricing/controller/" + sub
+}
+
+// walletURL 组装 shop-user-trade 钱包接口地址（accountController → account/controller，与 pricing 同 RouteRegister 推导规则）
+func walletURL(sub string) string {
+	return "shop-user-trade/account/controller/" + sub
+}
+
+// modelBillable 调用前门禁：模型须在 shop-user-trade 已配置且启用计价，否则阻塞调用。
+// 替换原 CheckTenantBalance（admin-go 租户余额门禁）；未配置→config/get 返回错误，未启用→enabled!=1。
+// minBalance>0 时追加最低余额门禁：钱包须存在且可用余额 >= 门限（与 shop-user-trade open_order 同语义，0=不校验）。
+func modelBillable(ctx context.Context, modelId int64) error {
+	var cfg shopPricingConfig
+	err := commonHttp.Get(ctx, pricingURL("config/get"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &cfg,
+		"subjectType", "model", "subjectId", fmt.Sprintf("%d", modelId))
+	if err != nil {
+		return fmt.Errorf("模型未配置计价，无法调用: %w", err)
+	}
+	if cfg.Enabled != 1 {
+		return fmt.Errorf("模型未启用计价，无法调用")
+	}
+	if cfg.MinBalance <= 0 {
+		return nil
+	}
+	user, e := utils.GetUserInfo(ctx)
+	if e != nil || user == nil || user.Id == 0 {
+		return fmt.Errorf("取不到用户，无法校验最低余额")
+	}
+	var acc shopWalletAccount
+	if e = commonHttp.Get(ctx, walletURL("get"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &acc,
+		"userId", fmt.Sprintf("%d", user.Id)); e != nil {
+		return fmt.Errorf("获取钱包失败，无法校验最低余额: %w", e)
+	}
+	if acc.Status != 1 {
+		return fmt.Errorf("钱包不可用，无法调用")
+	}
+	if acc.Balance < cfg.MinBalance {
+		return fmt.Errorf("余额不足：可用余额须不低于 %.2f 元才能发起", cfg.MinBalance)
+	}
+	return nil
+}
+
+// buildModelUsage 组装算费用量 JSON 对象（ChargeUsage 形状）。mediaType 为 shop 计费词汇
+// （audio/video，空=无媒体引用走默认价；DetectMediaType/异步快照已直接为该词汇，不再二次转换）。
+// per_char 模型把输出字数映射到 completionTokens（TokenMapping），随该字段传给 shop /calc 计价。
+func buildModelUsage(prompt, completion, cached int64, mediaType string, durationSec int64) map[string]any {
+	if durationSec < 0 {
+		durationSec = 0
+	}
+	return map[string]any{
+		"promptTokens":     prompt,
+		"completionTokens": completion,
+		"cachedTokens":     cached,
+		"mediaType":        mediaType,
+		"durationSec":      durationSec,
+	}
+}
+
+// calcModelCost 调 shop-user-trade /calc 按用量算费（不建单不扣费）。
+// 调用前门禁已保证配置存在；此处失败（配置中途删除/网络抖动）→ 记日志返回 0，不拖垮已完成的模型调用。
+func calcModelCost(ctx context.Context, modelId int64, usage map[string]any) float64 {
+	var res shopCalcFeeRes
+	err := commonHttp.Post(ctx, pricingURL("calc"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &res, &struct {
+		SubjectType string         `json:"subjectType"`
+		SubjectID   string         `json:"subjectId"`
+		Usage       map[string]any `json:"usage"`
+	}{
+		SubjectType: "model", SubjectID: fmt.Sprintf("%d", modelId), Usage: usage,
+	})
+	if err != nil {
+		g.Log().Errorf(ctx, "[算费] 调用 shop-user-trade 失败 modelId=%d: %v", modelId, err)
+		return 0
+	}
+	return res.Cost
+}

service/queue_gate.go

@@ -1,107 +0,0 @@
-package service
-
-import (
-	"context"
-	"fmt"
-	"math"
-	"time"
-
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-// ===== 严格 queue_limit：Redis 原子闸门 =====
-//
-// 背景：原来的 queue_limit 通过“Count + Insert”做近似控制，分布式并发创建时会短暂超限。
-// 目标：以 Redis Lua 脚本实现原子校验 + 入队占位，做到严格不超限。
-//
-// 计数口径与原逻辑保持一致：只统计 state=0/1（排队中/执行中）。
-// - CreateTask 成功入库后占用 1 个 slot
-// - 任务成功/失败（state->2/3）释放 slot
-// - 失败任务重试（state 3->0）需要再次占用 slot，若占位失败则暂不重试（留在 state=3，下次 cleaner 再尝试）
-//
-// 说明：为避免极端情况下“占位泄漏”导致永久占满，采用 ZSET + 过期时间的方式自动回收。
-// 只要任务实际生命周期远小于 gateTTLSeconds，就可保持严格。
-
-const (
-	queueGateKeyPrefix = "asynch:qgate:" // asynch:qgate:{modelName}
-)
-
-// Lua：清理过期 slot，然后按 limit 做原子判定并占位
-var queueGateAcquireLua = `
-local key = KEYS[1]
-local now = tonumber(ARGV[1])
-local limit = tonumber(ARGV[2])
-local expireAt = tonumber(ARGV[3])
-local member = ARGV[4]
-local keyTTL = tonumber(ARGV[5])
-
--- 先清理过期的占位
-redis.call("ZREMRANGEBYSCORE", key, "-inf", now)
-
-local current = tonumber(redis.call("ZCARD", key) or "0")
-if current >= limit then
-  return 0
-end
-redis.call("ZADD", key, expireAt, member)
-redis.call("EXPIRE", key, keyTTL)
-return 1
-`
-
-// Lua：释放 slot（幂等）
-var queueGateReleaseLua = `
-local key = KEYS[1]
-local member = ARGV[1]
-redis.call("ZREM", key, member)
-return 1
-`
-
-func queueGateKey(modelName string) string {
-	return fmt.Sprintf("%s%s", queueGateKeyPrefix, modelName)
-}
-
-// calcGateTTLSeconds 计算闸门占位的“自动回收 TTL”
-// 取 expectedSeconds 的倍数并做上下限，避免任务异常导致永久占位。
-func calcGateTTLSeconds(expectedSeconds int) int {
-	// 默认至少 1 小时；最多 24 小时
-	minTTL := 3600
-	maxTTL := 24 * 3600
-	if expectedSeconds <= 0 {
-		return minTTL
-	}
-	ttl := int(math.Ceil(float64(expectedSeconds) * 10)) // 预计耗时 * 10 做兜底
-	if ttl < minTTL {
-		ttl = minTTL
-	}
-	if ttl > maxTTL {
-		ttl = maxTTL
-	}
-	return ttl
-}
-
-// AcquireQueueSlot 严格入队：原子占位（成功返回 true）
-func AcquireQueueSlot(ctx context.Context, modelName, taskId string, limit int, expectedSeconds int) (bool, error) {
-	if limit <= 0 {
-		return true, nil
-	}
-	key := queueGateKey(modelName)
-	now := time.Now().Unix()
-	ttl := calcGateTTLSeconds(expectedSeconds)
-	expireAt := now + int64(ttl)
-	// keyTTL 要略大于 member TTL，避免 key 先过期导致计数丢失
-	keyTTL := ttl + 60
-	r, err := g.Redis().Do(ctx, "EVAL", queueGateAcquireLua, 1, key, now, limit, expireAt, taskId, keyTTL)
-	if err != nil {
-		return false, fmt.Errorf("queue gate acquire failed: %w", err)
-	}
-	return gconv.Int(r) == 1, nil
-}
-
-// ReleaseQueueSlot 释放占位（幂等）
-func ReleaseQueueSlot(ctx context.Context, modelName, taskId string) {
-	if taskId == "" || modelName == "" {
-		return
-	}
-	key := queueGateKey(modelName)
-	_, _ = g.Redis().Do(ctx, "EVAL", queueGateReleaseLua, 1, key, taskId)
-}

service/retry.go

@@ -0,0 +1,43 @@
+package service
+
+import (
+	"context"
+	"strings"
+	"time"
+
+	"gitea.redpowerfuture.com/red-future/common/oss"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// modelCallMaxRetries 上游调用最大重试次数
+const modelCallMaxRetries = 10
+
+// retryWait 指数退避等待（第 attempt 次重试，等待 1<<attempt 秒）。
+// 返回 nil 表示可继续重试；ctx 已取消返回 ctx.Err()，调用方应停止。
+func retryWait(ctx context.Context, attempt int) error {
+	wait := time.Duration(1<<attempt) * time.Second
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-time.After(wait):
+		return nil
+	}
+}
+
+// firstText 取任意值首位文本：数组取首个元素，其余原样转字符串
+func firstText(v any) string {
+	if arr, ok := v.([]any); ok && len(arr) > 0 {
+		return gconv.String(arr[0])
+	}
+	return gconv.String(v)
+}
+
+// extractChunkText 从流式分片字段取值并转存 OSS，返回首位文本（数组取首个元素，去掉首尾空白）。
+// 空值 / 仅空白 / 数组全空 返回空串。
+func extractChunkText(ctx context.Context, v any) string {
+	if v == nil || g.IsEmpty(v) {
+		return ""
+	}
+	return strings.TrimSpace(firstText(oss.TempURLToOSS(ctx, v)))
+}

service/runtime_tune.go

@@ -1,83 +0,0 @@
-package service
-
-import (
-	"context"
-
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-// 运行时调参存储在 Redis，不修改 asynch_models 中的 cap（最大上限）。
-// 上层每小时调用 /model/autoTune 写入运行时值；Worker/CreateTask 读取运行时值生效。
-
-const (
-	runtimeMaxCKeyPrefix   = "asynch:runtime:max_concurrency:" // + model_name
-	runtimeQueueKeyPrefix  = "asynch:runtime:queue_limit:"     // + model_name
-	runtimeTTLSeconds      = 2 * 3600                           // 2小时，避免一次调参失败导致立即回退
-)
-
-func runtimeMaxConcurrencyKey(modelName string) string {
-	return runtimeMaxCKeyPrefix + modelName
-}
-func runtimeQueueLimitKey(modelName string) string {
-	return runtimeQueueKeyPrefix + modelName
-}
-
-func getRuntimeInt(ctx context.Context, key string) (int, bool) {
-	v, err := g.Redis().Do(ctx, "GET", key)
-	if err != nil || v == nil {
-		return 0, false
-	}
-	iv := gconv.Int(v)
-	if iv <= 0 {
-		return 0, false
-	}
-	return iv, true
-}
-
-func setRuntimeInt(ctx context.Context, key string, val int) {
-	if val <= 0 {
-		return
-	}
-	// SETEX key ttl val
-	_, _ = g.Redis().Do(ctx, "SETEX", key, runtimeTTLSeconds, val)
-}
-
-// GetRuntimeMaxConcurrency 返回运行时并发上限（<= cap）。若不存在运行时值，则返回 cap。
-func GetRuntimeMaxConcurrency(ctx context.Context, modelName string, cap int) int {
-	if cap <= 0 {
-		return cap
-	}
-	if v, ok := getRuntimeInt(ctx, runtimeMaxConcurrencyKey(modelName)); ok {
-		if v > cap {
-			return cap
-		}
-		return v
-	}
-	return cap
-}
-
-// GetRuntimeQueueLimit 返回运行时队列上限（<= cap）。若不存在运行时值，则返回 cap。
-func GetRuntimeQueueLimit(ctx context.Context, modelName string, cap int) int {
-	if cap <= 0 {
-		return cap
-	}
-	if v, ok := getRuntimeInt(ctx, runtimeQueueLimitKey(modelName)); ok {
-		if v > cap {
-			return cap
-		}
-		return v
-	}
-	return cap
-}
-
-func clampInt(v, minV, maxV int) int {
-	if v < minV {
-		return minV
-	}
-	if v > maxV {
-		return maxV
-	}
-	return v
-}
-

service/schema_mapping_service.go

@@ -0,0 +1,260 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"reflect"
+	"regexp"
+	"strings"
+	"time"
+
+	"model-gateway/model/domain"
+	"model-gateway/model/dto"
+
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var SchemaMapping = &schemaMappingService{}
+
+type schemaMappingService struct{}
+
+// buildFieldDescriptions 从结构体中反射读取字段定义，构建提示词中的目标字段说明
+func buildFieldDescriptions(t reflect.Type) string {
+	var b strings.Builder
+	for i := 0; i < t.NumField(); i++ {
+		f := t.Field(i)
+		jsonName := f.Tag.Get("json")
+		desc := f.Tag.Get("dc")
+		typeName := f.Type.String()
+		if jsonName == "" || jsonName == "-" {
+			continue
+		}
+		if b.Len() > 0 {
+			b.WriteByte('\n')
+		}
+		b.WriteString("- **" + jsonName + "** (" + typeName + "): " + desc)
+	}
+	return b.String()
+}
+
+// getDomainTypeByModelType 根据模型类型返回对应的业务字段结构体反射类型
+// 如果找不到匹配，返回 nil
+func getDomainTypeByModelType(modelType int) reflect.Type {
+	switch modelType {
+	case 100, 101, 102, 103, 500, 501, 502, 503:
+		return reflect.TypeOf((*domain.ChatFieldsReq)(nil)).Elem()
+	case 600, 601, 602, 603, 604:
+		return reflect.TypeOf((*domain.VideoFields)(nil)).Elem()
+	default:
+		return nil
+	}
+}
+
+// BuildSchemaMapping 根据模型类型和 schema JSON，自动构建 schema_mapping（补充已有 mapping 的缺失字段）
+func (s *schemaMappingService) BuildSchemaMapping(ctx context.Context, req *dto.BuildSchemaMappingReq) (res *dto.BuildSchemaMappingRes, err error) {
+	if g.IsEmpty(req.Schema) {
+		return nil, fmt.Errorf("schema 不能为空")
+	}
+
+	// 1. 根据模型类型获取对应的业务字段结构体
+	domainType := getDomainTypeByModelType(req.ModelType)
+	if domainType == nil {
+		return nil, fmt.Errorf("不支持的模型类型: %d", req.ModelType)
+	}
+
+	// 3. 构建 LLM 提示词 输出的 JSON 对象键是 json 字段名。每个字段的值是定位到该位置的完整点号路径。
+	fieldDescs := buildFieldDescriptions(domainType)
+	systemPrompt := fmt.Sprintf(`你是一个 JSON Schema 分析助手。我提供了一个 AI API 的完整 Schema JSON 和待填充的目标结构体。
+请你仔细阅读 Schema 中所有字段的名称、类型、description 描述、枚举值、约束范围等完整信息，
+结合对 API 功能的理解，将目标结构体的每个字段映射到 Schema 中恰当的位置。
+
+## 输出格式
+
+输出的 JSON 对象键是 json 字段名。每个字段的值有两类：
+
+第一类（Schema 路径）：若该概念在 Schema 中有直接定义位置（约束值或字段定义），输出定位到该位置的完整点号路径。若定位的是对象数组的特定元素及其属性，在路径后追加 ?实际筛选字段名=筛选值&实际值字段名=# 格式，其中 =# 标记的目标值字段名替换为 schema 中的实际字段名。
+
+第二类（推导字符串）：若该概念在 Schema 中没有直接对应的定义位置，输出根据 Schema 信息推导出的内容字符串。
+
+## 重要规则
+
+1. 输出的每个字段都必须出现在 JSON 中，一个都不能少
+2. 若无法从 Schema 推理出某个字段的值，就输出空字符串 ""
+
+## 目标字段说明
+
+%s`, fieldDescs)
+
+	userPrompt := fmt.Sprintf("请分析以下 Schema JSON，生成对应的 schema_mapping：\n\n%s", req.Schema)
+
+	// 4. 调用 LLM
+	llmResp, err := callLLM(ctx, systemPrompt, userPrompt)
+	if err != nil {
+		return nil, err
+	}
+
+	// 5. 归一化所有路径为固定点号语法（无论模型输出哪种写法）
+	rawMap := gconv.Map(llmResp)
+	for k, v := range rawMap {
+		if s, ok := v.(string); ok {
+			rawMap[k] = normalizeSchemaPath(s)
+		}
+	}
+
+	return &dto.BuildSchemaMappingRes{
+		SchemaMapping: rawMap,
+	}, nil
+}
+
+// regNumIndexStar 匹配数字下标 [0]、[1] 等
+var regNumIndexStar = regexp.MustCompile(`\[\d+]`)
+
+// normalizeSchemaPath 将 LLM 生成的 Schema 路径统一为固定点号语法：
+//   - 移除模板包装字段 attrs / properties / items / defaultValue / required
+//   - enumValues、items 及 attrs[数字] 标记上一字段为数组，补 [*]
+//   - [数字] 下标统一转为 [*]
+//
+// 示例：
+//
+//	messages.attrs.enumValues.attrs.content.enumValues?type=image_url&image_url.url=#
+//	  → messages[*].content[*]?type=image_url&image_url.url=#
+//	choices.attrs[0].attrs.message.attrs.content → choices[*].message.content
+func normalizeSchemaPath(p string) string {
+	path, suffix := p, ""
+	if i := strings.Index(p, "?"); i >= 0 {
+		path, suffix = p[:i], p[i:]
+	}
+	segs := strings.Split(path, ".")
+	var out []string
+	for _, seg := range segs {
+		seg = strings.TrimSpace(seg)
+		switch {
+		case seg == "":
+			continue
+		case seg == "attrs" || seg == "properties" || seg == "defaultValue" || seg == "required":
+			continue
+		case seg == "enumValues" || seg == "items" || (strings.HasPrefix(seg, "attrs[") && regNumIndexStar.MatchString(seg)):
+			markPrevAsArray(&out)
+			continue
+		}
+		seg = regNumIndexStar.ReplaceAllString(seg, "[*]")
+		out = append(out, seg)
+	}
+	return strings.Join(out, ".") + suffix
+}
+
+// markPrevAsArray 将输出序列最后一个字段标记为数组（补 [*]）
+func markPrevAsArray(out *[]string) {
+	if len(*out) == 0 {
+		return
+	}
+	last := (*out)[len(*out)-1]
+	if !strings.HasSuffix(last, "[]") && !strings.HasSuffix(last, "[*]") {
+		(*out)[len(*out)-1] = last + "[*]"
+	}
+}
+
+// callLLM 调用大模型聊天接口（OpenAI 兼容格式）。
+// 模型地址/密钥走配置 schemaMapping 段，本地开发无配置时用默认值兜底。
+func callLLM(ctx context.Context, systemPrompt, userPrompt string) (string, error) {
+	modelName := g.Cfg().MustGet(ctx, "schemaMapping.modelName", "doubao-seed-2-0-lite-260428").String()
+	baseURL := g.Cfg().MustGet(ctx, "schemaMapping.baseUrl", "https://ark.cn-beijing.volces.com/api/v3/chat/completions").String()
+	apiKey := g.Cfg().MustGet(ctx, "schemaMapping.apiKey", "ark-9df744e8-a0de-4c54-9db3-18379bccd523-e6733").String()
+
+	body := map[string]any{
+		"model": modelName,
+		"messages": []map[string]string{
+			{"role": "system", "content": systemPrompt},
+			{"role": "user", "content": userPrompt},
+		},
+		"max_tokens":  2048,
+		"temperature": 0.1,
+	}
+
+	jsonBody, err := json.Marshal(body)
+	if err != nil {
+		return "", fmt.Errorf("marshal request body failed: %w", err)
+	}
+
+	url := strings.TrimRight(baseURL, "/")
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonBody))
+	if err != nil {
+		return "", fmt.Errorf("create request failed: %w", err)
+	}
+	httpReq.Header.Set("Authorization", "Bearer "+apiKey)
+	httpReq.Header.Set("Content-Type", "application/json")
+
+	client := &http.Client{Timeout: 120 * time.Second}
+	resp, err := client.Do(httpReq)
+	if err != nil {
+		return "", fmt.Errorf("request failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return "", fmt.Errorf("read response failed (status=%d): %w", resp.StatusCode, err)
+	}
+
+	if resp.StatusCode != 200 {
+		return "", fmt.Errorf("API error status=%d body=%s", resp.StatusCode, string(respBody))
+	}
+
+	var apiResp struct {
+		Choices []struct {
+			Message struct {
+				Content string `json:"content"`
+			} `json:"message"`
+		} `json:"choices"`
+		Error *struct {
+			Message string `json:"message"`
+		} `json:"error,omitempty"`
+	}
+
+	if err = json.Unmarshal(respBody, &apiResp); err != nil {
+		return "", fmt.Errorf("parse response failed: %s", string(respBody))
+	}
+
+	if apiResp.Error != nil {
+		return "", fmt.Errorf("API error: %s", apiResp.Error.Message)
+	}
+
+	if len(apiResp.Choices) == 0 {
+		return "", fmt.Errorf("empty response")
+	}
+
+	return apiResp.Choices[0].Message.Content, nil
+}
+
+// extractJSONObject 从字符串中提取第一个完整的 JSON 对象（{...}）
+func extractJSONObject(s string) string {
+	start := strings.Index(s, "{")
+	if start < 0 {
+		return s
+	}
+	for start > 0 {
+		ch := s[start-1]
+		if ch == ' ' || ch == '\t' || ch == '\n' || ch == '\r' {
+			start--
+		} else {
+			break
+		}
+	}
+	end := strings.LastIndex(s, "}")
+	if end <= start {
+		return s
+	}
+
+	snippet := s[start : end+1]
+	snippet = strings.TrimPrefix(snippet, "```json")
+	snippet = strings.TrimPrefix(snippet, "```")
+	snippet = strings.TrimSuffix(snippet, "```")
+	snippet = strings.TrimSpace(snippet)
+	return snippet
+}

service/semaphore.go

@@ -1,56 +0,0 @@
-package service
-
-import (
-	"context"
-	"fmt"
-
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-)
-
-var acquireLua = `
-local current = tonumber(redis.call("GET", KEYS[1]) or "0")
-local max = tonumber(ARGV[1])
-local ttl = tonumber(ARGV[2])
-if current >= max then
-  return 0
-end
-current = redis.call("INCR", KEYS[1])
-if current == 1 then
-  redis.call("EXPIRE", KEYS[1], ttl)
-end
-if current > max then
-  redis.call("DECR", KEYS[1])
-  return 0
-end
-return 1
-`
-
-var releaseLua = `
-local current = tonumber(redis.call("DECR", KEYS[1]) or "0")
-if current <= 0 then
-  redis.call("DEL", KEYS[1])
-end
-return 1
-`
-
-func acquireSemaphore(ctx context.Context, key string, max int, ttlSeconds int64) (bool, error) {
-	if max <= 0 {
-		// 不限制
-		return true, nil
-	}
-	if ttlSeconds <= 0 {
-		ttlSeconds = 3600
-	}
-	r, err := g.Redis().Do(ctx, "EVAL", acquireLua, 1, key, max, ttlSeconds)
-	if err != nil {
-		return false, fmt.Errorf("获取并发令牌失败: %w", err)
-	}
-	return gconv.Int(r) == 1, nil
-}
-
-func releaseSemaphore(ctx context.Context, key string) error {
-	_, err := g.Redis().Do(ctx, "EVAL", releaseLua, 1, key)
-	return err
-}
-

service/session_stream.go

@@ -0,0 +1,362 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"model-gateway/dao"
+	"model-gateway/model/domain"
+	"model-gateway/model/dto"
+	"model-gateway/service/httpclient"
+	modelUtils "model-gateway/service/utils"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// CreateSessionStreamOnce 流式调用上游模型 → 缓冲全量后一次返回（走 gf 框架正常返回）。
+// 与同步请求一致：上游返回错误时按 shouldRetryWithMemory 判定是否指数退避重试（最多 modelCallMaxRetries 次）。
+func (s *modelSessionService) CreateSessionStreamOnce(ctx context.Context, req *dto.CallModelSessionReq) (docMsg *dto.ModelCallRes, err error) {
+	startTime := time.Now()
+
+	id := req.Id
+	modelInfo := req.ModelInfo
+	newRequestParams := req.RequestParams
+
+	attempt := 0
+LOOP:
+	// 获取上游流式 reader（stream=false → w 不会被使用，传 nil）。
+	// 非 2xx 状态/网络错误在此返回；按 shouldRetryWithMemory 判定是否指数退避重试，与同步请求一致。
+	streamReader, err := httpclient.ModelHttpStreamRequest(ctx, nil, modelInfo.BaseURL, modelInfo.RequestHeadMapping, modelInfo.HttpMethod, newRequestParams)
+	if err != nil {
+		if attempt < modelCallMaxRetries {
+			if code, msg := streamErrorInfoOfError(err); shouldRetryWithMemory(ctx, modelInfo, code, msg, "") {
+				attempt++
+				wait := time.Duration(1<<attempt) * time.Second
+				g.Log().Warningf(ctx, "模型流式请求异常，第 %d 次重试(等待 %v): code=%s err=%v", attempt+1, wait, code, err)
+				if waitErr := retryWait(ctx, attempt); waitErr != nil {
+					recordSessionError(context.WithoutCancel(ctx), id, startTime, "调用取消: "+waitErr.Error())
+					return nil, waitErr
+				}
+				goto LOOP
+			}
+		}
+		// 非重试错误/重试耗尽：请求失败即返回，把失败信息写入模型会话记录
+		recordSessionError(ctx, id, startTime, err.Error())
+		return nil, err
+	}
+
+	docMsg = new(dto.ModelCallRes)
+	docMsg.TaskId = id
+	var contentBuf strings.Builder
+
+	// 记录流内 error 事件（OpenAI 兼容 error 分片），供流结束后统一判定重试/报错
+	var streamErrCode, streamErrMsg, streamErrBody string
+
+	// 路径预处理
+	respMapping := make(map[string]string, len(modelInfo.ResponseBodyMapping))
+	for k, _ := range modelInfo.ResponseBodyMapping {
+		respMapping[k] = modelUtils.CleanFieldPath(k)
+	}
+	totalTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.TotalTokens)
+	promptTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.PromptTokens)
+	completionTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.CompletionTokens)
+
+	httpclient.ParseSSEStream(ctx, streamReader, func(ctx context.Context, chunk map[string]any) error {
+		// 流内错误事件（OpenAI 兼容 error 分片）：暂存错误码/消息，不做内容累加，由流结束后统一判定
+		if code, msg := streamErrorOfChunk(chunk); code != "" {
+			streamErrCode, streamErrMsg, streamErrBody = code, msg, gconv.String(chunk["error"])
+			return nil
+		}
+		// 基于统一字段路径（GetByPath）在分片对象上取值，取首个数组元素文本累加
+		for _, jsonPath := range respMapping {
+			if realText := extractChunkText(ctx, modelUtils.GetByPathValue(chunk, jsonPath)); realText != "" {
+				contentBuf.WriteString(realText)
+			}
+		}
+
+		// Token 累加
+		docMsg.TotalTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, totalTokenPath))
+		docMsg.PromptTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, promptTokenPath))
+		docMsg.CompletionTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, completionTokenPath))
+		return nil
+	})
+
+	// 流内返回错误：丢弃本次部分内容，指数退避后重新请求（判定交给 shouldRetryWithMemory）
+	if streamErrCode != "" {
+		if attempt < modelCallMaxRetries && shouldRetryWithMemory(ctx, modelInfo, streamErrCode, streamErrMsg, streamErrBody) {
+			attempt++
+			wait := time.Duration(1<<attempt) * time.Second
+			g.Log().Warningf(ctx, "模型流式调用异常，第 %d 次重试(等待 %v): code=%s msg=%s", attempt+1, wait, streamErrCode, streamErrMsg)
+			if waitErr := retryWait(ctx, attempt); waitErr != nil {
+				recordSessionError(context.WithoutCancel(ctx), id, startTime, "调用取消: "+waitErr.Error())
+				return nil, waitErr
+			}
+			goto LOOP
+		}
+		// 与同步一致：不可重试的错误码记录到 ErrorMsg 后正常走组装返回，不中断流程
+		docMsg.ErrorMsg = streamErrMsg
+	}
+
+	// 流结束后组装（流内出错时内容为空，与同步一致不再组装/上传空内容）
+	if streamErrCode == "" {
+		for k, _ := range modelInfo.ResponseBodyMapping {
+			docMsg.Content = map[string]any{k: contentBuf.String()}
+		}
+	}
+
+	// 补充更新会话记录
+	updateModelSessionReq := dto.UpdateModelSessionReq{
+		Id:               id,
+		DurationSeconds:  int64(time.Since(startTime).Seconds()),
+		TotalTokens:      docMsg.TotalTokens,
+		PromptTokens:     docMsg.PromptTokens,
+		CompletionTokens: docMsg.CompletionTokens,
+		ErrorMsg:         docMsg.ErrorMsg,
+	}
+	if !g.IsEmpty(docMsg.Content) {
+		uploadNewResp, uploadErr := Upload(ctx, &dto.UploadFileBytesReq{
+			FileBytes: gconv.Bytes(gconv.String(docMsg.Content)),
+			FileName:  fmt.Sprintf("modelNewRespParams:%v.json", time.Now().UnixMilli()),
+		})
+		if uploadErr != nil {
+			return nil, fmt.Errorf("上传模型返回参数文件失败：%v", uploadErr)
+		}
+		updateModelSessionReq.ResponsePath = uploadNewResp.FileURL
+	}
+	updateModelSessionReq.DurationSeconds = int64(time.Since(startTime).Seconds())
+	// 调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	docMsg.ModelId = modelInfo.Id // 引用行=系统模型 id，供 per_token 结算按系统模型计价
+	docMsg.MediaType = mediaType
+	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
+	updateModelSessionReq.TotalCost = docMsg.Cost
+	if _, updateErr := dao.ModelSession.Update(ctx, &updateModelSessionReq); updateErr != nil {
+		g.Log().Errorf(ctx, "更新流式会话信息失败: %v", updateErr)
+	}
+
+	return docMsg, nil
+}
+
+// CreateSessionStream 流式调用上游模型 → SSE 逐分片推送给前端；流结束返回本次调用的 token/费用（docMsg.Cost）供调用方扣减。
+func (s *modelSessionService) CreateSessionStream(ctx context.Context, w http.ResponseWriter, req *dto.CallModelSessionReq) (*dto.ModelCallRes, error) {
+	startTime := time.Now()
+
+	id := req.Id
+	modelInfo := req.ModelInfo
+	newRequestParams := req.RequestParams
+
+	// 获取上游流式 reader 并设置 SSE 响应头
+	streamReader, err := httpclient.ModelHttpStreamRequest(ctx, w, modelInfo.BaseURL, modelInfo.RequestHeadMapping, modelInfo.HttpMethod, newRequestParams)
+	if err != nil {
+		// 请求建立前失败：把错误写入模型会话记录，避免留半截无错误信息记录
+		recordSessionError(ctx, id, startTime, err.Error())
+		return nil, err
+	}
+
+	flusher := w.(http.Flusher)
+
+	docMsg := new(dto.ModelCallRes)
+	var contentBuf strings.Builder
+
+	// tool_calls 按 index 累加（OpenAI 兼容 delta），流末随 done 事件一次性返回
+	toolAcc := make(map[int]*streamToolCallAcc)
+
+	// 路径预处理
+	respMapping := make(map[string]string, len(modelInfo.ResponseBodyMapping))
+	for k, _ := range modelInfo.ResponseBodyMapping {
+		respMapping[modelUtils.CleanFieldPath(k)] = modelUtils.CleanFieldPath(k)
+	}
+	totalTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.TotalTokens)
+	promptTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.PromptTokens)
+	completionTokenPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.CompletionTokens)
+
+	// 解析 ResponseBusinessFieldMapping 字段
+	businessFieldRes := new(domain.ChatFieldsRes)
+	err = gconv.Struct(modelInfo.ResponseBusinessFieldMapping, businessFieldRes)
+	if err != nil {
+		g.Log().Errorf(ctx, "解析 ResponseBusinessFieldMapping 失败: %v", err)
+	}
+	// tool_calls 读取路径：优先走 ResponseBusinessFieldMapping 的 tools 配置，缺省兜底 OpenAI 兼容路径
+	//toolsPath := "choices[0].delta.tool_calls"
+	toolsPath := businessFieldRes.Tools
+	// reasoning_content 读取路径：走 ResponseBusinessFieldMapping 的 reasoning_content 配置，未配置则不返回思考内容
+	reasoningPath := modelUtils.CleanFieldPath(businessFieldRes.ReasoningContent)
+
+	httpclient.ParseSSEStream(ctx, streamReader, func(ctx context.Context, chunk map[string]any) error {
+		// 基于统一字段路径（GetByPath）在分片对象上取值，取首个数组元素文本
+		content := make(map[string]any, len(respMapping))
+		for bizKey, jsonPath := range respMapping {
+			if realText := extractChunkText(ctx, modelUtils.GetByPathValue(chunk, jsonPath)); realText != "" {
+				content[bizKey] = realText
+				contentBuf.WriteString(realText)
+			}
+		}
+
+		// Token 累加（记录增量：usage 常在无文本/思考的末分片出现，需据此放行推送）
+		prevTotal, prevPrompt, prevCompletion := docMsg.TotalTokens, docMsg.PromptTokens, docMsg.CompletionTokens
+		docMsg.TotalTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, totalTokenPath))
+		docMsg.PromptTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, promptTokenPath))
+		docMsg.CompletionTokens += gconv.Int64(modelUtils.GetByPathValue(chunk, completionTokenPath))
+		tokenDelta := docMsg.TotalTokens != prevTotal || docMsg.PromptTokens != prevPrompt || docMsg.CompletionTokens != prevCompletion
+
+		accumulateStreamToolCallsByPath(chunk, toolsPath, toolAcc)
+
+		// 思考内容提取（独立业务字段，不进回答全文）
+		var reasoningContent string
+		if reasoningPath != "" {
+			if v := modelUtils.GetByPathValue(chunk, reasoningPath); v != nil && !g.IsEmpty(v) {
+				reasoningContent = firstText(v)
+			}
+		}
+
+		// 纯 token 分片（无文本/思考）也放行：否则末分片 usage 被过滤，调用方拿不到 token 值
+		if len(content) == 0 && reasoningContent == "" && !tokenDelta {
+			return nil
+		}
+
+		// 逐 chunk SSE 推送给前端（字段名由 ModelCallStreamEvent 统一管理）
+		event := &dto.ModelCallStreamEvent{
+			Content:          content,
+			ReasoningContent: reasoningContent,
+			TotalTokens:      docMsg.TotalTokens,
+			PromptTokens:     docMsg.PromptTokens,
+			CompletionTokens: docMsg.CompletionTokens,
+		}
+		outBytes, err := json.Marshal(event)
+		if err != nil {
+			g.Log().Errorf(ctx, "[SSE] marshal response failed: %v", err)
+			return nil
+		}
+		_, err = fmt.Fprintf(w, "data: %s\n\n", outBytes)
+		if err != nil {
+			g.Log().Errorf(ctx, "[SSE] write client failed: %v", err)
+			return err
+		}
+		flusher.Flush()
+		return nil
+	})
+
+	// 流结束：调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	docMsg.ModelId = modelInfo.Id // 引用行=系统模型 id，供 per_token 结算按系统模型计价
+	docMsg.MediaType = mediaType
+	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
+
+	// 流末 done 事件：携带该步最终 token 与费用。工具调用时附带完整 tool_calls，
+	// 纯文本流同样补发，使调用方拿到最终费用与 token；不识别 type=done 的消费方忽略该事件。
+	event := &dto.ModelCallStreamEvent{
+		Type:             "done",
+		TotalTokens:      docMsg.TotalTokens,
+		PromptTokens:     docMsg.PromptTokens,
+		CompletionTokens: docMsg.CompletionTokens,
+		Cost:             docMsg.Cost,
+	}
+	if tools := finalizeStreamToolCalls(toolAcc); len(tools) > 0 {
+		var toolModels []dto.ModelTool
+		if err := gconv.Structs(tools, &toolModels); err != nil {
+			g.Log().Errorf(ctx, "[SSE] convert tools failed: %v", err)
+		} else {
+			event.Tools = toolModels
+		}
+	}
+	outBytes, err := json.Marshal(event)
+	if err == nil {
+		_, _ = fmt.Fprintf(w, "data: %s\n\n", outBytes)
+		flusher.Flush()
+	}
+
+	// 流结束后补充更新会话记录
+	updateModelSessionReq := dto.UpdateModelSessionReq{
+		Id:               id,
+		DurationSeconds:  int64(time.Since(startTime).Seconds()),
+		TotalTokens:      docMsg.TotalTokens,
+		PromptTokens:     docMsg.PromptTokens,
+		CompletionTokens: docMsg.CompletionTokens,
+		TotalCost:        docMsg.Cost,
+	}
+	if !g.IsEmpty(contentBuf.String()) {
+		uploadNewResp, uploadErr := Upload(ctx, &dto.UploadFileBytesReq{
+			FileBytes: gconv.Bytes(gconv.String(map[string]any{"respBody": contentBuf.String()})),
+			FileName:  fmt.Sprintf("modelNewRespParams:%v.json", time.Now().UnixMilli()),
+		})
+		if uploadErr != nil {
+			return nil, fmt.Errorf("上传模型返回参数文件失败：%v", uploadErr)
+		}
+		updateModelSessionReq.ResponsePath = uploadNewResp.FileURL
+	}
+	updateModelSessionReq.DurationSeconds = int64(time.Since(startTime).Seconds())
+	if _, updateErr := dao.ModelSession.Update(ctx, &updateModelSessionReq); updateErr != nil {
+		return nil, fmt.Errorf("更新会话信息失败: %v", updateErr)
+	}
+
+	return docMsg, nil
+}
+
+// recordSessionError 请求建立前失败（上游不可达/非 2xx 且非重试/重试耗尽/调用取消）时，
+// 把错误与耗时写入模型会话记录，避免流式调用留半截无错误信息记录。
+// 仅写 ErrorMsg/DurationSeconds（OmitEmpty 不会影响已落库字段）；ctx 已取消时须传 WithoutCancel(ctx)。
+func recordSessionError(ctx context.Context, id int64, startTime time.Time, errMsg string) {
+	if _, updateErr := dao.ModelSession.Update(ctx, &dto.UpdateModelSessionReq{
+		Id:              id,
+		DurationSeconds: int64(time.Since(startTime).Seconds()),
+		ErrorMsg:        errMsg,
+	}); updateErr != nil {
+		g.Log().Errorf(ctx, "更新模型会话错误信息失败: %v", updateErr)
+	}
+}
+
+// streamErrorOfChunk 从流式分片提取错误码与消息：优先 OpenAI 兼容 error 事件，顶层 code 兜底。
+func streamErrorOfChunk(chunk map[string]any) (code, msg string) {
+	if errObj := gconv.Map(chunk["error"]); errObj != nil {
+		code = gconv.String(errObj["code"])
+		msg = gconv.String(errObj["message"])
+	}
+	if code == "" {
+		code = gconv.String(chunk["code"])
+	}
+	return
+}
+
+// streamErrorInfoOfError 从流式请求错误中提取错误码与消息（不做固定清单过滤，交由 shouldRetryWithMemory 判定）：
+// 优先解析错误体 error.code/顶层 code 与 message，其次取非 2xx 的 HTTP 状态码。
+// 纯网络错误等无错误码场景返回空串。
+func streamErrorInfoOfError(err error) (code, msg string) {
+	if err == nil {
+		return "", ""
+	}
+	e := err.Error()
+	// 非 2xx 时 httpclient.ModelHttpStreamRequest 返回 "[HTTP][Stream] 状态码异常: %d, body={...}"
+	if idx := strings.Index(e, "body="); idx >= 0 {
+		body := e[idx+len("body="):]
+		var errResp struct {
+			Error struct {
+				Code    string `json:"code"`
+				Message string `json:"message"`
+			} `json:"error"`
+			Code    string `json:"code"`
+			Message string `json:"message"`
+		}
+		if json.Unmarshal([]byte(body), &errResp) == nil {
+			if errResp.Error.Code != "" {
+				return errResp.Error.Code, errResp.Error.Message
+			}
+			if errResp.Code != "" {
+				return errResp.Code, errResp.Message
+			}
+		}
+	}
+	if idx := strings.Index(e, "状态码异常: "); idx >= 0 {
+		codeStr := strings.TrimSpace(e[idx+len("状态码异常: "):])
+		if comma := strings.IndexByte(codeStr, ','); comma >= 0 {
+			codeStr = codeStr[:comma]
+		}
+		return codeStr, ""
+	}
+	return "", ""
+}

service/session_sync.go

@@ -0,0 +1,135 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"model-gateway/consts/model"
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/service/httpclient"
+	modelUtils "model-gateway/service/utils"
+	"time"
+
+	"gitea.redpowerfuture.com/red-future/common/oss"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var ModelSession = &modelSessionService{}
+
+type modelSessionService struct{}
+
+// CreateSession 创建会话（同步调用，非流式）
+func (s *modelSessionService) CreateSession(ctx context.Context, req *dto.CallModelSessionReq) (res *dto.ModelCallRes, err error) {
+	startTime := time.Now()
+	attempt := 0
+	id := req.Id
+	modelInfo := req.ModelInfo
+	newRequestParams := req.RequestParams
+LOOP:
+	// 6) 模型请求
+	modelRespBody, err := httpclient.ModelHttpNormalRequest(ctx, modelInfo.BaseURL, modelInfo.RequestHeadMapping, modelInfo.HttpMethod, newRequestParams)
+	if err != nil {
+		return nil, err
+	}
+	if modelRespBody == nil {
+		return nil, fmt.Errorf("模型返回参数是空")
+	}
+	// 7) 上传模型返回参数文件
+	uploadOriginalResp, err := Upload(ctx, &dto.UploadFileBytesReq{
+		FileBytes: modelRespBody,
+		FileName:  fmt.Sprintf("modelRespParams:%v.json", time.Now().UnixMilli()),
+	})
+	if err != nil {
+		return nil, fmt.Errorf("上传模型返回参数文件失败：%v", err)
+	}
+	// 8) 更新模型会话信息
+	updateModelSessionReq := dto.UpdateModelSessionReq{
+		Id:                   id,
+		OriginalResponsePath: uploadOriginalResp.FileURL,
+	}
+	docMsg := new(dto.ModelCallRes)
+	docMsg.TaskId = id
+	errCode, errMsg, err := parseModelError(modelRespBody, modelInfo.ErrorMessageMapping)
+	if err != nil {
+		return nil, fmt.Errorf("模型返回参数解析失败：%v", err)
+	}
+	if errCode != "" {
+		if attempt < modelCallMaxRetries && shouldRetryWithMemory(ctx, modelInfo, errCode, errMsg, string(modelRespBody)) {
+			attempt++
+			wait := time.Duration(1<<attempt) * time.Second
+			g.Log().Warningf(ctx, "模型上游调用异常，第 %d 次重试(等待 %v): code=%s err=%v", attempt+1, wait, errCode, errMsg)
+			if waitErr := retryWait(ctx, attempt); waitErr != nil {
+				return nil, waitErr
+			}
+			goto LOOP
+		}
+		docMsg.ErrorMsg = errMsg
+		updateModelSessionReq.ErrorMsg = docMsg.ErrorMsg
+	} else {
+		if *model.ResponseTypeSync.Code() == *modelInfo.ResponseType {
+			respBodyMap := make(map[string]string, len(modelInfo.ResponseBodyMapping))
+			for k, _ := range modelInfo.ResponseBodyMapping {
+				respBodyMap[k] = modelUtils.CleanFieldPath(k)
+			}
+			// 基于统一字段路径（GetByPath）按映射取值组装结果
+			var respObj map[string]any
+			if err = json.Unmarshal(modelRespBody, &respObj); err != nil {
+				return nil, fmt.Errorf("模型返回参数解析失败：%v", err)
+			}
+			content := make(map[string]any, len(respBodyMap))
+			for bizKey, jsonPath := range respBodyMap {
+				content[bizKey] = oss.TempURLToOSS(ctx, modelUtils.GetByPathValue(respObj, jsonPath))
+			}
+
+			businessField := make(map[string]any, len(modelInfo.ResponseBusinessFieldMapping))
+			for key, value := range modelInfo.ResponseBusinessFieldMapping {
+				businessField[key] = modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(value))
+			}
+			err = gconv.Struct(businessField, docMsg)
+			if err != nil {
+				return nil, fmt.Errorf("模型返回参数解析失败：%v", err)
+			}
+
+			docMsg.Content = content
+			docMsg.TotalTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(modelInfo.TokenMapping.TotalTokens)))
+			docMsg.PromptTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(modelInfo.TokenMapping.PromptTokens)))
+			docMsg.CompletionTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(modelInfo.TokenMapping.CompletionTokens)))
+
+			updateModelSessionReq.PromptTokens = docMsg.PromptTokens
+			updateModelSessionReq.CompletionTokens = docMsg.CompletionTokens
+			updateModelSessionReq.TotalTokens = docMsg.TotalTokens
+		} else {
+			docMsg.Content = map[string]any{
+				"respBody": modelRespBody,
+			}
+		}
+	}
+	if !g.IsEmpty(docMsg.Content) {
+		// 9) 上传模型返回参数文件
+		uploadNewResp, err := Upload(ctx, &dto.UploadFileBytesReq{
+			FileBytes: gconv.Bytes(gconv.String(docMsg.Content)),
+			FileName:  fmt.Sprintf("modelNewRespParams:%v.json", time.Now().UnixMilli()),
+		})
+		if err != nil {
+			return nil, fmt.Errorf("上传模型返回参数文件失败：%v", err)
+		}
+		updateModelSessionReq.ResponsePath = uploadNewResp.FileURL
+	}
+	updateModelSessionReq.DurationSeconds = int64(time.Since(startTime).Seconds())
+	// 9.5) 调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	docMsg.ModelId = modelInfo.Id // 引用行=系统模型 id，供 per_token 结算按系统模型计价
+	docMsg.MediaType = mediaType
+	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
+	updateModelSessionReq.TotalCost = docMsg.Cost
+	// 10) 更新模型会话信息
+	_, err = dao.ModelSession.Update(ctx, &updateModelSessionReq)
+	if err != nil {
+		return nil, fmt.Errorf("更新模型会话信息失败: %v", err)
+	}
+
+	return docMsg, nil
+}

service/stat_service.go

@@ -1,39 +0,0 @@
-package service
-
-import (
-	"context"
-
-	"model-gateway/dao"
-	"model-gateway/model/dto"
-)
-
-type statService struct{}
-
-var Stat = &statService{}
-
-func (s *statService) List(ctx context.Context, req *dto.ListModelStatReq) (res *dto.ListModelStatRes, err error) {
-	pageNum, pageSize := 1, 10
-	if req != nil {
-		if req.PageNum > 0 {
-			pageNum = req.PageNum
-		}
-		if req.PageSize > 0 {
-			pageSize = req.PageSize
-		}
-	}
-	startDay, endDay := "", ""
-	var tenantID *int64
-	creator, modelName := "", ""
-	if req != nil {
-		startDay = req.StartDay
-		endDay = req.EndDay
-		tenantID = req.TenantID
-		creator = req.Creator
-		modelName = req.ModelName
-	}
-	list, total, err := dao.Stat.List(ctx, pageNum, pageSize, startDay, endDay, tenantID, creator, modelName)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.ListModelStatRes{List: list, Total: total}, nil
-}

service/storage.go

@@ -1,18 +0,0 @@
-package service
-
-import (
-	"context"
-	"errors"
-
-	"model-gateway/model/entity"
-)
-
-// StorageService 结果存储（OSS/MinIO）抽象
-type StorageService interface {
-	UploadByTask(ctx context.Context, t *entity.AsynchTask, data []byte, fileExt string, contentType string) (ossURL string, err error)
-}
-
-// Storage 默认存储实现（优先对接你们的 oss 文件服务；必要时也可以切到 MinIO）
-var Storage StorageService = &ossStorage{}
-
-var ErrStorageNotConfigured = errors.New("存储未配置")

service/storage_oss.go

@@ -1,81 +0,0 @@
-package service
-
-import (
-	"bytes"
-	"context"
-	"fmt"
-	"mime/multipart"
-	"time"
-
-	"model-gateway/model/entity"
-
-	commonHttp "gitea.com/red-future/common/http"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
-	"github.com/gogf/gf/v2/util/guid"
-)
-
-// 对接你们的 oss 文件服务：POST oss/file/uploadFile (multipart/form-data)
-type ossStorage struct{}
-
-type uploadFileResponse struct {
-	FileURL           string `json:"fileURL"`           // 文件 URL
-	FileSize          int    `json:"fileSize"`          // 文件大小（字节）
-	FileName          string `json:"fileName"`          // 文件名
-	FileFormat        string `json:"fileFormat"`        // 文件格式
-	FileAddressPrefix string `json:"fileAddressPrefix"` // 文件地址前缀
-}
-
-func (s *ossStorage) UploadByTask(ctx context.Context, _ *entity.AsynchTask, data []byte, fileExt string, _ string) (ossURL string, err error) {
-	// multipart
-	body := &bytes.Buffer{}
-	writer := multipart.NewWriter(body)
-
-	ext := fileExt
-	if ext == "" {
-		ext = ".bin"
-	}
-	if ext[0] != '.' {
-		ext = "." + ext
-	}
-
-	filename := fmt.Sprintf("asynch_%d_%s%s", time.Now().Unix(), guid.S(), ext)
-	part, err := writer.CreateFormFile("file", filename)
-	if err != nil {
-		return "", err
-	}
-	if _, err := part.Write(data); err != nil {
-		return "", err
-	}
-	contentType := writer.FormDataContentType()
-	if err := writer.Close(); err != nil {
-		return "", err
-	}
-
-	headers := forwardHeaders(ctx)
-	headers["Content-Type"] = contentType
-
-	fullURL := "oss/file/uploadFile"
-	g.Log().Infof(ctx, "[OSS] upload start url=%s filename=%s size=%d", fullURL, filename, len(data))
-
-	var resp uploadFileResponse
-	if err := commonHttp.Post(ctx, fullURL, headers, &resp, body.Bytes()); err != nil {
-		return "", err
-	}
-	g.Log().Infof(ctx, "[OSS] upload success url=%s size=%d format=%s", resp.FileURL, resp.FileSize, resp.FileFormat)
-	return resp.FileURL, nil
-}
-
-// setTaskHeadersToCtx 把任务入库时保存的 header 信息注入 ctx，给 worker 调 OSS 用
-func setTaskHeadersToCtx(ctx context.Context, headers map[string]string) context.Context {
-	if headers == nil {
-		return ctx
-	}
-	if v := gconv.String(headers["Authorization"]); v != "" {
-		ctx = context.WithValue(ctx, "token", v)
-	}
-	if v := gconv.String(headers["X-User-Info"]); v != "" {
-		ctx = context.WithValue(ctx, "xUserInfo", v)
-	}
-	return ctx
-}

service/stream_tool_calls.go

@@ -0,0 +1,101 @@
+package service
+
+import (
+	"sort"
+	"strings"
+
+	modelUtils "model-gateway/service/utils"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// streamToolCallAcc 流式 tool_call 按 index 累加的碎片（OpenAI 兼容 delta 格式）
+type streamToolCallAcc struct {
+	id     string
+	typ    string
+	fnName string
+	fnArgs strings.Builder
+}
+
+// streamToolCallDelta OpenAI 兼容流式 tool_call 增量片段，字段名集中于此
+type streamToolCallDelta struct {
+	Index    int    `json:"index"`
+	Id       string `json:"id"`
+	Type     string `json:"type"`
+	Function struct {
+		Name      string `json:"name"`
+		Arguments string `json:"arguments"`
+	} `json:"function"`
+}
+
+// toStreamToolCallDeltas 把 JSON 反序列化的 any 数组在边界转成强类型片段
+func toStreamToolCallDeltas(rawCalls []any) []streamToolCallDelta {
+	var deltas []streamToolCallDelta
+	if err := gconv.Structs(rawCalls, &deltas); err != nil {
+		return nil
+	}
+	return deltas
+}
+
+// accumulateStreamToolCallsByPath 按配置路径从 chunk 读取 tool_calls 数组并累加。
+// 路径未命中或值非数组时无副作用（不创建任何槽位）。
+func accumulateStreamToolCallsByPath(chunk map[string]any, toolsPath string, acc map[int]*streamToolCallAcc) {
+	raw := modelUtils.GetByPathValue(chunk, modelUtils.CleanFieldPath(toolsPath))
+	rawCalls, _ := raw.([]any)
+	accumulateToolCallFragments(toStreamToolCallDeltas(rawCalls), acc)
+}
+
+// accumulateToolCallFragments 按 index 累加 tool_calls 增量片段：
+// id/type/function.name 首片段补齐，function.arguments 为字符串片段需按 index 拼接。
+func accumulateToolCallFragments(rawCalls []streamToolCallDelta, acc map[int]*streamToolCallAcc) {
+	for _, d := range rawCalls {
+		slot, ok := acc[d.Index]
+		if !ok {
+			slot = &streamToolCallAcc{}
+			acc[d.Index] = slot
+		}
+		if d.Id != "" {
+			slot.id = d.Id
+		}
+		if d.Type != "" {
+			slot.typ = d.Type
+		}
+		if d.Function.Name != "" {
+			slot.fnName = d.Function.Name
+		}
+		slot.fnArgs.WriteString(d.Function.Arguments)
+	}
+}
+
+// finalizeStreamToolCalls 把累加结果按 index 升序转为 []map[string]any，形状对齐 dto.ModelTool。
+// 无有效工具返回 nil。
+func finalizeStreamToolCalls(acc map[int]*streamToolCallAcc) []map[string]any {
+	if len(acc) == 0 {
+		return nil
+	}
+	idx := make([]int, 0, len(acc))
+	for i := range acc {
+		idx = append(idx, i)
+	}
+	sort.Ints(idx)
+	tools := make([]map[string]any, 0, len(idx))
+	for _, i := range idx {
+		s := acc[i]
+		fn := map[string]any{}
+		if s.fnName != "" {
+			fn["name"] = s.fnName
+		}
+		if s.fnArgs.Len() > 0 {
+			fn["arguments"] = s.fnArgs.String()
+		}
+		tool := map[string]any{"function": fn}
+		if s.id != "" {
+			tool["id"] = s.id
+		}
+		if s.typ != "" {
+			tool["type"] = s.typ
+		}
+		tools = append(tools, tool)
+	}
+	return tools
+}

service/task_service.go

@@ -1,266 +0,0 @@
-package service
-
-import (
-	"context"
-	"errors"
-	"fmt"
-	"time"
-
-	"model-gateway/dao"
-	"model-gateway/model/dto"
-	"model-gateway/model/entity"
-
-	"github.com/gogf/gf/v2/database/gdb"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/os/gtime"
-	"github.com/google/uuid"
-)
-
-var Task = &taskService{}
-
-type taskService struct{}
-
-func (s *taskService) Create(ctx context.Context, req *dto.CreateTaskReq) (res *dto.CreateTaskRes, err error) {
-	fmt.Printf("打印请求:%+v", req)
-	startAt := time.Now()
-	// 固化 token/user 等信息
-	ctx = asyncCtx(ctx)
-
-	// 1) 检查模型配置
-	m, err := dao.Model.GetByModelName(ctx, req.ModelName)
-	if err != nil {
-		return nil, err
-	}
-	if m == nil || (m.Enabled != nil && *m.Enabled != 1) {
-		return nil, errors.New("模型不存在或未启用")
-	}
-
-	taskID := uuid.NewString()
-	// 2) 排队上限（严格控制：Redis 原子闸门）
-	limit := GetRuntimeQueueLimit(ctx, req.ModelName, m.QueueLimit)
-	if limit > 0 {
-		ok, err := AcquireQueueSlot(ctx, req.ModelName, taskID, limit, m.ExpectedSeconds)
-		if err != nil {
-			return nil, err
-		}
-		if !ok {
-			return nil, errors.New("任务排队已满，请稍后再试")
-		}
-	}
-
-	// 将调用模型的 payload 与透传头信息一起存入 request_payload，供后台 worker 使用
-	storedPayload := map[string]any{
-		"payload": req.RequestPayload,
-		"headers": forwardHeaders(ctx),
-	}
-
-	t := &entity.AsynchTask{
-		ModelName:      req.ModelName,
-		TaskID:         taskID,
-		State:          0,
-		BizName:        req.BizName,
-		CallbackURL:    req.CallbackUrl,
-		ModelKey:       m.ApiKey,
-		InputRef:       req.InputRef,
-		RequestPayload: storedPayload,
-		EpicycleId:     req.EpicycleId,
-	}
-	_, err = dao.Task.Insert(ctx, t)
-	if err != nil {
-		// 入库失败：回滚闸门占位
-		ReleaseQueueSlot(ctx, req.ModelName, taskID)
-		return nil, err
-	}
-
-	// 3) 写操作日志（尽量不影响主流程，失败忽略）
-	ip := ""
-	ua := ""
-	apiPath := "/task/createTask"
-	httpMethod := "POST"
-	if r := g.RequestFromCtx(ctx); r != nil {
-		ip = r.GetClientIp()
-		ua = r.UserAgent()
-		apiPath = r.URL.Path
-		httpMethod = r.Method
-	}
-	_, _ = dao.OpLog.Insert(ctx, &entity.LogsModelOp{
-		IP:             ip,
-		UserAgent:      ua,
-		APIPath:        apiPath,
-		HttpMethod:     httpMethod,
-		BizName:        req.BizName,
-		ModelName:      req.ModelName,
-		TaskID:         taskID,
-		OpType:         "createTask",
-		Success:        1,
-		ErrorMsg:       "",
-		CostMs:         time.Since(startAt).Milliseconds(),
-		RequestPayload: storedPayload,
-		ResponsePayload: gdb.Map{
-			"taskId": taskID,
-		},
-	})
-
-	// 4) 创建成功后立即异步尝试执行当前任务，并仅在任务仍处于 pending(state=0) 时做定向轮询。
-	// 一旦任务进入 running/success/failed/downloaded，就停止轮询，避免一直空转。
-	go s.pollAndRunUntilPicked(context.WithoutCancel(ctx), taskID, req.EpicycleId)
-
-	return &dto.CreateTaskRes{TaskID: taskID}, nil
-}
-
-// pollAndRunUntilPicked 用于 createTask 创建后的“轻量级定向轮询”：
-// - 目标：尽快把刚创建的任务拉起来执行
-// - 只在任务仍为 pending(state=0) 时继续尝试抢占
-// - 一旦任务进入 running(1) / success(2) / failed(3) / downloaded(4)，立即停止
-// - 这样不会无限轮询；runWork 仍负责处理积压队列和未处理到的任务
-func (s *taskService) pollAndRunUntilPicked(ctx context.Context, taskID string, epicycleId int64) {
-	if taskID == "" {
-		return
-	}
-	interval := g.Cfg().MustGet(ctx, "asynch.worker.intervalSeconds").Int()
-	if interval <= 0 {
-		interval = 5
-	}
-	g.Log().Infof(ctx, "[task-auto-run][start] taskId=%s interval=%ds", taskID, interval)
-
-	ticker := time.NewTicker(time.Duration(interval) * time.Second)
-	defer ticker.Stop()
-
-	tryRun := func() bool {
-		t, err := dao.Task.GetByTaskID(ctx, taskID)
-		if err != nil {
-			g.Log().Warningf(ctx, "[task-auto-run][stop] taskId=%s reason=query_failed err=%v", taskID, err)
-			return true
-		}
-		if t == nil {
-			g.Log().Warningf(ctx, "[task-auto-run][stop] taskId=%s reason=task_not_found", taskID)
-			return true
-		}
-		switch t.State {
-		case 0:
-			if err := AsyncWorker.RunByTaskID(ctx, taskID, epicycleId); err != nil {
-				g.Log().Warningf(ctx, "[task-auto-run][retry] taskId=%s state=0 err=%v", taskID, err)
-			} else {
-				g.Log().Infof(ctx, "[task-auto-run][triggered] taskId=%s state=0", taskID)
-			}
-			return false
-		case 1:
-			g.Log().Infof(ctx, "[task-auto-run][stop] taskId=%s reason=running", taskID)
-			return true
-		case 2, 3, 4:
-			g.Log().Infof(ctx, "[task-auto-run][stop] taskId=%s reason=terminal state=%d", taskID, t.State)
-			return true
-		default:
-			g.Log().Infof(ctx, "[task-auto-run][stop] taskId=%s reason=unknown_state state=%d", taskID, t.State)
-			return true
-		}
-	}
-
-	// 先立即尝试一次
-	if stop := tryRun(); stop {
-		return
-	}
-
-	for {
-		select {
-		case <-ctx.Done():
-			g.Log().Infof(ctx, "[task-auto-run][stop] taskId=%s reason=context_done", taskID)
-			return
-		case <-ticker.C:
-			if stop := tryRun(); stop {
-				return
-			}
-		}
-	}
-}
-
-func (s *taskService) GetResult(ctx context.Context, taskID string) (res *dto.GetTaskResultRes, err error) {
-	t, err := dao.Task.GetByTaskID(ctx, taskID)
-	if err != nil {
-		return nil, err
-	}
-	if t == nil {
-		return nil, errors.New("任务不存在")
-	}
-	return &dto.GetTaskResultRes{
-		OssFile: t.OssFile,
-		State:   t.State,
-	}, nil
-}
-
-// GetBatch 批量查询任务；将成功(state=2)的任务更新为已下载(state=4)，并写入过期时间
-func (s *taskService) GetBatch(ctx context.Context, req *dto.GetTaskBatchReq) (res *dto.GetTaskBatchRes, err error) {
-	if req == nil || len(req.TaskIDs) == 0 {
-		return &dto.GetTaskBatchRes{List: []dto.GetTaskBatchItem{}}, nil
-	}
-	// 1) 先查当前租户下的任务列表
-	list, err := dao.Task.ListByTaskIDs(ctx, req.TaskIDs)
-	if err != nil {
-		return nil, err
-	}
-
-	// 2) 对成功(state=2)的任务：标记为已下载(state=4)并写入 expire_at
-	now := time.Now()
-	for _, t := range list {
-		if t == nil {
-			continue
-		}
-		if t.State != 2 {
-			continue
-		}
-		// 按模型配置决定保留时间
-		m, err := dao.Model.GetByModelName(ctx, t.ModelName)
-		if err != nil {
-			return nil, err
-		}
-		retainSeconds := 86400
-		if m != nil && m.AutoCleanSeconds > 0 {
-			retainSeconds = m.AutoCleanSeconds
-		}
-		expireAt := gtime.New(now.Add(time.Duration(retainSeconds) * time.Second))
-		_ = dao.Task.MarkDownloadedByID(ctx, t.Id, expireAt)
-
-		// 为了本次返回一致性，内存里也更新
-		t.State = 4
-		t.ExpireAt = expireAt
-	}
-
-	// 3) 组装返回
-	items := make([]dto.GetTaskBatchItem, 0, len(list))
-	for _, t := range list {
-		if t == nil {
-			continue
-		}
-		items = append(items, dto.GetTaskBatchItem{
-			TaskID:  t.TaskID,
-			State:   t.State,
-			OssFile: t.OssFile,
-		})
-	}
-	return &dto.GetTaskBatchRes{List: items}, nil
-}
-
-func (s *taskService) List(ctx context.Context, req *dto.ListTaskReq) (res *dto.ListTaskRes, err error) {
-	pageNum, pageSize := 1, 10
-	if req != nil {
-		if req.PageNum > 0 {
-			pageNum = req.PageNum
-		}
-		if req.PageSize > 0 {
-			pageSize = req.PageSize
-		}
-	}
-	modelName := ""
-	taskID := ""
-	var state *int
-	if req != nil {
-		modelName = req.ModelName
-		taskID = req.TaskID
-		state = req.State
-	}
-	list, total, err := dao.Task.List(ctx, pageNum, pageSize, modelName, taskID, state)
-	if err != nil {
-		return nil, err
-	}
-	return &dto.ListTaskRes{List: list, Total: total}, nil
-}

service/tmp_store.go

@@ -1,38 +0,0 @@
-package service
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-)
-
-// 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”。
-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
-	}
-	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 path, nil
-}
-
-func loadTmpResult(path string) ([]byte, error) {
-	return os.ReadFile(path)
-}
-
-func deleteTmpResult(path string) {
-	if path == "" {
-		return
-	}
-	_ = os.Remove(path)
-}
-

service/utils.go

@@ -1,113 +0,0 @@
-package service
-
-import (
-	"encoding/json"
-	"strings"
-
-	"github.com/gogf/gf/v2/container/gvar"
-)
-
-func normalizeFormValue(v any) any {
-	// 目标：对外永远返回 JSON 数组/对象，而不是字符串。
-	if v == nil {
-		return []any{}
-	}
-	switch t := v.(type) {
-	case string:
-		s := strings.TrimSpace(t)
-		if s == "" {
-			return []any{}
-		}
-		return normalizeFormValueFromJSONString(s)
-	case []byte:
-		if len(t) == 0 {
-			return []any{}
-		}
-		return normalizeFormValueFromJSONBytes(t)
-	case *gvar.Var:
-		// goframe 常见的 DB 返回类型
-		if t == nil {
-			return []any{}
-		}
-		b := t.Bytes()
-		if len(b) > 0 {
-			return normalizeFormValueFromJSONBytes(b)
-		}
-		s := strings.TrimSpace(t.String())
-		if s == "" {
-			return []any{}
-		}
-		return normalizeFormValueFromJSONString(s)
-	default:
-		// 尝试兼容其他“像 JSON 的值类型”（例如实现了 Bytes/String 的包装类型）
-		if vb, ok := v.(interface{ Bytes() []byte }); ok {
-			if b := vb.Bytes(); len(b) > 0 {
-				return normalizeFormValueFromJSONBytes(b)
-			}
-		}
-		if vs, ok := v.(interface{ String() string }); ok {
-			if s := strings.TrimSpace(vs.String()); s != "" {
-				return normalizeFormValueFromJSONString(s)
-			}
-		}
-		// 已经是 []any / map[string]any 等结构
-		return v
-	}
-}
-
-// 兼容“JSONB 里存了 JSON 字符串”的历史数据：
-// 例如 form_json = '"[]"' 或 '"[{...}]"'（外层是字符串，内层才是数组/对象）
-func normalizeFormValueFromJSONString(s string) any {
-	var out any
-	if err := json.Unmarshal([]byte(s), &out); err != nil || out == nil {
-		return []any{}
-	}
-	// 如果解出来还是 string，且看起来是 JSON，再解一层
-	if inner, ok := out.(string); ok {
-		inner = strings.TrimSpace(inner)
-		if inner == "" {
-			return []any{}
-		}
-		if strings.HasPrefix(inner, "[") || strings.HasPrefix(inner, "{") {
-			var out2 any
-			if err := json.Unmarshal([]byte(inner), &out2); err == nil && out2 != nil {
-				return out2
-			}
-		}
-		return []any{}
-	}
-	return out
-}
-
-func normalizeFormValueFromJSONBytes(b []byte) any {
-	var out any
-	if err := json.Unmarshal(b, &out); err != nil || out == nil {
-		return []any{}
-	}
-	// bytes 解出来也可能是 string（同上）
-	if inner, ok := out.(string); ok {
-		return normalizeFormValueFromJSONString(inner)
-	}
-	return out
-}
-
-func ParseJSONField(field any) any {
-	var v *gvar.Var
-	switch val := field.(type) {
-	case *gvar.Var:
-		v = val
-	default:
-		return field
-	}
-
-	if v == nil || v.IsNil() || v.IsEmpty() {
-		return nil
-	}
-
-	str := v.String()
-	var result any
-	if json.Unmarshal([]byte(str), &result) == nil {
-		return result
-	}
-	return str
-}

service/utils/business_fields.go

@@ -0,0 +1,628 @@
+package utils
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// ============================================================
+// 业务字段读写：TakeBusinessFields 把 businessParams 按映射解析为写入路径，
+// WriteBusinessFields 按路径写入最终请求体（路径语法见 SetByPath）。
+// ============================================================
+
+// TakeBusinessFields 把业务参数（businessParams）按映射解析为写入路径：
+//   - 调用方按业务字段名（RequestBusinessFieldMapping 的 key）传值，这里是独立的 businessParams map，
+//     不再与模板字段混在 requestParams 中
+//   - 业务字段名未配置映射 → 返回错误（不静默忽略）
+//   - 解包 {type,value} 包裹格式为原始值
+//   - 跳过空值（空串/空数组），避免写入请求体污染
+//
+// 返回 map[映射路径]原始值，构建完成后由 WriteBusinessFields 按路径写入请求体。
+func TakeBusinessFields(businessParams map[string]any, bizMapping map[string]string) (map[string]any, error) {
+	if len(businessParams) == 0 {
+		return nil, nil
+	}
+	keyToPath := make(map[string]string, len(bizMapping))
+	for key, path := range bizMapping {
+		if key != "" && path != "" {
+			keyToPath[key] = path
+		}
+	}
+	bizValues := make(map[string]any)
+	for key, raw := range businessParams {
+		path, isBiz := keyToPath[key]
+		if !isBiz {
+			return nil, fmt.Errorf("业务字段 [%s] 未配置映射（RequestBusinessFieldMapping 中不存在该业务字段名）", key)
+		}
+		v := unwrapBizValue(raw)
+		if isEmptyBizValue(v) {
+			continue
+		}
+		bizValues[path] = v
+	}
+	return bizValues, nil
+}
+
+// unwrapBizValue 解包模板包裹格式 {type, value}，返回原始值；非包裹格式原样返回
+func unwrapBizValue(v any) any {
+	if m, ok := v.(map[string]any); ok {
+		if _, hasType := m["type"]; hasType {
+			if val, hasVal := m["value"]; hasVal {
+				return val
+			}
+		}
+	}
+	return v
+}
+
+// isEmptyBizValue 判断业务字段值是否为空（空值不写入请求体）
+func isEmptyBizValue(v any) bool {
+	switch t := v.(type) {
+	case nil:
+		return true
+	case string:
+		return t == ""
+	case []any:
+		return len(t) == 0
+	case map[string]any:
+		return len(t) == 0
+	}
+	return false
+}
+
+// ============================================================
+// 统一字段路径语法（读/写共用，见 NormalizeFieldPath）：
+//
+//	a.b.c                       普通点号路径
+//	a[*].b                      [*] 表示数组段
+//	a[*].b[*]?k=v&t=#           单层选择器：在数组中按 k==v 匹配元素，值/读取目标为 t
+//	a[*]?k=v&b[*]?k2=v2&t=#     多级选择器：选择器体内可再嵌 [*]?选择器，级数不限。
+//	                            每级 k=v 既是匹配条件（命中已存在元素时），
+//	                            也是新建元素时写入该元素的字段（如 role=user 直接落为 role 字段）；
+//	                            只有带 t=# 的那级是叶子目标（写值/读值的位置）。
+//
+// SetByPath（写，构建请求体）与 GetByPath（读，解析响应）共用 parsePath；
+// 读方向语义：数组段非末段取第 0 个元素继续下钻，[*] 为末段返回整个数组，选择器定位匹配元素；
+// 写方向语义：数组段非末段作用于最后一个元素，末段前置追加（业务值在前），选择器 upsert（命中更新/未命中新建），
+// 多值（[]any）仅在叶子选择器展开为多个独立元素（多个参考图/视频等）。
+// ============================================================
+
+// SetByPath 按业务字段映射路径把值写入请求结构（请求侧构建）。
+// 路径语法与 BuildSchemaMapping 输出一致（干净形态，无需 attrs 剔除）。
+//
+// 写入语义（业务字段一律前置追加，不覆盖已有值）：
+//   - 目标字段已存在且是数组 → 业务值（或其元素）前置插入数组头部，原元素依次后移
+//   - 目标字段已存在且非数组（普通叶子路径）→ 业务值前置拼接（字符串拼接/数组包裹），原值保留在后
+//   - 目标字段不存在 → 新建
+//   - 数组段/选择器段目标字段不是数组（如字符串 content）→ 返回错误，不覆盖已有值
+//   - 中间路径遇到非对象字段 → 返回错误
+//   - 数组段无选择器且非末尾 → 作用于最后一个元素（追加语义），数组为空则补一个空元素
+//   - 数组段带选择器 → 命中则更新目标字段，未命中则按选择器字段构造新元素追加；
+//     选择器段即使未标 [*] 也按数组处理（如 input.media?type=first_frame&url=#）
+//   - 多级选择器 → 递归：中间级选择器定位/新建容器元素并继续下钻，叶子选择器写值
+//   - 值本身是数组 → 叶子选择器逐值追加；普通数组段/点号路径前置追加（业务值在前，原值保留在后，不覆盖）
+func SetByPath(root map[string]any, path string, value any) error {
+	steps := parsePath(NormalizeFieldPath(path))
+	if len(steps) == 0 {
+		return nil
+	}
+	return setBySteps(root, steps, value)
+}
+
+// setBySteps 按步骤序列写入；选择器步骤（可能带嵌套）递归处理，非选择器步骤逐层下钻
+func setBySteps(cur map[string]any, steps []step, value any) error {
+	first := steps[0]
+	last := len(steps) == 1
+	if first.sel != nil {
+		// 选择器段：目标字段按数组处理（upsert），路径段未标 [*] 也按数组匹配
+		arr, err := existingArray(cur, first.key)
+		if err != nil {
+			return err
+		}
+		newArr, err := upsertStep(arr, first, value)
+		if err != nil {
+			return err
+		}
+		cur[first.key] = newArr
+		return nil
+	}
+	if !first.isArray {
+		if last {
+			setLeaf(cur, first.key, value)
+			return nil
+		}
+		next, err := ensureMap(cur, first.key)
+		if err != nil {
+			return err
+		}
+		return setBySteps(next, steps[1:], value)
+	}
+	// 数组段（无选择器）
+	arr, err := existingArray(cur, first.key)
+	if err != nil {
+		return err
+	}
+	if last {
+		cur[first.key] = prependToArray(arr, value)
+		return nil
+	}
+	// 无选择器数组段：作用于最后一个元素（追加语义）
+	if len(arr) == 0 {
+		arr = append(arr, map[string]any{})
+		cur[first.key] = arr
+	}
+	lastElem, ok := arr[len(arr)-1].(map[string]any)
+	if !ok {
+		return fmt.Errorf("字段 [%s] 数组元素不是对象，无法继续下钻（当前类型 %T）", first.key, arr[len(arr)-1])
+	}
+	return setBySteps(lastElem, steps[1:], value)
+}
+
+// WriteBusinessFields 把业务字段值写入最终请求体。
+// bizValues 的键为映射路径（如 input.media?type=reference_video&url=#），值由调用方按路径传入。
+// 按字典序升序写入：父路径是子路径的前缀（短者靠前），保证容器先写、子路径再 upsert，
+// 避免子路径先建出的结构被父路径整体覆盖（如 messages 容器与 messages[*].content[*] 内嵌目标并存）。
+// 任一路径写入失败（如数组段目标不是数组）→ 返回错误，由调用方拒绝本次请求。
+func WriteBusinessFields(out map[string]any, bizValues map[string]any) error {
+	paths := make([]string, 0, len(bizValues))
+	for path := range bizValues {
+		paths = append(paths, path)
+	}
+	sort.Strings(paths)
+	for _, path := range paths {
+		if err := SetByPath(out, path, bizValues[path]); err != nil {
+			return fmt.Errorf("业务字段写入失败 [%s]: %w", path, err)
+		}
+	}
+	return nil
+}
+
+// GetByPath 按字段路径读取响应值（与 SetByPath 同一套路径语法，读方向语义）：
+//   - 普通段：逐层进入对象取字段
+//   - 数组段 [*]：非末段取数组第 0 个元素继续下钻；[*] 为末段返回整个数组
+//   - 选择器段 ?k=v&t=#：定位 k==v 的元素，返回该元素 t 字段的值；多级选择器递归下钻
+//
+// 未命中（路径缺失 / 中间类型不符）返回 (nil, nil)，不视为错误；语法错误返回 error。
+func GetByPath(root map[string]any, path string) (any, error) {
+	steps := parsePath(NormalizeFieldPath(path))
+	if len(steps) == 0 {
+		return nil, nil
+	}
+	return getBySteps(root, steps)
+}
+
+// getBySteps 按步骤序列读取；选择器步骤（可能带嵌套）递归处理
+func getBySteps(cur any, steps []step) (any, error) {
+	if len(steps) == 0 {
+		return nil, nil
+	}
+	first := steps[0]
+	rest := steps[1:]
+	m, ok := cur.(map[string]any)
+	if !ok {
+		return nil, nil
+	}
+	// 选择器段：定位匹配元素，返回叶子目标或递归嵌套下钻
+	if first.sel != nil {
+		arr, ok := m[first.key].([]any)
+		if !ok {
+			return nil, nil
+		}
+		for _, e := range arr {
+			em, ok := e.(map[string]any)
+			if !ok {
+				continue
+			}
+			if matchFilters(em, first.sel) {
+				return getSelValue(em, first.sel, rest)
+			}
+		}
+		return nil, nil
+	}
+	v, ok := m[first.key]
+	if !ok {
+		return nil, nil
+	}
+	if first.isArray {
+		arr, ok := v.([]any)
+		if !ok {
+			return nil, nil
+		}
+		if len(rest) == 0 {
+			return arr, nil
+		}
+		if len(arr) == 0 {
+			return nil, nil
+		}
+		return getBySteps(arr[0], rest)
+	}
+	if len(rest) == 0 {
+		return v, nil
+	}
+	return getBySteps(v, rest)
+}
+
+// getSelValue 选择器命中元素后取值：有嵌套路径则递归下钻，否则取叶子目标字段
+// （target 可为点号路径，如 image_url.url=#，按 parseSteps 拆级下钻）
+func getSelValue(em map[string]any, sel *selNode, rest []step) (any, error) {
+	if len(sel.nested) > 0 {
+		return getBySteps(em, append(sel.nested, rest...))
+	}
+	if sel.target != "" {
+		return getBySteps(em, append(parseSteps(sel.target), rest...))
+	}
+	return getBySteps(em, rest)
+}
+
+// GetByPathValue 读取路径值，未命中或出错返回 nil（免去调用方处理双返回值）
+func GetByPathValue(root map[string]any, path string) any {
+	v, err := GetByPath(root, path)
+	if err != nil {
+		return nil
+	}
+	return v
+}
+
+// GetByPathAll 按字段路径读取响应值（与 GetByPath 同一套语法），返回路径下**所有**命中值。
+// 与 GetByPath 的区别：GetByPath 命中即返回第一个匹配；GetByPathAll 遍历数组段/选择器段的全部
+// 元素并展开收集。适用于通配路径（messages[*]...[*]...）取全部匹配值（如收集所有图片 url）。
+// 无命中返回 nil。
+func GetByPathAll(root map[string]any, path string) []any {
+	steps := parsePath(NormalizeFieldPath(path))
+	if len(steps) == 0 {
+		return nil
+	}
+	return getAllBySteps(root, steps)
+}
+
+// getAllBySteps 按步骤序列收集全部匹配值；数组段/选择器段遍历所有元素展开，普通段单值包裹返回
+func getAllBySteps(cur any, steps []step) []any {
+	if len(steps) == 0 {
+		return nil
+	}
+	first := steps[0]
+	rest := steps[1:]
+	m, ok := cur.(map[string]any)
+	if !ok {
+		return nil
+	}
+	// 选择器段：遍历命中元素收集
+	if first.sel != nil {
+		arr, ok := m[first.key].([]any)
+		if !ok {
+			return nil
+		}
+		var out []any
+		for _, e := range arr {
+			em, ok := e.(map[string]any)
+			if !ok {
+				continue
+			}
+			if matchFilters(em, first.sel) {
+				out = append(out, getSelValueAll(em, first.sel, rest)...)
+			}
+		}
+		return out
+	}
+	v, ok := m[first.key]
+	if !ok {
+		return nil
+	}
+	if first.isArray {
+		arr, ok := v.([]any)
+		if !ok {
+			return nil
+		}
+		if len(rest) == 0 {
+			return arr
+		}
+		var out []any
+		for _, e := range arr {
+			out = append(out, getAllBySteps(e, rest)...)
+		}
+		return out
+	}
+	if len(rest) == 0 {
+		return []any{v}
+	}
+	return getAllBySteps(v, rest)
+}
+
+// getSelValueAll 选择器命中元素后收集：有嵌套路径递归下钻，否则取叶子目标字段（全部）
+func getSelValueAll(em map[string]any, sel *selNode, rest []step) []any {
+	if len(sel.nested) > 0 {
+		return getAllBySteps(em, append(sel.nested, rest...))
+	}
+	if sel.target != "" {
+		return getAllBySteps(em, append(parseSteps(sel.target), rest...))
+	}
+	return getAllBySteps(em, rest)
+}
+
+// step 路径段；sel 非空表示该段带选择器（按数组处理）
+type step struct {
+	key     string
+	isArray bool
+	sel     *selNode
+}
+
+// selNode 选择器：
+//   - filters：k=v 匹配条件，新建元素时也作为字段写入
+//   - target：叶子目标字段（k=#），值/读取目标；target 为空且 nested 非空时为中间级选择器
+//   - nested：下钻子路径（多级嵌套选择器，级数不限）
+type selNode struct {
+	filters [][2]string
+	target  string
+	nested  []step
+}
+
+// parsePath 解析路径为步骤序列。选择器体挂到最后一个步骤上；选择器体中的嵌套 [*]?选择器
+// 递归解析为 nested（级数不限）。
+func parsePath(p string) []step {
+	base, suffix := p, ""
+	if i := strings.Index(p, "?"); i >= 0 {
+		base, suffix = p[:i], p[i+1:]
+	}
+	steps := parseSteps(base)
+	if suffix != "" {
+		parseSelector(suffix, &steps)
+	}
+	return steps
+}
+
+// parseSteps 解析点号分隔的普通步骤（含 [*] 数组段）
+func parseSteps(s string) []step {
+	var steps []step
+	for _, raw := range strings.Split(s, ".") {
+		raw = strings.TrimSpace(raw)
+		if raw == "" {
+			continue
+		}
+		st := step{}
+		if strings.HasSuffix(raw, "[*]") {
+			st.key = strings.TrimSuffix(raw, "[*]")
+			st.isArray = true
+		} else {
+			st.key = raw
+		}
+		steps = append(steps, st)
+	}
+	return steps
+}
+
+// parseSelector 解析选择器体（? 之后的内容）并挂到最后一个步骤上。
+// 元素用顶层 & 分隔（? 之后的 & 属于嵌套选择器）；k=v 为过滤/写入对，k=# 为叶子目标，
+// 含 [*] 或路径的块为嵌套下钻子路径（递归 parsePath）。
+func parseSelector(selStr string, steps *[]step) {
+	if len(*steps) == 0 {
+		return
+	}
+	sel := &selNode{}
+	var nested []step
+	for _, el := range splitTopLevel(selStr) {
+		if isPair(el) {
+			k, v, _ := strings.Cut(el, "=")
+			if v == "#" {
+				sel.target = k
+			} else {
+				sel.filters = append(sel.filters, [2]string{k, v})
+			}
+			continue
+		}
+		// 嵌套路径（含自己的选择器）：级数不限，递归解析
+		nested = append(nested, parsePath(el)...)
+	}
+	if len(sel.filters) == 0 && sel.target == "" && len(nested) == 0 {
+		return
+	}
+	last := &(*steps)[len(*steps)-1]
+	if last.sel == nil {
+		last.sel = sel
+	}
+	if len(nested) > 0 {
+		last.sel.nested = nested
+	}
+}
+
+// splitTopLevel 按顶层 & 拆分选择器体；? 之后的 & 属于嵌套选择器，不在此层拆分
+func splitTopLevel(s string) []string {
+	var elems []string
+	var cur strings.Builder
+	inNested := false
+	for _, ch := range s {
+		if ch == '?' {
+			inNested = true
+		}
+		if ch == '&' && !inNested {
+			elems = append(elems, cur.String())
+			cur.Reset()
+			continue
+		}
+		cur.WriteRune(ch)
+	}
+	if cur.Len() > 0 {
+		elems = append(elems, cur.String())
+	}
+	return elems
+}
+
+// isPair 判断元素是否为 k=v 对：= 出现在任何 [ ? 之前则是 pair，否则为嵌套路径。
+// 目标字段 k 本身可以是点号路径（image_url.url=#），故 . 不参与判别。
+func isPair(el string) bool {
+	for i := 0; i < len(el); i++ {
+		switch el[i] {
+		case '=':
+			return true
+		case '[', '?':
+			return false
+		}
+	}
+	return false
+}
+
+// existingArray 返回数组字段的当前数组：
+//   - 字段不存在 → 空数组（允许按追加语义新建）
+//   - 字段是数组 → 原样
+//   - 字段是其他类型（如字符串 content）→ 返回错误，调用方拒绝写入，不覆盖已有值
+func existingArray(cur map[string]any, key string) ([]any, error) {
+	v, ok := cur[key]
+	if !ok {
+		return []any{}, nil
+	}
+	if arr, ok := v.([]any); ok {
+		return arr, nil
+	}
+	return nil, fmt.Errorf("字段 [%s] 不是数组，无法按数组路径写入（当前类型 %T）", key, v)
+}
+
+// upsertStep 选择器 upsert，返回追加后的数组：
+//   - 叶子选择器（有 target）且值为数组 → 每个值追加一个独立元素
+//   - 命中（所有 filters 匹配）→ 把值写入现有元素（叶子写 target，中间级递归 nested）
+//   - 未命中 → 按选择器字段构造新元素并追加
+//
+// 返回新切片（append 可能重新分配底层数组），调用方需用返回值覆盖写回。
+func upsertStep(arr []any, st step, value any) ([]any, error) {
+	sel := st.sel
+	// 叶子选择器：多值逐个展开为独立元素（多个参考图/视频等）
+	if sel.target != "" {
+		if vals, ok := value.([]any); ok && len(vals) > 0 {
+			for _, v := range vals {
+				elem, err := buildStepElement(sel, v)
+				if err != nil {
+					return nil, err
+				}
+				arr = append(arr, elem)
+			}
+			return arr, nil
+		}
+	}
+	for _, e := range arr {
+		m, ok := e.(map[string]any)
+		if !ok {
+			continue
+		}
+		if matchFilters(m, sel) {
+			if err := writeStepValue(m, sel, value); err != nil {
+				return nil, err
+			}
+			return arr, nil
+		}
+	}
+	elem, err := buildStepElement(sel, value)
+	if err != nil {
+		return nil, err
+	}
+	return append(arr, elem), nil
+}
+
+// matchFilters 判断元素是否匹配选择器全部过滤条件；无过滤条件时命中第一个元素
+func matchFilters(m map[string]any, sel *selNode) bool {
+	if len(sel.filters) == 0 {
+		return true
+	}
+	for _, f := range sel.filters {
+		if gconv.String(m[f[0]]) != f[1] {
+			return false
+		}
+	}
+	return true
+}
+
+// writeStepValue 把值写入已定位元素：中间级递归 nested 下钻，叶子写 target 字段
+func writeStepValue(m map[string]any, sel *selNode, value any) error {
+	if len(sel.nested) > 0 {
+		return setBySteps(m, sel.nested, value)
+	}
+	if sel.target != "" {
+		return setLeafPath(m, sel.target, value)
+	}
+	return nil
+}
+
+// buildStepElement 按选择器构造新元素：{filterKey: filterVal, ...} + 叶子写 target / 中间级递归 nested
+func buildStepElement(sel *selNode, value any) (map[string]any, error) {
+	elem := make(map[string]any, len(sel.filters)+1)
+	for _, f := range sel.filters {
+		elem[f[0]] = f[1]
+	}
+	if err := writeStepValue(elem, sel, value); err != nil {
+		return nil, err
+	}
+	return elem, nil
+}
+
+// setLeafPath 在对象内按点号路径写入值（叶子用 setLeaf 语义）
+func setLeafPath(m map[string]any, path string, value any) error {
+	cur := m
+	segs := strings.Split(path, ".")
+	for i, k := range segs {
+		if i == len(segs)-1 {
+			setLeaf(cur, k, value)
+			return nil
+		}
+		next, err := ensureMap(cur, k)
+		if err != nil {
+			return err
+		}
+		cur = next
+	}
+	return nil
+}
+
+// setLeaf 叶子写入（业务字段语义：值前置追加而非覆盖）：
+// 字段不存在 → 直接写入；已有值 → 业务字段值前置到原有值前面（数组前插/字符串拼接），原值保留在后。
+func setLeaf(parent map[string]any, key string, value any) {
+	existing, ok := parent[key]
+	if !ok {
+		parent[key] = value
+		return
+	}
+	parent[key] = prependValue(value, existing)
+}
+
+// prependValue 把 value 前置到 existing 前（业务值在前，原值保留在后，不覆盖）：
+//   - existing 是数组 → value（或其元素）前插到数组头部
+//   - value 是数组（existing 为标量）→ value 各元素在前，existing 作为末位元素
+//   - 其余标量 → 字符串拼接，业务值在前
+func prependValue(value, existing any) any {
+	if arr, isArr := existing.([]any); isArr {
+		return prependToArray(arr, value)
+	}
+	if vals, isArr := value.([]any); isArr {
+		out := make([]any, 0, len(vals)+1)
+		out = append(out, vals...)
+		out = append(out, existing)
+		return out
+	}
+	return gconv.String(value) + gconv.String(existing)
+}
+
+// prependToArray 把 value（或其元素）插到数组头部，原元素依次后移
+func prependToArray(arr []any, value any) []any {
+	if vals, isArr := value.([]any); isArr {
+		out := make([]any, 0, len(vals)+len(arr))
+		out = append(out, vals...)
+		out = append(out, arr...)
+		return out
+	}
+	out := make([]any, 0, len(arr)+1)
+	out = append(out, value)
+	out = append(out, arr...)
+	return out
+}
+
+// ensureMap 确保键对应 map，不存在则新建；已存在但非对象 → 返回错误
+func ensureMap(parent map[string]any, key string) (map[string]any, error) {
+	if v, ok := parent[key]; ok {
+		if m, isMap := v.(map[string]any); isMap {
+			return m, nil
+		}
+		return nil, fmt.Errorf("字段 [%s] 不是对象，无法按路径写入（当前类型 %T）", key, v)
+	}
+	m := map[string]any{}
+	parent[key] = m
+	return m, nil
+}

service/utils/media_type.go

@@ -0,0 +1,30 @@
+package utils
+
+// DetectMediaType 按模型业务字段映射从请求体推导输入媒体类型（替代硬编码的 media.type 路径）。
+// 直接返回 shop-user-trade 计费词汇（audio/video，对齐 ChargeUsage.MediaType），
+// 无需二次转换（原 mgMediaTypeToShop 已删除）：
+//   - reference_audio 映射路径在请求体中有值 → "audio"
+//   - reference_video 映射路径在请求体中有值 → "video"
+//   - 否则 → ""（无媒体引用，shop 侧 pickModelPrice 查不到 mediaPrices 键 → 落默认价）
+//
+// 判定完全由模型配置（RequestBusinessFieldMapping，业务字段名见 ChatFieldsReq/VideoFields）驱动，
+// 无请求结构硬编码；映射路径值即 GetByPathAll 路径（如 input.media?type=audio&url=#）。
+// 媒体类型仅供 shop-user-trade 算费用量（见 service/pricing_client.go buildModelUsage）。
+func DetectMediaType(reqBizMapping map[string]string, reqParams map[string]any) string {
+	if hasMediaValue(reqBizMapping, reqParams, "reference_audio") {
+		return "audio"
+	}
+	if hasMediaValue(reqBizMapping, reqParams, "reference_video") {
+		return "video"
+	}
+	return ""
+}
+
+// hasMediaValue 业务字段映射路径在请求体中是否命中值
+func hasMediaValue(reqBizMapping map[string]string, reqParams map[string]any, bizField string) bool {
+	path := reqBizMapping[bizField]
+	if path == "" {
+		return false
+	}
+	return len(GetByPathAll(reqParams, path)) > 0
+}

service/utils/model_resolve.go

@@ -0,0 +1,150 @@
+package utils
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"model-gateway/dao"
+	"model-gateway/model/dto"
+	"model-gateway/model/entity"
+
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+// ====================== 模型引用解析层 ======================
+// 引用行(ref_system_model_id>0) 调用时实时取系统模型配置 + 本人 apiKey 合成可执行配置；
+// 系统模型调整零同步。Id 被覆盖为系统模型 id → 计价/并发键按系统模型走（会话/任务落库仍用引用行 id）。
+
+const apiKeyPlaceholder = "{apiKey}"
+
+// replacePlaceholder 递归替换 map/slice/string 中的占位符（泛化自 task_end 的 replaceTaskPlaceholder，非破坏式）。
+func replacePlaceholder(v any, from, to string) any {
+	switch val := v.(type) {
+	case string:
+		return strings.ReplaceAll(val, from, to)
+	case map[string]any:
+		m := make(map[string]any, len(val))
+		for k, x := range val {
+			m[k] = replacePlaceholder(x, from, to)
+		}
+		return m
+	case map[string]string:
+		m := make(map[string]string, len(val))
+		for k, x := range val {
+			m[k] = strings.ReplaceAll(x, from, to)
+		}
+		return m
+	case []any:
+		arr := make([]any, len(val))
+		for i, x := range val {
+			arr[i] = replacePlaceholder(x, from, to)
+		}
+		return arr
+	default:
+		return v
+	}
+}
+
+// copyAndReplaceStringMap 非破坏式替换 map[string]string 值（新建 map，不污染入参）
+func copyAndReplaceStringMap(src map[string]string, from, to string) map[string]string {
+	if src == nil {
+		return nil
+	}
+	m := make(map[string]string, len(src))
+	for k, v := range src {
+		m[k] = strings.ReplaceAll(v, from, to)
+	}
+	return m
+}
+
+// substituteAPIPlaceholder 把输入侧配置中的 {apiKey} 替换为生效 key（非破坏式：新建 map/struct，不污染入参）。
+// 覆盖 BaseURL / RequestHeadMapping / RequestBodyMapping / RequestBusinessFieldMapping /
+// AsyncTaskMapping(Url/RequestHeadMapping/RequestBodyMapping)。
+func substituteAPIPlaceholder(m *entity.ModelManage, key string) {
+	m.BaseURL = strings.ReplaceAll(m.BaseURL, apiKeyPlaceholder, key)
+	m.RequestHeadMapping = copyAndReplaceStringMap(m.RequestHeadMapping, apiKeyPlaceholder, key)
+	m.RequestBusinessFieldMapping = copyAndReplaceStringMap(m.RequestBusinessFieldMapping, apiKeyPlaceholder, key)
+	if v, ok := replacePlaceholder(m.RequestBodyMapping, apiKeyPlaceholder, key).(map[string]any); ok {
+		m.RequestBodyMapping = v
+	}
+	if a := m.AsyncTaskMapping; a != nil {
+		ac := *a
+		ac.Url = strings.ReplaceAll(a.Url, apiKeyPlaceholder, key)
+		ac.RequestHeadMapping = copyAndReplaceStringMap(a.RequestHeadMapping, apiKeyPlaceholder, key)
+		if v, ok := replacePlaceholder(a.RequestBodyMapping, apiKeyPlaceholder, key).(map[string]any); ok {
+			ac.RequestBodyMapping = v
+		}
+		m.AsyncTaskMapping = &ac
+	}
+}
+
+// mergeReferenceConfig 引用行 + 系统行 → 有效配置（纯函数，便于单测）。
+// 配置字段取系统行；个人字段(apiKey/enabled/chatModel)取引用行；enabled 取 AND（系统停用=引用失效）。
+func mergeReferenceConfig(stub, sys *entity.ModelManage) *entity.ModelManage {
+	out := *sys
+	out.RefSystemModelId = stub.RefSystemModelId // 保留引用标记，调用侧据此判断引用行门禁（Id 已覆盖为系统模型 id）
+	out.ApiKey = stub.ApiKey
+	if stub.Enabled != nil {
+		out.Enabled = stub.Enabled
+	}
+	if stub.ChatModel != nil {
+		out.ChatModel = stub.ChatModel
+	}
+	if sys.Enabled != nil && !*sys.Enabled {
+		out.Enabled = gconv.PtrBool(false)
+	}
+	return &out
+}
+
+// MergeReferenceConfigForQuery 管理端 Get 查询展示用：以引用行为基底，把系统行的配置列合入，
+// 保留引用行自身 Id/RefSystemModelId/SystemModel/Creator/时间戳与个人字段(apiKey/enabled/chatModel)。
+// 与 mergeReferenceConfig 的区别：不替换 {apiKey}（Get 非引用行也不替换，展示模板），
+// enabled 不做 AND（展示引用行个人开关，调用时才按系统行生效状态门禁）。
+func MergeReferenceConfigForQuery(stub, sys *entity.ModelManage) *entity.ModelManage {
+	out := *stub
+	out.BaseURL = sys.BaseURL
+	out.HttpMethod = sys.HttpMethod
+	out.ResponseType = sys.ResponseType
+	out.RequestHeadMapping = sys.RequestHeadMapping
+	out.RequestBodyMapping = sys.RequestBodyMapping
+	out.RequestBusinessFieldMapping = sys.RequestBusinessFieldMapping
+	out.ResponseMapping = sys.ResponseMapping
+	out.ResponseBodyMapping = sys.ResponseBodyMapping
+	out.ResponseBusinessFieldMapping = sys.ResponseBusinessFieldMapping
+	out.MaxConcurrency = sys.MaxConcurrency
+	out.TokenMapping = sys.TokenMapping
+	out.AsyncTaskMapping = sys.AsyncTaskMapping
+	out.TokenPredictPrice = sys.TokenPredictPrice
+	out.TokenPredictPriceUnit = sys.TokenPredictPriceUnit
+	out.MaxTokens = sys.MaxTokens
+	out.MinDuration = sys.MinDuration
+	out.MaxDuration = sys.MaxDuration
+	out.LastFrame = sys.LastFrame
+	out.ErrorMessageMapping = sys.ErrorMessageMapping
+	return &out
+}
+
+// ResolveModelConfig 把请求命中的模型行解析为可执行配置：
+// 引用行 → 系统行配置 + 引用行 apiKey（Id 覆盖为系统模型 id）；非引用行 → 原配置 + 自身 apiKey 替换占位。
+// 引用系统模型已删除 → 报错（调用方阻塞）。
+func ResolveModelConfig(ctx context.Context, m *entity.ModelManage) (*entity.ModelManage, error) {
+	if m == nil {
+		return nil, nil
+	}
+	if m.RefSystemModelId > 0 {
+		sys, err := dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{Id: m.RefSystemModelId})
+		if err != nil {
+			return nil, err
+		}
+		if sys == nil {
+			return nil, fmt.Errorf("引用的系统模型已删除")
+		}
+		out := mergeReferenceConfig(m, sys)
+		substituteAPIPlaceholder(out, out.ApiKey)
+		return out, nil
+	}
+	out := *m
+	substituteAPIPlaceholder(&out, out.ApiKey)
+	return &out, nil
+}

service/utils/pool.go

@@ -0,0 +1,58 @@
+package utils
+
+import (
+	"context"
+	"sync"
+	"sync/atomic"
+
+	"github.com/gogf/gf/v2/os/grpool"
+)
+
+var (
+	defaultPool atomic.Pointer[grpool.Pool]
+	once        sync.Once
+	wg          sync.WaitGroup
+)
+
+const DefaultWorkerNum = 100
+
+// Init 初始化全局协程池，首次调用生效，后续调用忽略。
+func Init(workerNum int) {
+	once.Do(func() {
+		if workerNum <= 0 {
+			workerNum = DefaultWorkerNum
+		}
+		defaultPool.Store(grpool.New(workerNum))
+	})
+}
+
+// Submit 提交异步任务，上下文透传至 grpool。
+// Submit 也可在 Init 前调用（自动 Init），但 Shutdown 后返回 ErrPoolClosed。
+func Submit(ctx context.Context, task func(ctx context.Context)) error {
+	p := defaultPool.Load()
+	if p == nil {
+		Init(DefaultWorkerNum)
+		p = defaultPool.Load()
+	}
+
+	wg.Add(1)
+	err := p.AddWithRecover(ctx, func(ctx context.Context) {
+		defer wg.Done()
+		task(ctx)
+	}, nil)
+	if err != nil {
+		wg.Done()
+		return err
+	}
+	return nil
+}
+
+// Shutdown 优雅关闭：停止新任务，等待全部已完成/排队任务完成。
+func Shutdown() {
+	p := defaultPool.Swap(nil)
+	if p == nil {
+		return
+	}
+	wg.Wait()
+	p.Close()
+}

service/utils/template_merge.go

@@ -0,0 +1,260 @@
+package utils
+
+import "github.com/gogf/gf/v2/util/gconv"
+
+// MergeNode 按模板节点的元数据把用户数据节点合并为模板可解析的节点：
+//   - 用户节点无 type 键 → 套上模板类型包装（标量取用户值，空则回落模板 value/defaultValue）
+//   - 用户节点有 type 键 → 递归合并子节点（子节点缺 type 时同样补模板类型，避免模板包装泄漏）
+//   - 对象：模板 attrs 缺省字段补进用户容器（保留模板默认值）
+//   - 数组：用户 enumValues/attrs 元素逐个与模板元素原型合并（模板未提供的槽位克隆原型）
+func MergeNode(user, tmpl any) any {
+	tmplMap, ok := tmpl.(map[string]any)
+	if !ok {
+		return user
+	}
+	fieldType, _ := tmplMap["type"].(string)
+
+	// 用户节点已带 type：按用户类型递归合并子节点；标量叶子已完备，直接返回
+	if userMap, ok := user.(map[string]any); ok {
+		if ut, isTpl := userMap["type"]; isTpl {
+			switch gconv.String(ut) {
+			case TypeObject:
+				return mergeObject(userMap, tmplMap)
+			case TypeArray:
+				return mergeArray(userMap, tmplMap)
+			default:
+				return userMap
+			}
+		}
+	}
+
+	switch fieldType {
+	case TypeString, TypeBool, TypeNumber, TypeNumberInt, TypeNumberFloat:
+		return wrapScalarNode(tmplMap, user)
+	case TypeObject:
+		return mergeObject(user, tmplMap)
+	case TypeArray:
+		return mergeArray(user, tmplMap)
+	default:
+		return user
+	}
+}
+
+// wrapScalarNode 把用户标量值包装为 {type, value} 节点；用户值为空时回落模板 value/defaultValue
+func wrapScalarNode(tmplMap map[string]any, user any) map[string]any {
+	val := user
+	if m, ok := user.(map[string]any); ok {
+		if v, has := m["value"]; has {
+			val = v
+		} else {
+			val = nil
+		}
+	}
+	node := map[string]any{"type": normalizeScalarType(gconv.String(tmplMap["type"]))}
+	if hasUsableValue(val) {
+		node["value"] = val
+	} else if v, has := tmplMap["value"]; has {
+		node["value"] = v
+	} else if d, has := tmplMap["defaultValue"]; has {
+		node["value"] = d
+	}
+	return node
+}
+
+// normalizeScalarType integer/float 统一为 number（resolveField 只分发 string/boolean/number）
+func normalizeScalarType(t string) string {
+	if t == TypeNumberInt || t == TypeNumberFloat {
+		return TypeNumber
+	}
+	return t
+}
+
+// mergeObject 合并对象节点：模板 attrs 缺省字段补进用户容器（模板默认值保留）
+func mergeObject(user any, tmplMap map[string]any) map[string]any {
+	tmplAttrs, _ := tmplMap["attrs"].(map[string]any)
+	node := map[string]any{"type": TypeObject}
+	if tmplAttrs == nil {
+		node["attrs"] = user
+		return node
+	}
+	var container map[string]any
+	switch u := user.(type) {
+	case map[string]any:
+		if a, has := u["attrs"]; has {
+			if am, ok := a.(map[string]any); ok {
+				container = copyMap(am)
+			} else {
+				container = map[string]any{}
+			}
+		} else if v, has := u["value"]; has {
+			if vm, ok := v.(map[string]any); ok {
+				container = copyMap(vm)
+			} else {
+				container = map[string]any{}
+			}
+		} else {
+			container = copyMap(u)
+		}
+	default:
+		container = map[string]any{}
+	}
+	for k, subTmpl := range tmplAttrs {
+		if _, has := container[k]; !has {
+			container[k] = DeepCopyNode(subTmpl)
+		} else {
+			container[k] = MergeNode(container[k], subTmpl)
+		}
+	}
+	node["attrs"] = container
+	return node
+}
+
+// mergeArray 合并数组节点：用户 enumValues/attrs 元素逐个与模板元素原型合并。
+// 用户未填的槽位（sjson null 填充）用模板对应槽位原型补位，保留下标不塌缩。
+func mergeArray(user any, tmplMap map[string]any) map[string]any {
+	proto := arrayElementTemplate(tmplMap)
+	tmplSlots, _ := tmplMap["enumValues"].([]any) // 模板各槽位原型，按下标一一对应
+	node := map[string]any{"type": TypeArray}
+	slotProto := func(i int) any {
+		if i < len(tmplSlots) {
+			return tmplSlots[i]
+		}
+		return proto
+	}
+	switch u := user.(type) {
+	case map[string]any:
+		if evs, ok := u["enumValues"].([]any); ok {
+			out := make([]any, len(evs))
+			for i, ev := range evs {
+				if ev == nil {
+					// 保留 null 槽位以维持下标；解析阶段 resolveArray 会丢弃空元素，
+					// 避免用模板原型填充 null 而物化出"幽灵"元素（用户未填的数组槽位）
+					out[i] = nil
+					continue
+				}
+				out[i] = mergeArrayElement(ev, slotProto(i))
+			}
+			node["enumValues"] = out
+		} else if a, ok := u["attrs"].([]any); ok {
+			out := make([]any, len(a))
+			for i, item := range a {
+				out[i] = mergeArrayElement(item, proto)
+			}
+			node["attrs"] = out
+		} else if v, has := u["value"]; has {
+			node["value"] = v
+		} else if len(u) > 0 {
+			node["attrs"] = []any{mergeArrayElement(u, proto)}
+		}
+	case []any:
+		if isSchemaElementList(u) {
+			out := make([]any, len(u))
+			for i, item := range u {
+				out[i] = mergeArrayElement(item, proto)
+			}
+			node["attrs"] = out
+		} else {
+			node["value"] = u
+		}
+	default:
+		node["value"] = user
+	}
+	return node
+}
+
+// mergeArrayElement 合并单个数组元素：元素缺 type 时按槽位原型补默认字段。
+// slotProto 可为 nil（模板未提供槽位原型时）→ 原样返回用户元素
+func mergeArrayElement(ev any, slotProto any) any {
+	if ev == nil {
+		return ev
+	}
+	proto, _ := slotProto.(map[string]any)
+	if proto == nil {
+		return ev
+	}
+	evMap, ok := ev.(map[string]any)
+	if !ok {
+		return ev
+	}
+	if _, isTpl := evMap["type"]; isTpl {
+		return ev
+	}
+	// 元素原型自身是包装节点（嵌套对象/数组）→ 按模板类型合并
+	if t, has := proto["type"].(string); has && t != "" {
+		return MergeNode(ev, proto)
+	}
+	protoAttrs, _ := proto["attrs"].(map[string]any)
+	var container map[string]any
+	if a, has := evMap["attrs"]; has {
+		if am, ok := a.(map[string]any); ok {
+			container = copyMap(am)
+		} else {
+			container = map[string]any{}
+		}
+	} else {
+		container = copyMap(evMap)
+	}
+	for k, subTmpl := range protoAttrs {
+		if _, has := container[k]; !has {
+			container[k] = DeepCopyNode(subTmpl)
+		} else {
+			container[k] = MergeNode(container[k], subTmpl)
+		}
+	}
+	return map[string]any{"attrs": container}
+}
+
+// isSchemaElementList 判断 []any 是否为 schema 元素列表（元素均为 map，且带 type 或 attrs 包装）
+func isSchemaElementList(list []any) bool {
+	if len(list) == 0 {
+		return false
+	}
+	for _, item := range list {
+		m, ok := item.(map[string]any)
+		if !ok {
+			return false
+		}
+		if _, hasType := m["type"]; !hasType {
+			if _, hasAttrs := m["attrs"]; !hasAttrs {
+				return false
+			}
+		}
+	}
+	return true
+}
+
+// DeepCopyNode 深拷贝任意嵌套节点（模板节点整体拷贝进 src 时使用，避免共享引用被后续回填修改）
+func DeepCopyNode(v any) any {
+	switch val := v.(type) {
+	case map[string]any:
+		out := make(map[string]any, len(val))
+		for k, sub := range val {
+			out[k] = DeepCopyNode(sub)
+		}
+		return out
+	case []any:
+		out := make([]any, len(val))
+		for i, sub := range val {
+			out[i] = DeepCopyNode(sub)
+		}
+		return out
+	default:
+		return v
+	}
+}
+
+// hasUsableValue 值是否可写入节点 value；0/false 视为有效值
+func hasUsableValue(v any) bool {
+	switch val := v.(type) {
+	case nil:
+		return false
+	case string:
+		return val != ""
+	case []any:
+		return len(val) > 0
+	case map[string]any:
+		return len(val) > 0
+	default:
+		return true
+	}
+}

service/utils/template_parse.go

@@ -0,0 +1,605 @@
+package utils
+
+import (
+	"regexp"
+	"strings"
+
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	"github.com/gogf/gf/v2/util/gconv"
+)
+
+var (
+	// 匹配 [数字]
+	regNumIndex = regexp.MustCompile(`\[\d+\]`)
+	// 匹配 .attrs
+	regAttrs = regexp.MustCompile(`\.attrs`)
+)
+
+// NormalizeFieldPath 归一化字段路径到统一语法（[*] 数组段）：
+//   - 移除模板残留 .attrs
+//   - [数字] 下标 → [*]（choices[0] → choices[*]）
+//   - 兼容 gjson 风格 .# / .数字 下标 → [*]（choices.#、choices.0 → choices[*]）
+//
+// 统一语法见 business_fields.go 的 SetByPath / GetByPath：
+//
+//	a.b.c                    普通点号路径
+//	a[*].b                   [*] 表示数组段
+//	a[*].b[*]?k=v&t=#        选择器：数组元素按 k==v 定位，值/读取目标为 t
+//	a[*]?k=v&b[*]?k2=v2&t=#  多级选择器：级数不限，中间级定位容器元素，叶子写值
+//
+// 正则归一（.attrs / [数字] / .#）作用于整个路径（含多级选择器中的数组段）；
+// 纯数字段（gjson 下标）归一只作用于首个 ? 之前的 base 路径。
+//
+// 示例：
+//
+//	usage.attrs.total_tokens → usage.total_tokens
+//	choices.attrs[0].attrs.message.attrs.content → choices[*].message.content
+//	choices.#.message.content → choices[*].message.content
+//	choices.0.message.content → choices[*].message.content
+func NormalizeFieldPath(path string) string {
+	s := regAttrs.ReplaceAllString(path, "")
+	s = regNumIndex.ReplaceAllString(s, "[*]")
+	s = strings.ReplaceAll(s, ".#", "[*]")
+	base, suffix := s, ""
+	if i := strings.Index(s, "?"); i >= 0 {
+		base, suffix = s[:i], s[i:]
+	}
+	// 逐段把纯数字段（gjson 下标）归一为 [*]：附着到前一段字段（choices.0 → choices[*]），
+	// 避免误伤数字开头的字段名；选择器体用 # 作目标、不用数字段下标，故只归一 base
+	segs := strings.Split(base, ".")
+	var out []string
+	for _, seg := range segs {
+		if seg == "" {
+			continue
+		}
+		if isAllDigits(seg) {
+			if len(out) > 0 {
+				out[len(out)-1] += "[*]"
+			} else {
+				out = append(out, "[*]")
+			}
+			continue
+		}
+		out = append(out, seg)
+	}
+	return strings.Join(out, ".") + suffix
+}
+
+// isAllDigits 判断字符串是否全部为数字字符
+func isAllDigits(s string) bool {
+	if s == "" {
+		return false
+	}
+	for _, r := range s {
+		if r < '0' || r > '9' {
+			return false
+		}
+	}
+	return true
+}
+
+// CleanFieldPath 清理字段路径（等价于 NormalizeFieldPath，保留旧名兼容）
+func CleanFieldPath(path string) string {
+	return NormalizeFieldPath(path)
+}
+
+// CleanMapFieldPath 清理字段路径（Map）
+func CleanMapFieldPath(m map[string]string) map[string]string {
+	if m == nil {
+		return nil
+	}
+	newMap := make(map[string]string, len(m))
+	for k, _ := range m {
+		newMap[k] = CleanFieldPath(k)
+	}
+	return newMap
+}
+
+// ParseConfigTemplate 解析配置模板生成简化请求结构
+//
+// 输入: config 模板（含 type/value/defaultValue/attrs/enumValues 等元数据字段）
+// 输出: 简化后的请求结构体
+//
+// 规则:
+//   - 标量字段（string/number/boolean）: value 非零则用 value，为空则跳过（不再取 defaultValue）
+//   - 对象字段（object）: 递归处理 attrs
+//   - 数组字段（array）: 遍历 enumValues，每个 enumValue 独立判断是否产出元素
+//   - 数组展开: enumValue 内某叶子字段 value 为数组时，按数组元素展开为多个项
+func ParseConfigTemplate(cfg map[string]interface{}) map[string]interface{} {
+	var flattenJSON map[string]interface{}
+	flatMap := utils.IsFlatMap(cfg)
+	if flatMap {
+		var err error
+		flattenJSON, err = utils.UnFlatBySjson(cfg)
+		if err != nil {
+			return nil
+		}
+	} else {
+		flattenJSON = cfg
+	}
+	result := make(map[string]interface{})
+	for key, val := range flattenJSON {
+		field, ok := val.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		if v := resolveField(field); v != nil {
+			result[key] = v
+		}
+	}
+	return result
+}
+
+// resolveField 按 type 分发解析
+func resolveField(field map[string]interface{}) interface{} {
+	fieldType, _ := field["type"].(string)
+	switch fieldType {
+	case TypeString, TypeBool, TypeNumber:
+		return resolveScalar(field)
+	case TypeObject:
+		return resolveObject(field)
+	case TypeArray:
+		return resolveArray(field)
+	}
+	return nil
+}
+
+// resolveScalar 解析标量字段: value 非空则用 value，否则回落 defaultValue
+// （模板只声明结构、值由业务字段给出时，defaultValue 生效）
+func resolveScalar(field map[string]interface{}) interface{} {
+	if v, has := field["value"]; has && v != nil {
+		switch vv := v.(type) {
+		case string:
+			if vv != "" {
+				return vv
+			}
+		case bool:
+			return vv
+		default:
+			// 数值零值（int/float 各类型）视为未提供，跳过；bool/string 已在上方处理
+			if isNumericZero(v) {
+				return nil
+			}
+			return vv
+		}
+	}
+	if d, has := field["defaultValue"]; has && d != nil {
+		switch dv := d.(type) {
+		case string:
+			if dv != "" {
+				return dv
+			}
+		case bool:
+			return dv
+		default:
+			if isNumericZero(d) {
+				return nil
+			}
+			return dv
+		}
+	}
+	return nil
+}
+
+// isNumericZero 判断是否为数值零值（模板 value 常为 int 字面量，经 gconv 可能为 float64）
+func isNumericZero(v interface{}) bool {
+	switch vv := v.(type) {
+	case int:
+		return vv == 0
+	case int8:
+		return vv == 0
+	case int16:
+		return vv == 0
+	case int32:
+		return vv == 0
+	case int64:
+		return vv == 0
+	case uint:
+		return vv == 0
+	case uint8:
+		return vv == 0
+	case uint16:
+		return vv == 0
+	case uint32:
+		return vv == 0
+	case uint64:
+		return vv == 0
+	case float32:
+		return vv == 0
+	case float64:
+		return vv == 0
+	default:
+		return false
+	}
+}
+
+// resolveObject 解析对象字段，递归处理 attrs
+//
+// 特殊处理「参数定义」结构：当 attrs 含 default 字段时，说明该对象是一个
+// 参数定义（含 default/description/min/max/type/enum/required 等元数据），
+// 此时只提取 default 的值作为该参数的值，其余元数据字段忽略。
+func resolveObject(field map[string]interface{}) interface{} {
+	attrs, ok := field["attrs"].(map[string]interface{})
+	if !ok {
+		return nil
+	}
+
+	// 参数定义：只提取 default 值，跳过元数据
+	if defaultField, hasDefault := attrs["default"]; hasDefault {
+		if df, ok := defaultField.(map[string]interface{}); ok {
+			return extractRawValueKeepZero(df)
+		}
+		return nil
+	}
+
+	// 普通对象：递归处理所有 attrs
+	result := make(map[string]interface{})
+	for key, val := range attrs {
+		subField, ok := val.(map[string]interface{})
+		if !ok {
+			result[key] = val // 纯值字段原样保留
+			continue
+		}
+		if subType, _ := subField["type"].(string); subType == "" {
+			result[key] = val // 无 type 键的纯对象原样保留
+			continue
+		}
+		if v := resolveField(subField); v != nil {
+			result[key] = v
+		}
+	}
+	if len(result) == 0 {
+		return nil
+	}
+	return result
+}
+
+// resolveArray 解析数组字段，遍历 enumValues 或 attrs 生成元素列表
+func resolveArray(field map[string]interface{}) []interface{} {
+	// 实际数据在 value（schema-editor 数据存放处），直接返回
+	if v, has := field["value"]; has {
+		if arr, ok := v.([]interface{}); ok && len(arr) > 0 {
+			return arr
+		}
+	}
+
+	enumValues, ok := field["enumValues"].([]interface{})
+	if ok {
+		var result []interface{}
+		for _, ev := range enumValues {
+			evMap, ok := ev.(map[string]interface{})
+			if !ok {
+				continue
+			}
+			items := resolveEnumObject(evMap)
+			result = append(result, items...)
+		}
+		if len(result) > 0 {
+			return result
+		}
+	}
+
+	// enumValues 取不到或为空时，尝试从 attrs（数组）中取
+	attrs, ok := field["attrs"].([]interface{})
+	if !ok {
+		return nil
+	}
+	var result []interface{}
+	for _, item := range attrs {
+		itemMap, ok := item.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		if v := resolveField(itemMap); v != nil {
+			result = append(result, v)
+		}
+	}
+	return result
+}
+
+// resolveEnumObject 解析 enumValue 对象，支持数组展开
+func resolveEnumObject(ev map[string]interface{}) []interface{} {
+	attrs, ok := ev["attrs"].(map[string]interface{})
+	if !ok {
+		return nil
+	}
+	// 将 enumValue 级别的 value 注入 attrs.type.value（如果 type.value 为空）
+	if evVal, has := ev["value"]; has && evVal != nil {
+		if s, ok := evVal.(string); ok && s != "" {
+			if typeField, has := attrs["type"]; has {
+				if typeMap, ok := typeField.(map[string]interface{}); ok {
+					if existing, has := typeMap["value"]; !has || existing == nil || existing == "" {
+						typeMap["value"] = s
+					}
+				}
+			}
+		}
+	}
+	return resolveAttrs(attrs)
+}
+
+// resolveAttrs 递归解析 attrs map，支持字段级数组展开
+func resolveAttrs(attrs map[string]interface{}) []interface{} {
+	currentItems := []map[string]interface{}{{}}
+	hasValue := false
+
+	for key, val := range attrs {
+		subField, isMap := val.(map[string]interface{})
+		var subType string
+		if isMap {
+			subType, _ = subField["type"].(string)
+		}
+
+		var nextItems []map[string]interface{}
+
+		// 非包裹字段（纯值/纯对象，无 type 键）：原样保留，数组值仍参与展开
+		if !isMap || subType == "" {
+			raw := val
+			if raw == nil {
+				nextItems = currentItems
+				currentItems = nextItems
+				continue
+			}
+			hasValue = true
+			if arr, ok := raw.([]interface{}); ok && len(arr) > 0 {
+				for _, item := range currentItems {
+					for _, elem := range arr {
+						cp := copyMap(item)
+						cp[key] = elem
+						nextItems = append(nextItems, cp)
+					}
+				}
+			} else {
+				for _, item := range currentItems {
+					cp := copyMap(item)
+					cp[key] = raw
+					nextItems = append(nextItems, cp)
+				}
+			}
+			currentItems = nextItems
+			continue
+		}
+
+		switch subType {
+		case TypeString, TypeBool, TypeNumber:
+			raw := extractRawValue(subField)
+			if raw == nil {
+				nextItems = currentItems
+				continue
+			}
+			hasValue = true
+			if arr, ok := raw.([]interface{}); ok && len(arr) > 0 {
+				for _, item := range currentItems {
+					for _, elem := range arr {
+						cp := copyMap(item)
+						cp[key] = elem
+						nextItems = append(nextItems, cp)
+					}
+				}
+			} else {
+				for _, item := range currentItems {
+					cp := copyMap(item)
+					cp[key] = raw
+					nextItems = append(nextItems, cp)
+				}
+			}
+
+		case TypeObject:
+			subAttrs, ok := subField["attrs"].(map[string]interface{})
+			if !ok {
+				nextItems = currentItems
+				continue
+			}
+			subItems := resolveAttrs(subAttrs)
+			if len(subItems) == 0 {
+				nextItems = currentItems
+				continue
+			}
+			hasValue = true
+			for _, item := range currentItems {
+				for _, subI := range subItems {
+					cp := copyMap(item)
+					cp[key] = subI
+					nextItems = append(nextItems, cp)
+				}
+			}
+
+		case TypeArray:
+			items := resolveArray(subField)
+			if len(items) == 0 {
+				nextItems = currentItems
+				continue
+			}
+			hasValue = true
+			for _, item := range currentItems {
+				cp := copyMap(item)
+				cp[key] = items
+				nextItems = append(nextItems, cp)
+			}
+
+		default:
+			nextItems = currentItems
+		}
+
+		currentItems = nextItems
+	}
+
+	if !hasValue {
+		return nil
+	}
+	result := make([]interface{}, len(currentItems))
+	for i, item := range currentItems {
+		result[i] = item
+	}
+	return result
+}
+
+// extractRawValue 提取原始值（保留数组值供上层展开）
+func extractRawValue(field map[string]interface{}) interface{} {
+	if v, has := field["value"]; has && v != nil {
+		switch vv := v.(type) {
+		case string:
+			if vv != "" {
+				return vv
+			}
+		case float64:
+			if vv != 0 {
+				return vv
+			}
+		case bool:
+			return vv
+		case []interface{}:
+			if len(vv) > 0 {
+				return vv
+			}
+		default:
+			return vv
+		}
+	}
+	return nil
+}
+
+// extractRawValueKeepZero 同 extractRawValue，但不过滤零值
+// 在参数定义场景下，default 可能是 false/0/""，需要保留
+func extractRawValueKeepZero(field map[string]interface{}) interface{} {
+	if v, has := field["value"]; has && v != nil {
+		switch vv := v.(type) {
+		case string:
+			return vv
+		case float64:
+			return vv
+		case bool:
+			return vv
+		case []interface{}:
+			if len(vv) > 0 {
+				return vv
+			}
+			return vv
+		default:
+			return vv
+		}
+	}
+	if dv, has := field["defaultValue"]; has && dv != nil {
+		switch dvv := dv.(type) {
+		case string:
+			if field["type"] == TypeBool {
+				if dvv == "true" {
+					return true
+				}
+				if dvv == "false" {
+					return false
+				}
+			}
+			return dvv
+		case float64:
+			return dvv
+		case bool:
+			return dvv
+		case []interface{}:
+			if len(dvv) > 0 {
+				return dvv
+			}
+			return dvv
+		default:
+			return dvv
+		}
+	}
+	return nil
+}
+
+// copyMap 浅拷贝 map
+func copyMap(src map[string]interface{}) map[string]interface{} {
+	dst := make(map[string]interface{}, len(src))
+	for k, v := range src {
+		dst[k] = v
+	}
+	return dst
+}
+
+// CoerceBodyTypes 按模板声明的 type 递归归一请求体字段值类型：
+//   - string → gconv.String；number → gconv.Float64；boolean → gconv.Bool
+//   - object → 按模板 attrs 递归子字段；array → 按元素模板逐个递归
+//   - 模板未声明的字段（业务字段写入且超出模板的部分）保持原样
+//
+// 用于构建请求体后统一修正：模板字段 value 与业务字段写入的值都可能携带与声明
+// 类型不一致的 Go 类型（如 number 字段 value 为字符串 "0.7"），在此统一转成模型
+// API 期望的 JSON 类型。仅做类型归一，不增删字段。
+func CoerceBodyTypes(out map[string]interface{}, templateParams map[string]interface{}) map[string]interface{} {
+	if len(templateParams) == 0 {
+		return out
+	}
+	for key, raw := range out {
+		if tmplNode, has := templateParams[key]; has {
+			out[key] = coerceNode(raw, tmplNode)
+		}
+	}
+	return out
+}
+
+// coerceNode 按单个模板节点归一值类型
+func coerceNode(value interface{}, tmplNode interface{}) interface{} {
+	tmplMap, ok := tmplNode.(map[string]interface{})
+	if !ok {
+		return value
+	}
+	fieldType, _ := tmplMap["type"].(string)
+	switch fieldType {
+	case TypeString:
+		return gconv.String(value)
+	case TypeNumber, TypeNumberInt, TypeNumberFloat:
+		return gconv.Float64(value)
+	case TypeBool:
+		return gconv.Bool(value)
+	case TypeObject:
+		sub, ok := value.(map[string]interface{})
+		if !ok {
+			return value
+		}
+		if attrs, ok := tmplMap["attrs"].(map[string]interface{}); ok {
+			return coerceObject(sub, attrs)
+		}
+		return value
+	case TypeArray:
+		arr, ok := value.([]interface{})
+		if !ok {
+			return value
+		}
+		proto := arrayElementTemplate(tmplMap)
+		if proto == nil {
+			return value
+		}
+		out := make([]interface{}, len(arr))
+		for i, elem := range arr {
+			out[i] = coerceNode(elem, proto)
+		}
+		return out
+	default:
+		return value
+	}
+}
+
+// coerceObject 按对象模板 attrs 归一对象子字段类型
+func coerceObject(sub, attrs map[string]interface{}) map[string]interface{} {
+	for key, raw := range sub {
+		if tmplNode, has := attrs[key]; has {
+			sub[key] = coerceNode(raw, tmplNode)
+		}
+	}
+	return sub
+}
+
+// arrayElementTemplate 从数组模板节点提取元素模板（attrs 优先，其次 enumValues）。
+// 与 arrayElementPrototype 语义一致，但直接工作在原始模板 map 上，供类型归一使用。
+func arrayElementTemplate(field map[string]interface{}) map[string]interface{} {
+	if attrs, ok := field["attrs"].([]interface{}); ok && len(attrs) > 0 {
+		if m, ok := attrs[0].(map[string]interface{}); ok {
+			return m
+		}
+	}
+	if evs, ok := field["enumValues"].([]interface{}); ok && len(evs) > 0 {
+		if m, ok := evs[0].(map[string]interface{}); ok {
+			return m
+		}
+	}
+	return nil
+}

service/utils/template_validate.go

@@ -0,0 +1,503 @@
+package utils
+
+import (
+	"fmt"
+	"model-gateway/model/dto"
+
+	"gitea.redpowerfuture.com/red-future/common/utils"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/util/gconv"
+	"github.com/gogf/gf/v2/util/gutil"
+)
+
+// 数据类型常量
+const (
+	TypeString      = "string"
+	TypeBool        = "boolean"
+	TypeNumber      = "number"
+	TypeNumberInt   = "integer"
+	TypeNumberFloat = "float"
+	TypeNull        = "null"
+	TypeObject      = "object"
+	TypeArray       = "array"
+)
+
+// CheckParams 校验用户入参并回填默认值：
+// 用户只传 key/value，约束参数（type/required/constraint）全部取模板定义。
+// 模板定义必填的字段，用户未传或传空值都报错；用户值为空时用模板 defaultValue 回填。
+// 严格模式：未知字段报错。
+func CheckParams(userParams map[string]interface{}, templateParams map[string]interface{}) error {
+	return checkParams(userParams, templateParams, true, true)
+}
+
+// CheckBody 校验构建完成的请求体（ParseConfigTemplate + WriteBusinessFields 之后）：
+// 业务字段按映射写入的路径可能超出模板声明，未知字段不报错；默认值已在构建期处理，不做回填。
+// 仍按模板约束校验必填/长度/范围。
+func CheckBody(body map[string]interface{}, templateParams map[string]interface{}) error {
+	return checkParams(body, templateParams, false, true)
+}
+
+// checkParams 按模板校验请求结构。strictUnknown：未知字段是否报错；backfill：空值是否回填 defaultValue。
+func checkParams(userParams map[string]interface{}, templateParams map[string]interface{}, strictUnknown, backfill bool) error {
+	// 兼容扁平路径入参：还原为嵌套结构
+	orig := userParams
+	if utils.IsFlatMap(userParams) {
+		nested, err := utils.UnFlatBySjson(userParams)
+		if err != nil {
+			return fmt.Errorf("无法解析用户参数: %w", err)
+		}
+		orig = nested
+	}
+	// 顶层未知字段检查
+	if strictUnknown {
+		for key := range orig {
+			if _, has := templateParams[key]; !has {
+				return fmt.Errorf("非法字段: %s 模板中不存在该字段", key)
+			}
+		}
+	}
+	for key, tmplNode := range templateParams {
+		if err := validateNode(orig, key, tmplNode, key, strictUnknown, backfill); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// validateNode 按模板节点校验用户值，空值回填 defaultValue。
+// parent 为用户原始结构（模板格式 {type,value/attrs} 或纯值），key 为字段名；回填写回 parent[key]。
+func validateNode(parent map[string]interface{}, key string, tmplNode interface{}, path string, strictUnknown, backfill bool) error {
+	raw, hasRaw := parent[key]
+
+	tmplMap, ok := tmplNode.(map[string]interface{})
+	if !ok {
+		return nil // 模板节点不是对象，无约束可校验
+	}
+	var tmpl dto.Template
+	if err := gconv.Struct(tmplMap, &tmpl); err != nil {
+		return fmt.Errorf("字段 [%s] 模板解析错误: %w", path, err)
+	}
+
+	label := tmpl.Label
+	if label == "" {
+		label = path
+	}
+
+	switch tmpl.Type {
+	case TypeObject:
+		userMap, hasUser := userObjectValue(raw, hasRaw)
+		if !hasUser {
+			if tmpl.Required {
+				return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", label)
+			}
+			return nil // 未传对象且非必填：跳过
+		}
+		attrs, ok := tmpl.Attrs.(map[string]interface{})
+		if !ok {
+			return nil
+		}
+		// 未知子字段检查
+		if strictUnknown {
+			for k := range userMap {
+				if _, has := attrs[k]; !has {
+					return fmt.Errorf("非法字段: %s 模板中不存在该字段", path+"."+k)
+				}
+			}
+		}
+		// 递归子字段（即使对象未传，子字段必填校验仍生效）
+		for subKey, subTmpl := range attrs {
+			if err := validateNode(userMap, subKey, subTmpl, path+"."+subKey, strictUnknown, backfill); err != nil {
+				return err
+			}
+		}
+		return nil
+
+	case TypeArray:
+		// 枚举项：逐项校验请求 enumValue.attrs 子字段并回填默认值
+		if err := validateEnumValues(raw, hasRaw, path, strictUnknown, backfill); err != nil {
+			return err
+		}
+		userArr, hasUser := userArrayValue(raw, hasRaw)
+		if !hasUser || len(userArr) == 0 {
+			// 先回填 defaultValue（必填字段也可由默认值兜底），回填后重新判空
+			if backfill {
+				backfillDefault(parent, key, raw, hasRaw, &tmpl)
+			}
+			if tmpl.Required && isValueEmptyByType(&tmpl) {
+				return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", label)
+			}
+			return nil
+		}
+		// 数组数量约束：上限取 Constraint.UploadTotalMaxCount 或各 UploadRule.MaxCount 之和（schema 构建时配置）
+		if limit := maxArrayCount(&tmpl); limit > 0 && len(userArr) > limit {
+			return fmt.Errorf("字段 [%s] 数量 %d 超过限制 %d", label, len(userArr), limit)
+		}
+		proto := arrayElementPrototype(&tmpl)
+		if proto == nil {
+			return nil
+		}
+		var protoTmpl dto.Template
+		if err := gconv.Struct(proto, &protoTmpl); err != nil {
+			return nil
+		}
+		switch protoTmpl.Type {
+		case TypeObject:
+			// 对象元素：以元素 attrs 为容器递归校验子字段（模板对象节点 {type:object,attrs:{...}} 的
+			// 子字段藏在 attrs 下；纯对象 map 直接以自身为容器）
+			for i, elem := range userArr {
+				elemMap, isMap := elem.(map[string]interface{})
+				if !isMap {
+					continue
+				}
+				container, has := userObjectValue(elemMap, true)
+				if !has {
+					continue
+				}
+				// 元素带 type 键（schema 包裹）：已由 validateEnumValues 按其自身 attrs 校验，跳过，
+				// 避免用枚举首原型（可能是必填字段模板）误报其他槽位元素缺失
+				if _, isWrapped := elemMap["type"]; isWrapped {
+					continue
+				}
+				// 纯对象元素（解析后数组下标可能塌缩）：按 attrs 键集结构匹配槽位原型，避免恒用首原型误报必填
+				subAttrs := matchArraySlotProto(&tmpl, container)
+				if subAttrs == nil {
+					continue
+				}
+				for subKey, subTmpl := range subAttrs {
+					subPath := fmt.Sprintf("%s[%d].%s", path, i, subKey)
+					if err := validateNode(container, subKey, subTmpl, subPath, strictUnknown, backfill); err != nil {
+						return err
+					}
+				}
+			}
+		default:
+			// 标量元素：逐元素校验（数组内元素不参与整体必填）
+			for i, elem := range userArr {
+				pt := protoTmpl
+				pt.Value = elem
+				pt.Required = false
+				if err := checkScalar(&pt, fmt.Sprintf("%s[%d]", path, i)); err != nil {
+					return err
+				}
+			}
+		}
+		return nil
+
+	default:
+		// 标量类型：先回填 defaultValue（必填字段也可由默认值兜底），回填后重新判空，再校验必填/约束
+		tmpl.Value = userScalarValue(raw, hasRaw)
+		if isValueEmptyByType(&tmpl) {
+			if backfill {
+				backfillDefault(parent, key, raw, hasRaw, &tmpl)
+				tmpl.Value = tmpl.DefaultValue
+			}
+			if isValueEmptyByType(&tmpl) {
+				if tmpl.Required {
+					return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", label)
+				}
+				return nil
+			}
+		}
+		return checkScalar(&tmpl, label)
+	}
+}
+
+// checkScalar 校验标量值：必填 + 约束
+func checkScalar(tmpl *dto.Template, label string) error {
+	if isValueEmptyByType(tmpl) {
+		if tmpl.Required {
+			return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", label)
+		}
+		return nil
+	}
+	switch tmpl.Type {
+	case TypeString:
+		return checkStringTmpl(tmpl)
+	case TypeNumber:
+		return checkNumberTmpl(tmpl)
+	case TypeBool:
+		return checkBoolTmpl(tmpl)
+	case TypeNull:
+		return nil
+	default:
+		return fmt.Errorf("字段 [%s] 不支持的模板类型: %s", label, tmpl.Type)
+	}
+}
+
+// userObjectValue 从原始请求节点提取对象值（模板格式取 attrs/value，纯值直接返回 map）
+func userObjectValue(raw interface{}, hasRaw bool) (map[string]interface{}, bool) {
+	if !hasRaw || raw == nil {
+		return nil, false
+	}
+	if m, ok := raw.(map[string]interface{}); ok {
+		if _, isTpl := m["type"]; isTpl {
+			if v, has := m["attrs"]; has {
+				if sub, ok := v.(map[string]interface{}); ok {
+					return sub, true
+				}
+			}
+			if v, has := m["value"]; has {
+				if sub, ok := v.(map[string]interface{}); ok {
+					return sub, true
+				}
+			}
+			return nil, false
+		}
+		return m, true
+	}
+	return nil, false
+}
+
+// userArrayValue 从原始请求节点提取数组值
+func userArrayValue(raw interface{}, hasRaw bool) ([]interface{}, bool) {
+	if !hasRaw || raw == nil {
+		return nil, false
+	}
+	if arr, ok := raw.([]interface{}); ok {
+		return arr, true
+	}
+	if m, ok := raw.(map[string]interface{}); ok {
+		if _, isTpl := m["type"]; isTpl {
+			if v, has := m["value"]; has {
+				if sub, ok := v.([]interface{}); ok {
+					return sub, true
+				}
+			}
+			v1, has1 := m["attrs"]
+			v2, has2 := m["enumValues"]
+			if has1 || has2 {
+				sub1, ok1 := v1.([]interface{})
+				sub2, ok2 := v2.([]interface{})
+				if ok1 {
+					if ok2 {
+						return sub2, true
+					}
+					return sub1, true
+				}
+				if ok2 {
+					return sub2, true
+				}
+			}
+		}
+	}
+	return nil, false
+}
+
+// userScalarValue 从原始请求节点提取标量值
+func userScalarValue(raw interface{}, hasRaw bool) interface{} {
+	if !hasRaw {
+		return nil
+	}
+	if m, ok := raw.(map[string]interface{}); ok {
+		if _, isTpl := m["type"]; isTpl {
+			return m["value"]
+		}
+	}
+	return raw
+}
+
+// backfillDefault 空值回填 defaultValue：
+// 模板格式节点写 value 键；纯值直接覆盖；字段缺失则补一个模板格式节点供下游产出默认值。
+func backfillDefault(parent map[string]interface{}, key string, raw interface{}, hasRaw bool, tmpl *dto.Template) {
+	if tmpl.DefaultValue == nil {
+		return
+	}
+	if m, ok := raw.(map[string]interface{}); ok {
+		if _, isTpl := m["type"]; isTpl {
+			m["value"] = tmpl.DefaultValue
+			return
+		}
+	}
+	if hasRaw {
+		parent[key] = tmpl.DefaultValue
+		return
+	}
+	parent[key] = map[string]interface{}{
+		"type":  tmpl.Type,
+		"value": tmpl.DefaultValue,
+	}
+}
+
+// validateEnumValues 校验数组枚举项：逐项取请求 enumValue.attrs 作为字段容器，
+// 递归校验每个子字段（必填/约束）并回填空值的 defaultValue。与旧 checkArrayTmpl 行为对齐。
+func validateEnumValues(raw interface{}, hasRaw bool, path string, strictUnknown, backfill bool) error {
+	if !hasRaw {
+		return nil
+	}
+	rawMap, ok := raw.(map[string]interface{})
+	if !ok {
+		return nil
+	}
+	evs, ok := rawMap["enumValues"].([]interface{})
+	if !ok {
+		return nil
+	}
+	for i, ev := range evs {
+		evMap, ok := ev.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		attrs, ok := evMap["attrs"].(map[string]interface{})
+		if !ok {
+			continue
+		}
+		for subKey, subTmpl := range attrs {
+			subPath := fmt.Sprintf("%s.enumValues[%d].%s", path, i, subKey)
+			if err := validateNode(attrs, subKey, subTmpl, subPath, strictUnknown, backfill); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}
+
+// maxArrayCount 取数组字段的数量上限：UploadTotalMaxCount 优先，其次各 UploadRule.MaxCount 之和；未配置返回 0
+func maxArrayCount(tmpl *dto.Template) int {
+	if tmpl.Constraint.UploadTotalMaxCount > 0 {
+		return tmpl.Constraint.UploadTotalMaxCount
+	}
+	total := 0
+	for _, rule := range tmpl.Constraint.UploadRules {
+		total += rule.MaxCount
+	}
+	return total
+}
+
+// arrayElementPrototype 取数组元素模板原型（attrs 优先，其次 enumValues）
+func arrayElementPrototype(tmpl *dto.Template) map[string]interface{} {
+	if attrs, ok := tmpl.Attrs.([]interface{}); ok && len(attrs) > 0 {
+		if m, ok := attrs[0].(map[string]interface{}); ok {
+			return m
+		}
+	}
+	if len(tmpl.EnumValues) > 0 {
+		if m, ok := tmpl.EnumValues[0].(map[string]interface{}); ok {
+			return m
+		}
+	}
+	return nil
+}
+
+// matchArraySlotProto 按元素 attrs 键集与各槽位原型 attrs 键集的重合度匹配最合适的槽位原型。
+// 解析后数组下标可能塌缩（resolveArray 丢弃空元素），不能按 index 对齐，故用结构匹配。
+// 键集完全无重合时返回 nil（跳过该校验，避免用错误原型误报必填）。
+func matchArraySlotProto(tmpl *dto.Template, container map[string]interface{}) map[string]interface{} {
+	var best map[string]interface{}
+	bestCount := -1
+	for _, ev := range tmpl.EnumValues {
+		evMap, ok := ev.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		attrs, ok := evMap["attrs"].(map[string]interface{})
+		if !ok {
+			continue
+		}
+		count := 0
+		for k := range container {
+			if _, has := attrs[k]; has {
+				count++
+			}
+		}
+		if count > bestCount {
+			bestCount = count
+			best = attrs
+		}
+	}
+	if bestCount <= 0 {
+		return nil
+	}
+	return best
+}
+
+// isValueEmptyByType 按 tmpl.Type 判断是否为"业务空值"
+func isValueEmptyByType(tmpl *dto.Template) bool {
+	switch tmpl.Type {
+	case TypeString:
+		return g.IsEmpty(gconv.String(tmpl.Value))
+	case TypeNumber:
+		return g.IsEmpty(gconv.Float64(tmpl.Value))
+	case TypeBool:
+		return tmpl.Value == nil
+	case TypeObject:
+		return g.IsEmpty(gconv.Map(tmpl.Value))
+	case TypeArray:
+		return g.IsEmpty(gconv.SliceAny(tmpl.Value))
+	case TypeNull:
+		return true
+	default:
+		return true
+	}
+}
+
+// checkStringTmpl 字符串类型校验
+func checkStringTmpl(tmpl *dto.Template) error {
+	val := gconv.String(tmpl.Value)
+	if tmpl.Required && gutil.IsEmpty(val) {
+		return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", tmpl.Label)
+	}
+	ct := tmpl.Constraint
+	if gutil.IsEmpty(ct) {
+		return nil
+	}
+	if tmpl.FieldType == "string" || tmpl.FieldType == "textarea" {
+		if ct.MinLength > 0 && len(val) < ct.MinLength {
+			return fmt.Errorf("字段 [%s] 长度应大于等于 %d，当前长度 %d", tmpl.Label, ct.MinLength, len(val))
+		}
+		if ct.MaxLength > 0 && len(val) > ct.MaxLength {
+			return fmt.Errorf("字段 [%s] 长度应小于等于 %d，当前长度 %d", tmpl.Label, ct.MaxLength, len(val))
+		}
+	}
+	return nil
+}
+
+// checkNumberTmpl 数字类型校验
+func checkNumberTmpl(tmpl *dto.Template) error {
+	ct := tmpl.Constraint
+	if gutil.IsEmpty(ct) {
+		return nil
+	}
+
+	switch ct.NumberType {
+	case TypeNumberInt:
+		val := gconv.Int(tmpl.Value)
+		if tmpl.Required && gutil.IsEmpty(val) {
+			return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", tmpl.Label)
+		}
+		minVal := gconv.Int(ct.Min)
+		maxVal := gconv.Int(ct.Max)
+		if !g.IsEmpty(minVal) && val < minVal {
+			return fmt.Errorf("字段 [%s] 值 %d 不应小于 最小值 %d", tmpl.Label, val, minVal)
+		}
+		if !g.IsEmpty(maxVal) && val > maxVal {
+			return fmt.Errorf("字段 [%s] 值 %d 不应大于 最大值 %d", tmpl.Label, val, maxVal)
+		}
+
+	case TypeNumberFloat:
+		val := gconv.Float64(tmpl.Value)
+		if tmpl.Required && gutil.IsEmpty(val) {
+			return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", tmpl.Label)
+		}
+		minVal := gconv.Float64(ct.Min)
+		maxVal := gconv.Float64(ct.Max)
+		if !g.IsEmpty(minVal) && val < minVal {
+			return fmt.Errorf("字段 [%s] 值 %.2f 不应小于 最小值 %.2f", tmpl.Label, val, minVal)
+		}
+		if !g.IsEmpty(maxVal) && val > maxVal {
+			return fmt.Errorf("字段 [%s] 值 %.2f 不应大于 最大值 %.2f", tmpl.Label, val, maxVal)
+		}
+
+	default:
+		return fmt.Errorf("字段 [%s] 数字类型 [%s] 错误，仅支持 int/float", tmpl.Label, ct.NumberType)
+	}
+	return nil
+}
+
+// checkBoolTmpl 布尔类型校验
+func checkBoolTmpl(tmpl *dto.Template) error {
+	val := gconv.Bool(tmpl.Value)
+	if tmpl.Required && gutil.IsEmpty(val) {
+		return fmt.Errorf("字段 [%s] 为必填项，但未提供有效值", tmpl.Label)
+	}
+	return nil
+}

service/worker.go

@@ -1,246 +0,0 @@
-package service
-
-import (
-	"context"
-	"fmt"
-	"strings"
-	"time"
-	"unicode/utf8"
-
-	"model-gateway/dao"
-	"model-gateway/model/entity"
-
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/os/grpool"
-	"github.com/tidwall/gjson"
-)
-
-var AsyncWorker = &asyncWorker{}
-
-type asyncWorker struct {
-}
-
-// RunOnce 由上层定时任务触发：一次性抢占并处理一批任务
-// - batchSize: 本次抢占数量
-// - goroutines: 本次并发数（协程池大小）
-func (w *asyncWorker) RunOnce(ctx context.Context, batchSize, goroutines int) (claimed int, err error) {
-	if batchSize <= 0 {
-		batchSize = 10
-	}
-	if goroutines <= 0 {
-		goroutines = 1
-	}
-	tasks, err := dao.Task.ClaimPendingGlobal(ctx, batchSize)
-	if err != nil {
-		return 0, err
-	}
-	if len(tasks) == 0 {
-		return 0, nil
-	}
-	pool := grpool.New(goroutines)
-	defer pool.Close()
-
-	claimed = len(tasks)
-	done := make(chan struct{}, claimed)
-	for _, t := range tasks {
-		task := t
-		_ = pool.AddWithRecover(ctx, func(ctx context.Context) {
-			w.handleOne(ctx, task, 0)
-			done <- struct{}{}
-		}, func(ctx context.Context, e error) {
-			if e != nil {
-				_ = dao.Task.UpdateFailedGlobal(ctx, task.Id, fmt.Sprintf("worker panic: %v", e))
-				ReleaseQueueSlot(ctx, task.ModelName, task.TaskID)
-			}
-			done <- struct{}{}
-		})
-	}
-	for i := 0; i < claimed; i++ {
-		<-done
-	}
-	return claimed, nil
-}
-
-// RunByTaskID 创建任务后立即异步尝试执行当前任务：
-// - 只定向抢占当前 taskId 对应的 pending 任务
-// - 若任务已被其它 worker 抢走/已不在 pending，则直接返回
-func (w *asyncWorker) RunByTaskID(ctx context.Context, taskID string, epicycleId int64) error {
-	task, err := dao.Task.ClaimPendingByTaskIDGlobal(ctx, taskID)
-	if err != nil {
-		return err
-	}
-	if task == nil {
-		return nil
-	}
-	w.handleOne(ctx, task, epicycleId)
-	return nil
-}
-
-func (w *asyncWorker) handleOne(ctx context.Context, t *entity.AsynchTask, epicycleId int64) {
-	// 从任务入库的 request_payload 里恢复 payload + headers
-	payload, headers := parseStoredPayload(t.RequestPayload)
-	if len(headers) > 0 {
-		ctx = setTaskHeadersToCtx(ctx, headers)
-	}
-
-	// 1) 拉取模型配置
-	m, err := dao.Model.GetByModelNameForTenant(ctx, t.TenantId, t.ModelName)
-	if err != nil {
-		_ = dao.Task.UpdateFailedGlobal(ctx, t.Id, err.Error())
-		ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-		// ============ 失败回调 ============
-		t.State = 3
-		t.ErrorMsg = err.Error()
-		go triggerCallback(context.WithoutCancel(ctx), t)
-		// ================================
-		return
-	}
-	if m == nil || (m.Enabled != nil && *m.Enabled != 1) {
-		errMsg := "模型不存在或未启用"
-		_ = dao.Task.UpdateFailedGlobal(ctx, t.Id, errMsg)
-		ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-		// ============ 失败回调 ============
-		t.State = 3
-		t.ErrorMsg = errMsg
-		go triggerCallback(context.WithoutCancel(ctx), t)
-		// ================================
-		return
-	}
-
-	// 2) 分布式并发限制
-	semKey := fmt.Sprintf("asynch:sem:%s", t.ModelName)
-	leaseSeconds := int64(3600)
-	maxC := GetRuntimeMaxConcurrency(ctx, t.ModelName, m.MaxConcurrency)
-	acquired, err := acquireSemaphore(ctx, semKey, maxC, leaseSeconds)
-	if err != nil {
-		_ = dao.Task.UpdateFailedGlobal(ctx, t.Id, err.Error())
-		ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-		// ============ 失败回调 ============
-		t.State = 3
-		t.ErrorMsg = err.Error()
-		go triggerCallback(context.WithoutCancel(ctx), t)
-		// ================================
-		return
-	}
-	if !acquired {
-		// 并发满了：放回排队，不回调（不是失败）
-		_ = w.rollbackToPending(ctx, t.Id)
-		return
-	}
-	defer func() {
-		_ = releaseSemaphore(ctx, semKey)
-	}()
-
-	// 3) 调用模型服务
-	if payload == nil {
-		payload = map[string]any{
-			"taskId":   t.TaskID,
-			"inputRef": t.InputRef,
-		}
-	}
-	var (
-		data        []byte
-		contentType string
-		ext         string
-		textResult  string
-	)
-
-	// phase=1 表示模型已成功但 OSS 上传失败：优先从临时文件加载
-	if t.Phase == 1 && strings.TrimSpace(t.TmpFile) != "" {
-		data, err = loadTmpResult(t.TmpFile)
-		if err == nil && len(data) > 0 {
-			contentType, ext = DetectFileType(data)
-		} else {
-			data = nil
-		}
-	}
-	if data == nil {
-		// 统计
-		_ = dao.Stat.IncRequestCount(ctx, time.Now(), int64(t.TenantId), t.Creator, t.ModelName)
-		// 核心调用
-		data, err = InvokeModel(ctx, m, payload, t.ModelKey)
-		if err != nil {
-			_ = dao.Task.UpdateFailedGlobal(ctx, t.Id, err.Error())
-			ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-			// ============ 失败回调 ============
-			t.State = 3
-			t.ErrorMsg = err.Error()
-			go triggerCallback(context.WithoutCancel(ctx), t)
-			// ================================
-			return
-		}
-		contentType, ext = DetectFileType(data)
-		if utf8.Valid(data) && (strings.HasPrefix(contentType, "text/") || contentType == "application/json") {
-			textResult = string(data)
-		}
-		tmpPath, err := saveTmpResult(t.TaskID, data, ext)
-		if err == nil && tmpPath != "" {
-			t.TmpFile = tmpPath
-			t.Phase = 1
-			_ = dao.Task.UpdateTmpAfterModelGlobal(ctx, t.Id, tmpPath)
-		}
-	}
-
-	// 4) 存储 OSS
-	ossURL, err := Storage.UploadByTask(ctx, t, data, ext, contentType)
-	if err != nil {
-		// OSS 阶段失败：保留临时文件，下一轮仅重试 OSS
-		_ = dao.Task.UpdateFailedKeepTmpGlobal(ctx, t.Id, err.Error())
-		ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-		// ============ OSS失败不回调（还会重试） ============
-		// 注意：OSS失败保留临时文件，下次重试，所以这里不触发最终回调
-		// 如果已经重试多次还没成功，需要在任务超时或超过最大重试次数时才回调失败
-		return
-	}
-
-	// 5) 更新任务状态成功
-	fileType := strings.TrimPrefix(ext, ".")
-	if fileType == "" {
-		fileType = contentType
-	}
-	if err := dao.Task.UpdateSuccessGlobal(
-		ctx,
-		t.Id,
-		ossURL,
-		fileType,
-		textResult,
-		int64(len(data)),
-		nil,
-		GetExpendTokens(m.TokenMapping, textResult),
-	); err != nil {
-		g.Log().Errorf(ctx, "[worker] update success failed: %v", err)
-		return
-	}
-
-	// 成功/失败均不再占用 queue_limit
-	ReleaseQueueSlot(ctx, t.ModelName, t.TaskID)
-
-	// 6) 成功回调
-	t.State = 2
-	t.OssFile = ossURL
-	t.FileType = fileType
-	t.TextResult = textResult
-	g.Log().Infof(ctx, "[CALLBACK][DISPATCH] taskId=%s bizName=%s callbackUrl=%s", t.TaskID, t.BizName, t.CallbackURL)
-	go triggerCallback(context.WithoutCancel(ctx), t)
-	// ============ 如果有 epicycleId，也触发业务回调 ============
-	if epicycleId != 0 {
-		go triggerPromptsCallback(context.WithoutCancel(ctx), t, epicycleId)
-	}
-
-	// 成功后清理临时文件
-	deleteTmpResult(t.TmpFile)
-}
-
-func (w *asyncWorker) rollbackToPending(ctx context.Context, id int64) error {
-	return dao.Task.RollbackToPendingGlobal(ctx, id)
-}
-
-// GetExpendTokens 根据映射路径从 textResult 中提取消耗 token 值
-func GetExpendTokens(tokenMapping string, textResult string) int {
-	value := gjson.Get(textResult, tokenMapping)
-	if value.Exists() {
-		return int(value.Int())
-	} else {
-		return len(textResult)
-	}
-}

timezone/timezone

@@ -1 +0,0 @@
-Asia/Shanghai

update.sql

@@ -1,264 +1,389 @@
--- model-asynch 核心表(pgsql)
--- 1) asynch_models：模型配置
--- 2) asynch_task：异步任务
--- 3) logs_model_op：操作日志(统计用)
--- 4) logs_model_stat：按天模型请求统计(限流/监控用)
-
 -- =========================
--- 1) asynch_models
+-- model_gateway_models
 -- =========================
-CREATE TABLE IF NOT EXISTS asynch_models (
+CREATE TABLE IF NOT EXISTS model_gateway_models (
-    -- 基础字段
+                                                    id              int8 PRIMARY KEY,
-    id                         BIGINT PRIMARY KEY,                 -- 主键ID(非自增)
+                                                    tenant_id       int8 NOT NULL DEFAULT 0,
-    tenant_id                  BIGINT NOT NULL DEFAULT 0,          -- 租户ID
+                                                    creator         varchar(64) NOT NULL,
-    creator                    VARCHAR(64) NOT NULL,               -- 创建人
+    created_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    created_at                 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP, -- 创建时间
+    updater         varchar(64) NOT NULL,
-    updater                    VARCHAR(64) NOT NULL,               -- 更新人
+    updated_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    updated_at                 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP, -- 更新时间
+    deleted_at      timestamp(6),
-    deleted_at                 TIMESTAMP(6),                       -- 删除时间(软删)
+    model_name      varchar(128) NOT NULL,
-    -- 业务字段
+    model_type      int2 NOT NULL DEFAULT 0,
-    model_name                 VARCHAR(128) NOT NULL,              -- 模型名称
+    operator_name   varchar(64) NOT NULL DEFAULT '',
-    model_type                SMALLINT NOT NULL DEFAULT 0,        -- 模型类型
+    base_url        varchar(256) NOT NULL,
-    base_url                   VARCHAR(256) NOT NULL,              -- 模型地址
+    http_method     varchar(8) NOT NULL DEFAULT 'POST',
-    http_method                VARCHAR(8) NOT NULL DEFAULT 'POST', -- 请求方式 GET/POST
+    head_msg        jsonb NOT NULL DEFAULT '{}',
-    head_msg                   VARCHAR(1024) DEFAULT '',           -- 请求头绑定(支持多个,逗号分隔)示例 X-API:xxx,operation:true
+    api_key         varchar(256) NOT NULL DEFAULT '',
-    is_private                 SMALLINT NOT NULL DEFAULT 0,        -- 是否私有化 0-私有 1-公共
+    is_private      int2 NOT NULL DEFAULT 0,
-    enabled                    SMALLINT NOT NULL DEFAULT 1,        -- 是否启用 0停用 1-启用
+    enabled         int2 NOT NULL DEFAULT 1,
-    is_chat_model              SMALLINT NOT NULL DEFAULT 0,        -- 是否为对话模型 0-否 1-是
+    is_chat_model   int2 NOT NULL DEFAULT 0,
-    is_owner              SMALLINT NOT NULL DEFAULT 99,        -- 1=当前用户创建的，0=超级管理员的
+    is_owner        int2 NOT NULL DEFAULT 99,
-    api_key                    VARCHAR(256) NOT NULL DEFAULT '',   -- 调用凭证,密钥
+    form_json       jsonb NOT NULL DEFAULT '{}',
-    prompt                     TEXT NOT NULL DEFAULT '',           -- 提示词内容(文本)
+    request_mapping  jsonb NOT NULL DEFAULT '{}',
-    form_json                  JSONB NOT NULL DEFAULT '{}'::jsonb, -- 表单结构(用于前端渲染)
+    response_mapping jsonb NOT NULL DEFAULT '{}',
-    request_mapping            JSONB NOT NULL DEFAULT '{}'::jsonb  -- 请求映射
+    response_body   varchar(128) NOT NULL DEFAULT '',
-    response_mapping           JSONB NOT NULL DEFAULT '{}'::jsonb, -- 返回映射
+    token_config    jsonb NOT NULL DEFAULT '{}',
-    response_body              JSONB NOT NULL DEFAULT '{}'::jsonb, -- 返回主体
+    extend_mapping  jsonb NOT NULL DEFAULT '{}',
-    max_concurrency            INT NOT NULL DEFAULT 10,            -- 单模型最大并发
+    query_config    jsonb NOT NULL DEFAULT '{}',
-    queue_limit                INT NOT NULL DEFAULT 1000,          -- 排队上限(近似控制)
+    stream_config   jsonb NOT NULL DEFAULT '{}',
-    timeout_seconds            INT NOT NULL DEFAULT 600,           -- 调用模型服务超时(秒)
+    first_frame     varchar(128) NOT NULL DEFAULT '',
-    expected_seconds           INT NOT NULL DEFAULT 600,           -- 模型预计执行时间(秒)
+    last_frame      varchar(128) NOT NULL DEFAULT '',
-    retry_times                SMALLINT NOT NULL DEFAULT 3,        -- 失败重试次数
+    max_concurrency int4 NOT NULL DEFAULT 10,
-    retry_queue_max_seconds    INT NOT NULL DEFAULT 600,           -- 失败重试最大排队时间(秒 0=插队到队首；>0=排队超过该时间后插队，否则仍到队尾)
+    timeout_seconds int4 NOT NULL DEFAULT 600,
-    auto_clean_seconds         INT NOT NULL DEFAULT 86400,         -- 已下载(state=4 后的保留时间(秒)，到期清理)
+    retry_times     int2 NOT NULL DEFAULT 3,
-    remark                     TEXT DEFAULT ''                     -- 备注
+    response_token_field varchar(128) NOT NULL DEFAULT '',
-    token_mapping              VARCHAR(128) NOT NULL DEFAULT '';   -- token 映射
+    call_mode       int2 NOT NULL DEFAULT 0,
-);
+    required_fields jsonb NOT NULL DEFAULT '[]',
+    max_tokens      int4 DEFAULT 0
+    );
 
-CREATE UNIQUE INDEX IF NOT EXISTS uk_asynch_models_tenant_creator_chat ON asynch_models(tenant_id, creator) WHERE is_chat_model = 1 AND deleted_at IS NULL;
+CREATE UNIQUE INDEX IF NOT EXISTS uk_model_gateway_models_tenant_creator_model ON model_gateway_models (tenant_id, creator, model_name);
-CREATE UNIQUE INDEX IF NOT EXISTS uk_asynch_models_tenant_model_name ON asynch_models(tenant_id, creator, model_name);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_models_model_name ON model_gateway_models (model_name);
-CREATE INDEX IF NOT EXISTS idx_asynch_models_tenant_id ON asynch_models(tenant_id);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_models_model_type ON model_gateway_models (model_type);
-CREATE INDEX IF NOT EXISTS idx_asynch_models_model_name ON asynch_models(model_name);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_models_tenant_id ON model_gateway_models (tenant_id);
-CREATE INDEX IF NOT EXISTS idx_asynch_models_model_type ON asynch_models(model_type);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_models_deleted_at ON model_gateway_models (deleted_at);
-CREATE INDEX IF NOT EXISTS idx_asynch_models_enabled ON asynch_models(enabled);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_models_enabled ON model_gateway_models (enabled);
-CREATE INDEX IF NOT EXISTS idx_asynch_models_deleted_at ON asynch_models(deleted_at);
 
-COMMENT ON TABLE asynch_models IS '模型配置表';
+COMMENT ON TABLE model_gateway_models IS '模型配置表';
-COMMENT ON COLUMN asynch_models.id IS '主键ID(非自增)';
+COMMENT ON COLUMN model_gateway_models.id IS '主键ID(非自增)';
-COMMENT ON COLUMN asynch_models.tenant_id IS '租户ID';
+COMMENT ON COLUMN model_gateway_models.tenant_id IS '租户ID';
-COMMENT ON COLUMN asynch_models.creator IS '创建人';
+COMMENT ON COLUMN model_gateway_models.creator IS '创建人';
-COMMENT ON COLUMN asynch_models.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_models.created_at IS '创建时间';
-COMMENT ON COLUMN asynch_models.updater IS '更新人';
+COMMENT ON COLUMN model_gateway_models.updater IS '更新人';
-COMMENT ON COLUMN asynch_models.updated_at IS '更新时间';
+COMMENT ON COLUMN model_gateway_models.updated_at IS '更新时间';
-COMMENT ON COLUMN asynch_models.deleted_at IS '删除时间(软删)';
+COMMENT ON COLUMN model_gateway_models.deleted_at IS '删除时间(软删)';
 
-COMMENT ON COLUMN asynch_models.model_name IS '模型名称';
+COMMENT ON COLUMN model_gateway_models.model_name IS '模型名称';
-COMMENT ON COLUMN asynch_models.model_type IS '模型类型';
+COMMENT ON COLUMN model_gateway_models.model_type IS '模型类型';
-COMMENT ON COLUMN asynch_models.base_url IS '模型地址';
+COMMENT ON COLUMN model_gateway_models.operator_name IS '运营商名称';
-COMMENT ON COLUMN asynch_models.http_method IS '请求方式 GET/POST';
+COMMENT ON COLUMN model_gateway_models.base_url IS '模型地址';
-COMMENT ON COLUMN asynch_models.head_msg IS '请求头绑定(支持多个,逗号分隔)示例 X-API:xxx,operation:true';
+COMMENT ON COLUMN model_gateway_models.http_method IS '请求方式 GET/POST';
-COMMENT ON COLUMN asynch_models.is_private IS '是否私有化 0-私有 1-公共';
+COMMENT ON COLUMN model_gateway_models.head_msg IS '请求头信息';
-COMMENT ON COLUMN asynch_models.enabled IS '是否启用 0停用 1-启用';
+COMMENT ON COLUMN model_gateway_models.api_key IS '调用凭证/密钥';
-COMMENT ON COLUMN asynch_models.is_chat_model IS '是否为对话模型 0-否 1-是';
+COMMENT ON COLUMN model_gateway_models.is_private IS '是否私有化：0-私有 1-公共';
-COMMENT ON COLUMN asynch_models.is_owner IS '1=当前用户创建的，0=超级管理员的';
+COMMENT ON COLUMN model_gateway_models.enabled IS '是否启用：0-停用 1-启用';
-COMMENT ON COLUMN asynch_models.api_key IS '调用凭证,密钥';
+COMMENT ON COLUMN model_gateway_models.is_chat_model IS '是否为对话模型：0-否 1-是';
-COMMENT ON COLUMN asynch_models.prompt IS '提示词内容(文本)';
+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 asynch_models.form_json IS '表单结构(用于前端渲染，也用于后端校验)';
+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 asynch_models.request_mapping IS '请求映射';
+COMMENT ON COLUMN model_gateway_models.form_json IS '动态表单结构';
-COMMENT ON COLUMN asynch_models.response_mapping IS '返回映射';
+COMMENT ON COLUMN model_gateway_models.request_mapping IS '请求映射';
-COMMENT ON COLUMN asynch_models.response_body IS '返回主体';
+COMMENT ON COLUMN model_gateway_models.response_mapping IS '返回映射';
-COMMENT ON COLUMN asynch_models.max_concurrency IS '单模型最大并发';
+COMMENT ON COLUMN model_gateway_models.response_body IS '返回主体';
-COMMENT ON COLUMN asynch_models.queue_limit IS '排队上限(近似控制)';
+COMMENT ON COLUMN model_gateway_models.token_config IS 'Token计算配置';
-COMMENT ON COLUMN asynch_models.timeout_seconds IS '调用模型服务超时(秒)';
+COMMENT ON COLUMN model_gateway_models.extend_mapping IS '附加映射';
-COMMENT ON COLUMN asynch_models.expected_seconds IS '模型预计执行时间(秒)';
+COMMENT ON COLUMN model_gateway_models.query_config IS '查询/回调配置';
-COMMENT ON COLUMN asynch_models.retry_times IS '失败重试次数';
+COMMENT ON COLUMN model_gateway_models.stream_config IS '流式输出配置';
-COMMENT ON COLUMN asynch_models.retry_queue_max_seconds IS '失败重试最大排队时间(秒 0=插队到队首；>0=排队超过该时间后插队，否则仍到队尾)';
+COMMENT ON COLUMN model_gateway_models.first_frame IS '首帧图片参数';
-COMMENT ON COLUMN asynch_models.auto_clean_seconds IS '已下载(state=4 后的保留时间(秒)，到期清理)';
+COMMENT ON COLUMN model_gateway_models.last_frame IS '尾帧图片参数';
-COMMENT ON COLUMN asynch_models.remark IS '备注';
+COMMENT ON COLUMN model_gateway_models.max_concurrency IS '最大并发数';
-COMMENT ON COLUMN asynch_models.token_mapping IS 'token映射';
+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.response_token_field IS '响应中消耗token的字段映射';
+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';
 
 
 
 -- =========================
--- 2) asynch_task
+-- model_gateway_task
 -- =========================
-CREATE TABLE IF NOT EXISTS asynch_task (
+CREATE TABLE IF NOT EXISTS model_gateway_task (
-    -- 基础字段
+                                                  id              int8 PRIMARY KEY,
-    id                         BIGINT PRIMARY KEY,                 -- 主键ID(非自增)
+                                                  tenant_id       int8 NOT NULL DEFAULT 0,
-    tenant_id                  BIGINT NOT NULL DEFAULT 0,          -- 租户ID
+                                                  creator         varchar(64) NOT NULL,
-    creator                    VARCHAR(64) NOT NULL,               -- 创建人
+    created_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    created_at                 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP, -- 创建时间
+    updater         varchar(64) NOT NULL,
-    updater                    VARCHAR(64) NOT NULL,               -- 更新人
+    updated_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    updated_at                 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP, -- 更新时间
+    deleted_at      timestamp(6),
-    deleted_at                 TIMESTAMP(6),                       -- 删除时间(软删)
+    model_name      varchar(128) NOT NULL,
+    task_id         varchar(64) NOT NULL,
+    biz_name        varchar(128) NOT NULL DEFAULT '',
+    callback_url    varchar(512) DEFAULT '',
+    state           int2 NOT NULL DEFAULT 0,
+    retry_count     int4 NOT NULL DEFAULT 0,
+    phase           int2 NOT NULL DEFAULT 0,
+    tmp_file        text DEFAULT '',
+    error_msg       text DEFAULT '',
+    result_file     jsonb NOT NULL DEFAULT '{}',
+    request_payload jsonb NOT NULL DEFAULT '{}',
+    text_result     jsonb NOT NULL DEFAULT '{}',
+    expend_tokens   int8 NOT NULL DEFAULT 0,
+    duration_seconds int8 NOT NULL DEFAULT 0,
+    epicycle_id     varchar(64) NOT NULL DEFAULT ''
+    );
 
-    -- 业务字段
+CREATE UNIQUE INDEX IF NOT EXISTS uk_model_gateway_task_tenant_creator_task_id ON model_gateway_task (tenant_id, creator, task_id);
-    model_name                 VARCHAR(128) NOT NULL,              -- 模型名称
+CREATE INDEX IF NOT EXISTS idx_model_gateway_task_task_id ON model_gateway_task (task_id);
-    task_id                    VARCHAR(64) NOT NULL,               -- 任务ID(对外返回)
+CREATE INDEX IF NOT EXISTS idx_model_gateway_task_state ON model_gateway_task (state);
-    biz_name                   VARCHAR(128) NOT NULL DEFAULT '',   -- 业务名称(调用方模块/系统)
+CREATE INDEX IF NOT EXISTS idx_model_gateway_task_deleted_at ON model_gateway_task (deleted_at);
-    callback_url               VARCHAR(512) DEFAULT '',            -- 回调地址(可选，用于后续业务通知)
-    model_key                  VARCHAR(1024) DEFAULT '',           -- 动态请求头(用于覆盖/补充模型配置 head_msg)，如 X-API-Key:xxx
-    state                      SMALLINT NOT NULL DEFAULT 0,        -- 0排队中/1执行中/2成功/3失败/4已下载
-    oss_file                   VARCHAR(512) DEFAULT '',            -- 结果文件OSS地址
-    file_type                  VARCHAR(32) DEFAULT '',             -- 文件类型(mp3/mp4/png/...)
-    file_size                  BIGINT NOT NULL DEFAULT 0,          -- 文件大小(字节)
-    error_msg                  TEXT DEFAULT '',                    -- 错误信息
-    started_at                 TIMESTAMP,                          -- 开始执行时间
-    finished_at                TIMESTAMP,                          -- 执行结束时间
-    duration_seconds           BIGINT NOT NULL DEFAULT 0,          -- 耗时(秒)：从创建到完成(成功/失败)整体耗时
-    expire_at                  TIMESTAMP,                          -- state=4 后写入，用于清理
-    retry_count                INT NOT NULL DEFAULT 0,             -- 已重试次数(不含首次)
-    enqueue_at                 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP, -- 入队时间(用于排队顺序)
-    phase                      SMALLINT NOT NULL DEFAULT 0,        -- 0模型阶段/1OSS阶段
-    tmp_file                   TEXT DEFAULT '',                    -- 临时结果文件路径(phase=1 时仅重试 OSS 上传)
-    input_ref                  TEXT DEFAULT '',                    -- 输入引用(如OSS/业务资源ID等)
-    request_payload            JSONB,                              -- 请求参数(可选)
-    text_result                TEXT DEFAULT '',                    -- 文本类结果(可选，支持直接回调)
-    epicycle_id                VARCHAR(64) DEFAULT '',             -- 轮次ID
-    expend_tokens              BIGINT NOT NULL DEFAULT 0           -- 消耗 token 数
-);
 
-CREATE UNIQUE INDEX IF NOT EXISTS uk_asynch_task_tenant_task_id ON asynch_task(tenant_id, task_id);
+COMMENT ON TABLE model_gateway_task IS '模型网关任务表';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_tenant_id ON asynch_task(tenant_id);
+COMMENT ON COLUMN model_gateway_task.id IS '主键ID';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_model_name ON asynch_task(model_name);
+COMMENT ON COLUMN model_gateway_task.tenant_id IS '租户ID';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_biz_name ON asynch_task(biz_name);
+COMMENT ON COLUMN model_gateway_task.creator IS '创建人';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_model_key ON asynch_task(model_key);
+COMMENT ON COLUMN model_gateway_task.created_at IS '创建时间';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_state ON asynch_task(state);
+COMMENT ON COLUMN model_gateway_task.updater IS '更新人';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_enqueue_at ON asynch_task(enqueue_at);
+COMMENT ON COLUMN model_gateway_task.updated_at IS '更新时间';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_updated_at ON asynch_task(updated_at);
+COMMENT ON COLUMN model_gateway_task.deleted_at IS '删除时间（软删）';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_expire_at ON asynch_task(expire_at);
+COMMENT ON COLUMN model_gateway_task.model_name IS '模型名称';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_deleted_at ON asynch_task(deleted_at);
+COMMENT ON COLUMN model_gateway_task.task_id IS '任务ID（对外返回）';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_epicycle_id ON asynch_task(epicycle_id);
+COMMENT ON COLUMN model_gateway_task.biz_name IS '业务名称（调用方模块/系统）';
-CREATE INDEX IF NOT EXISTS idx_asynch_task_expend_tokens ON asynch_task(expend_tokens);
+COMMENT ON COLUMN model_gateway_task.callback_url IS '回调地址';
-
+COMMENT ON COLUMN model_gateway_task.state IS '0排队中/1执行中/2成功/3失败/4已下载';
-COMMENT ON TABLE asynch_task IS '异步任务表';
+COMMENT ON COLUMN model_gateway_task.retry_count IS '已重试次数';
-COMMENT ON COLUMN asynch_task.id IS '主键ID(非自增)';
+COMMENT ON COLUMN model_gateway_task.phase IS '执行阶段：0模型阶段/1OSS阶段';
-COMMENT ON COLUMN asynch_task.tenant_id IS '租户ID';
+COMMENT ON COLUMN model_gateway_task.tmp_file IS '临时结果文件路径';
-COMMENT ON COLUMN asynch_task.creator IS '创建人';
+COMMENT ON COLUMN model_gateway_task.error_msg IS '错误信息';
-COMMENT ON COLUMN asynch_task.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_task.result_file IS '结果文件：{oss_file, file_type, file_size}';
-COMMENT ON COLUMN asynch_task.updater IS '更新人';
+COMMENT ON COLUMN model_gateway_task.request_payload IS '请求参数（JSON）';
-COMMENT ON COLUMN asynch_task.updated_at IS '更新时间';
+COMMENT ON COLUMN model_gateway_task.text_result IS '文本类结果';
-COMMENT ON COLUMN asynch_task.deleted_at IS '删除时间(软删)';
+COMMENT ON COLUMN model_gateway_task.expend_tokens IS '消耗token数';
-COMMENT ON COLUMN asynch_task.model_name IS '模型名称';
+COMMENT ON COLUMN model_gateway_task.duration_seconds IS '耗时（秒）';
-COMMENT ON COLUMN asynch_task.task_id IS '任务ID(对外返回)';
+COMMENT ON COLUMN model_gateway_task.epicycle_id IS '轮次ID';
-COMMENT ON COLUMN asynch_task.biz_name IS '业务名称(调用方模块/系统)';
-COMMENT ON COLUMN asynch_task.callback_url IS '回调地址(可选，用于后续业务通知)';
-COMMENT ON COLUMN asynch_task.model_key IS '动态请求头(用于覆盖/补充模型配置 head_msg)，如 X-API-Key:xxx';
-COMMENT ON COLUMN asynch_task.state IS '0排队中/1执行中/2成功/3失败/4已下载';
-COMMENT ON COLUMN asynch_task.oss_file IS '结果文件OSS地址';
-COMMENT ON COLUMN asynch_task.file_type IS '文件类型(mp3/mp4/png/...)';
-COMMENT ON COLUMN asynch_task.file_size IS '文件大小(字节)';
-COMMENT ON COLUMN asynch_task.error_msg IS '错误信息';
-COMMENT ON COLUMN asynch_task.started_at IS '开始执行时间';
-COMMENT ON COLUMN asynch_task.finished_at IS '执行结束时间';
-COMMENT ON COLUMN asynch_task.duration_seconds IS '耗时(秒)：从创建到完成(成功/失败)整体耗时';
-COMMENT ON COLUMN asynch_task.expire_at IS 'state=4 后写入，用于清理';
-COMMENT ON COLUMN asynch_task.retry_count IS '已重试次数(不含首次)';
-COMMENT ON COLUMN asynch_task.enqueue_at IS '入队时间(用于排队顺序)';
-COMMENT ON COLUMN asynch_task.phase IS '执行阶段 模型阶段/1OSS阶段(模型已成功,等待上传OSS)';
-COMMENT ON COLUMN asynch_task.tmp_file IS '临时结果文件路径(phase=1 时仅重试 OSS 上传)';
-COMMENT ON COLUMN asynch_task.input_ref IS '输入引用(如OSS/业务资源ID等)';
-COMMENT ON COLUMN asynch_task.request_payload IS '请求参数(可选,JSON)';
-COMMENT ON COLUMN asynch_task.text_result IS '文本类结果(可选，支持直接回调)';
-COMMENT ON COLUMN asynch_task.epicycle_id IS '轮次ID(用于标识同一轮次的任务)';
-COMMENT ON COLUMN asynch_task.expend_tokens IS '消耗 token 数';
 
 
 
 -- =========================
--- 3) logs_model_op
+-- model_gateway_log_stat
 -- =========================
-CREATE TABLE IF NOT EXISTS logs_model_op (
+CREATE TABLE IF NOT EXISTS model_gateway_log_stat (
-    -- 基础字段
+                                                      day             date NOT NULL,
-    id BIGINT PRIMARY KEY,
+                                                      tenant_id       int8 NOT NULL DEFAULT 0,
-    tenant_id BIGINT NOT NULL DEFAULT 0,
+                                                      creator         varchar(64) NOT NULL DEFAULT '',
-    creator VARCHAR(64) NOT NULL,
+    model_name      varchar(128) NOT NULL DEFAULT '',
+    request_count   int8 NOT NULL DEFAULT 0,
+    created_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    updated_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    PRIMARY KEY (day, tenant_id, creator, model_name)
+    );
+
+CREATE INDEX IF NOT EXISTS idx_model_gateway_log_stat_day ON model_gateway_log_stat (day);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_log_stat_creator ON model_gateway_log_stat (creator);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_log_stat_model_name ON model_gateway_log_stat (model_name);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_log_stat_tenant_day ON model_gateway_log_stat (tenant_id, day);
+
+COMMENT ON TABLE model_gateway_log_stat IS '按天统计表';
+COMMENT ON COLUMN model_gateway_log_stat.day IS '天（YYYY-MM-DD）';
+COMMENT ON COLUMN model_gateway_log_stat.tenant_id IS '租户ID';
+COMMENT ON COLUMN model_gateway_log_stat.creator IS '创建人';
+COMMENT ON COLUMN model_gateway_log_stat.model_name IS '模型名称';
+COMMENT ON COLUMN model_gateway_log_stat.request_count IS '请求次数';
+COMMENT ON COLUMN model_gateway_log_stat.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_log_stat.updated_at IS '更新时间';
+
+
+-- =========================
+-- model_gateway_logs_op
+-- =========================
+CREATE TABLE IF NOT EXISTS model_gateway_logs_op (
+                                                     id              int8 PRIMARY KEY,
+                                                     tenant_id       int8 NOT NULL DEFAULT 0,
+                                                     creator         varchar(64) NOT NULL,
+    created_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    updater         varchar(64) NOT NULL,
+    updated_at      timestamp(6) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    deleted_at      timestamp(6),
+    ip              varchar(64) DEFAULT '',
+    user_agent      varchar(256) DEFAULT '',
+    api_path        varchar(256) DEFAULT '',
+    http_method     varchar(16) DEFAULT '',
+    biz_name        varchar(128) NOT NULL DEFAULT '',
+    model_name      varchar(128) NOT NULL DEFAULT '',
+    task_id         varchar(64) NOT NULL DEFAULT '',
+    op_type         varchar(64) NOT NULL DEFAULT 'createTask',
+    success         int2 NOT NULL DEFAULT 1,
+    error_msg       text DEFAULT '',
+    cost_ms         int8 NOT NULL DEFAULT 0,
+    request_payload  jsonb,
+    response_payload jsonb
+    );
+
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_task_id ON model_gateway_logs_op (task_id);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_biz_name ON model_gateway_logs_op (biz_name);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_model_name ON model_gateway_logs_op (model_name);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_op_type ON model_gateway_logs_op (op_type);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_deleted_at ON model_gateway_logs_op (deleted_at);
+CREATE INDEX IF NOT EXISTS idx_model_gateway_logs_op_tenant_time ON model_gateway_logs_op (tenant_id, created_at);
+
+COMMENT ON TABLE model_gateway_logs_op IS '操作日志表';
+COMMENT ON COLUMN model_gateway_logs_op.id IS '主键ID（非自增）';
+COMMENT ON COLUMN model_gateway_logs_op.tenant_id IS '租户ID';
+COMMENT ON COLUMN model_gateway_logs_op.creator IS '创建人';
+COMMENT ON COLUMN model_gateway_logs_op.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_logs_op.updater IS '更新人';
+COMMENT ON COLUMN model_gateway_logs_op.updated_at IS '更新时间';
+COMMENT ON COLUMN model_gateway_logs_op.deleted_at IS '删除时间（软删）';
+COMMENT ON COLUMN model_gateway_logs_op.ip IS '客户端IP';
+COMMENT ON COLUMN model_gateway_logs_op.user_agent IS 'User-Agent';
+COMMENT ON COLUMN model_gateway_logs_op.api_path IS '接口路径';
+COMMENT ON COLUMN model_gateway_logs_op.http_method IS 'HTTP方法';
+COMMENT ON COLUMN model_gateway_logs_op.biz_name IS '业务名称（调用方模块/系统）';
+COMMENT ON COLUMN model_gateway_logs_op.model_name IS '模型名称';
+COMMENT ON COLUMN model_gateway_logs_op.task_id IS '任务ID';
+COMMENT ON COLUMN model_gateway_logs_op.op_type IS '操作类型';
+COMMENT ON COLUMN model_gateway_logs_op.success IS '是否成功：1成功/0失败';
+COMMENT ON COLUMN model_gateway_logs_op.error_msg IS '错误信息（失败时）';
+COMMENT ON COLUMN model_gateway_logs_op.cost_ms IS '耗时（毫秒）';
+COMMENT ON COLUMN model_gateway_logs_op.request_payload IS '请求 JSON';
+COMMENT ON COLUMN model_gateway_logs_op.response_payload IS '响应 JSON';
+
+
+-- =====================================================================================
+
+--------------------pgsql创建model_gateway_model_manage表语句---------------------------
+-- 模型管理表
+CREATE TABLE IF NOT EXISTS model_gateway_model_manage (
+      id BIGINT PRIMARY KEY,
+      tenant_id BIGINT NOT NULL DEFAULT 0,
+      creator VARCHAR(64) NOT NULL,
     created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
     updater VARCHAR(64) NOT NULL,
     updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    deleted_at TIMESTAMP(6),
+    deleted_at timestamp(6),
-    -- 基础审计信息
+
-    ip VARCHAR(64) DEFAULT '',
+    model_supplier VARCHAR(64) NOT NULL DEFAULT '',
-    user_agent VARCHAR(256) DEFAULT '',
-    api_path VARCHAR(256) DEFAULT '',
-    http_method VARCHAR(16) DEFAULT '',
-    -- 业务信息
-    biz_name VARCHAR(128) NOT NULL DEFAULT '', -- 调用方业务模块/系统
     model_name VARCHAR(128) NOT NULL DEFAULT '',
-    task_id VARCHAR(64) NOT NULL DEFAULT '',
+    model_type INT NOT NULL DEFAULT 0,
-    -- 统计字段
+    base_url VARCHAR(512) NOT NULL DEFAULT '',
-    op_type VARCHAR(64) NOT NULL DEFAULT 'createTask', -- 操作类型(默认创建任务)
+    http_method VARCHAR(32) NOT NULL DEFAULT '',
-    success SMALLINT NOT NULL DEFAULT 1, -- 1成功/0失败
+    system_model BOOLEAN NOT NULL DEFAULT false,
-    error_msg TEXT DEFAULT '',
+    private_model BOOLEAN NOT NULL DEFAULT false,
-    cost_ms BIGINT NOT NULL DEFAULT 0, -- 耗时(毫秒)
+    chat_model BOOLEAN NOT NULL DEFAULT false,
-    -- 请求/响应 JSON(用于后期统计分析)
+    invoke_type VARCHAR(32) NOT NULL DEFAULT '',
-    request_payload JSONB,
+    api_key VARCHAR(255) NOT NULL DEFAULT '',
-    response_payload JSONB
+    enabled BOOLEAN NOT NULL DEFAULT false,
-);
+    request_mapping JSONB DEFAULT '{}',
+    response_mapping JSONB DEFAULT '{}',
+    max_concurrency INT NOT NULL DEFAULT 0
+    );
 
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_tenant_time ON logs_model_op(tenant_id, created_at);
+-- 索引
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_model_name ON logs_model_op(model_name);
+CREATE INDEX idx_model_manage_tenant_id ON model_gateway_model_manage(tenant_id);
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_biz_name ON logs_model_op(biz_name);
+CREATE INDEX idx_model_manage_supplier ON model_gateway_model_manage(model_supplier);
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_task_id ON logs_model_op(task_id);
+CREATE INDEX idx_model_manage_type ON model_gateway_model_manage(model_type);
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_op_type ON logs_model_op(op_type);
+CREATE INDEX idx_model_manage_enabled ON model_gateway_model_manage(enabled);
-CREATE INDEX IF NOT EXISTS idx_logs_model_op_deleted_at ON logs_model_op(deleted_at);
+CREATE INDEX idx_model_manage_deleted_at ON model_gateway_model_manage(deleted_at);
 
-COMMENT ON TABLE logs_model_op IS '操作记录日志表(创建任务等，用于统计)';
+-- 字段与表注释
-COMMENT ON COLUMN logs_model_op.id IS '主键ID(非自增)';
+COMMENT ON TABLE model_gateway_model_manage IS '模型管理表';
-COMMENT ON COLUMN logs_model_op.tenant_id IS '租户ID';
+COMMENT ON COLUMN model_gateway_model_manage.id IS '主键ID';
-COMMENT ON COLUMN logs_model_op.creator IS '创建人';
+COMMENT ON COLUMN model_gateway_model_manage.tenant_id IS '租户ID';
-COMMENT ON COLUMN logs_model_op.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_model_manage.creator IS '创建人';
-COMMENT ON COLUMN logs_model_op.updater IS '更新人';
+COMMENT ON COLUMN model_gateway_model_manage.created_at IS '创建时间';
-COMMENT ON COLUMN logs_model_op.updated_at IS '更新时间';
+COMMENT ON COLUMN model_gateway_model_manage.updater IS '更新人';
-COMMENT ON COLUMN logs_model_op.deleted_at IS '删除时间(软删)';
+COMMENT ON COLUMN model_gateway_model_manage.updated_at IS '更新时间';
-COMMENT ON COLUMN logs_model_op.ip IS '客户端IP';
+COMMENT ON COLUMN model_gateway_model_manage.deleted_at IS '删除时间（软删）';
-COMMENT ON COLUMN logs_model_op.user_agent IS 'User-Agent';
+COMMENT ON COLUMN model_gateway_model_manage.model_supplier IS '模型供应商';
-COMMENT ON COLUMN logs_model_op.api_path IS '接口路径';
+COMMENT ON COLUMN model_gateway_model_manage.model_name IS '模型名称';
-COMMENT ON COLUMN logs_model_op.http_method IS 'HTTP方法';
+COMMENT ON COLUMN model_gateway_model_manage.model_type IS '模型类型';
-COMMENT ON COLUMN logs_model_op.biz_name IS '业务名称(调用方模块/系统)';
+COMMENT ON COLUMN model_gateway_model_manage.base_url IS '模型地址';
-COMMENT ON COLUMN logs_model_op.model_name IS '模型名称';
+COMMENT ON COLUMN model_gateway_model_manage.http_method IS 'http方法';
-COMMENT ON COLUMN logs_model_op.task_id IS '任务ID';
+COMMENT ON COLUMN model_gateway_model_manage.system_model IS '是否系统模型';
-COMMENT ON COLUMN logs_model_op.op_type IS '操作类型(如 createTask/getTaskResult/getTaskBatch 等)';
+COMMENT ON COLUMN model_gateway_model_manage.private_model IS '是否私有模型';
-COMMENT ON COLUMN logs_model_op.success IS '是否成功：1成功/0失败';
+COMMENT ON COLUMN model_gateway_model_manage.chat_model IS '是否聊天模型';
-COMMENT ON COLUMN logs_model_op.error_msg IS '错误信息(失败时)';
+COMMENT ON COLUMN model_gateway_model_manage.invoke_type IS '调用类型';
-COMMENT ON COLUMN logs_model_op.cost_ms IS '耗时(毫秒)';
+COMMENT ON COLUMN model_gateway_model_manage.api_key IS 'api key';
-COMMENT ON COLUMN logs_model_op.request_payload IS '请求 JSON';
+COMMENT ON COLUMN model_gateway_model_manage.enabled IS '是否启用';
-COMMENT ON COLUMN logs_model_op.response_payload IS '响应 JSON';
+COMMENT ON COLUMN model_gateway_model_manage.request_mapping IS '请求映射';
+COMMENT ON COLUMN model_gateway_model_manage.response_mapping IS '响应映射';
+COMMENT ON COLUMN model_gateway_model_manage.max_concurrency IS '最大并发数';
+--------------------pgsql创建model_gateway_model_manage表语句---------------------------
 
 
 -- =========================
--- 4) logs_model_stat
+-- 计费规则：model_manage 新增 price_config（JSONB），model_session 新增 total_cost（NUMERIC）
 -- =========================
-CREATE TABLE IF NOT EXISTS logs_model_stat (
+ALTER TABLE model_gateway_model_manage
-    day DATE NOT NULL,                     -- 天(YYYY-MM-DD)
+    ADD COLUMN IF NOT EXISTS price_config JSONB DEFAULT NULL;
-    tenant_id BIGINT NOT NULL DEFAULT 0,   -- 租户ID
+COMMENT ON COLUMN model_gateway_model_manage.price_config
-    creator VARCHAR(64) NOT NULL DEFAULT '', -- 创建人
+    IS '计费规则：{currency,unit,dimensions,rules,discount}，未配置为NULL（费用按0处理）';
-    model_name VARCHAR(128) NOT NULL DEFAULT '', -- 模型名称
-    request_count BIGINT NOT NULL DEFAULT 0, -- 请求次数
-    created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
-    PRIMARY KEY(day, tenant_id, creator, model_name)
-);
 
--- 便于时间段/租户/人/模型过滤
+ALTER TABLE model_gateway_session
-CREATE INDEX IF NOT EXISTS idx_logs_model_stat_tenant_day ON logs_model_stat(tenant_id, day);
+    ADD COLUMN IF NOT EXISTS total_cost NUMERIC DEFAULT 0;
-CREATE INDEX IF NOT EXISTS idx_logs_model_stat_day ON logs_model_stat(day);
+COMMENT ON COLUMN model_gateway_session.total_cost
-CREATE INDEX IF NOT EXISTS idx_logs_model_stat_model_name ON logs_model_stat(model_name);
+    IS '本次调用总费用（元），未配置计费规则为0';
-CREATE INDEX IF NOT EXISTS idx_logs_model_stat_creator ON logs_model_stat(creator);
 
-COMMENT ON TABLE logs_model_stat IS '按天模型请求统计(用于限流/监控)';
+-- =========================
-COMMENT ON COLUMN logs_model_stat.day IS '天(YYYY-MM-DD)';
+-- 异步任务计费：task_start 快照 media_type，task_end 记录 total_cost
-COMMENT ON COLUMN logs_model_stat.tenant_id IS '租户ID';
+-- 模型计费配置（price_config）任务完成时按 modelId 从 model_manage 现查，不在 task_start 快照
-COMMENT ON COLUMN logs_model_stat.creator IS '创建人';
+-- =========================
-COMMENT ON COLUMN logs_model_stat.model_name IS '模型名称';
+ALTER TABLE model_gateway_model_task_start
-COMMENT ON COLUMN logs_model_stat.request_count IS '请求次数';
+    ADD COLUMN IF NOT EXISTS media_type VARCHAR(32) DEFAULT NULL;
-COMMENT ON COLUMN logs_model_stat.created_at IS '创建时间';
+COMMENT ON COLUMN model_gateway_model_task_start.media_type
-COMMENT ON COLUMN logs_model_stat.updated_at IS '更新时间';
+    IS '输入媒体类型快照（audio/video，空=无媒体引用；shop 计费词汇，创建任务时按请求体参考媒体字段推导）';
+
+ALTER TABLE model_gateway_model_task_end
+    ADD COLUMN IF NOT EXISTS total_cost NUMERIC DEFAULT 0;
+COMMENT ON COLUMN model_gateway_model_task_end.total_cost
+    IS '本次调用总费用（元），未配置计费规则为0';
+
+-- =========================
+-- 模型引用化：model_manage 新增 ref_system_model_id（引用行指向系统模型）
+-- 系统模型 = 配置唯一来源；引用行只存 apiKey/enabled/chatModel，配置列留空
+-- =========================
+ALTER TABLE model_gateway_model_manage
+    ADD COLUMN IF NOT EXISTS ref_system_model_id BIGINT DEFAULT NULL;
+COMMENT ON COLUMN model_gateway_model_manage.ref_system_model_id
+    IS '引用的系统模型ID（NULL=非引用行；system_model=false 且该列非空=引用行）';
+CREATE INDEX IF NOT EXISTS idx_model_manage_ref_system_model_id
+    ON model_gateway_model_manage(ref_system_model_id);
+
+-- =========================
+-- 存量迁移：拷贝行 → 引用行（按同名系统模型匹配，一次性，幂等）
+-- 前提：存量拷贝均原封不动、无自定义（用户已确认）。若存在用户独立创建的同名自有模型会被误转，执行前复核一次：
+--   SELECT r.id, r.creator, r.model_name, r.ref_system_model_id, s.id AS sys_id
+--   FROM model_gateway_model_manage r
+--   JOIN model_gateway_model_manage s ON s.model_name = r.model_name AND s.system_model = true
+--   WHERE r.system_model = false AND r.ref_system_model_id IS NOT NULL;
+-- 列名以 entity orm 为准（update.sql 顶部旧 DDL 已过时）。
+-- =========================
+UPDATE model_gateway_model_manage r SET
+  ref_system_model_id = s.id,
+  base_url = NULL, http_method = NULL,
+  request_head_mapping = NULL, request_body_mapping = NULL,
+  request_business_field_mapping = NULL,
+  response_mapping = NULL, response_body_mapping = NULL,
+  response_business_field_mapping = NULL,
+  max_concurrency = 0, token_mapping = NULL, async_task_mapping = NULL,
+  token_predict_price = 0, token_predict_price_unit = NULL,
+  max_tokens = 0, min_duration = 0, max_duration = 0,
+  last_frame = NULL, error_message_mapping = NULL
+FROM model_gateway_model_manage s
+WHERE r.system_model = false AND s.system_model = true
+  AND r.model_name = s.model_name
+  AND r.ref_system_model_id IS NULL;
+
+-- =========================
+-- 错误消息映射：model_manage 新增 error_message_mapping（JSONB，schema 树形态，解析模型错误响应用）
+-- =========================
+ALTER TABLE model_gateway_model_manage
+    ADD COLUMN IF NOT EXISTS error_message_mapping JSONB DEFAULT NULL;
+COMMENT ON COLUMN model_gateway_model_manage.error_message_mapping
+    IS '错误消息映射：{code,message} 的 schema 树（type/attrs/value/defaultValue），解析模型错误响应，defaultValue 为成功码';
+
+
+-- =========================
+-- 错误重试记忆：LLM 分析上游模型错误是否可重试的持久知识库
+-- memory_key = SHA-256(upstream|error_code|归一化消息)，唯一；命中直接复用，永久有效
+-- =========================
+CREATE TABLE IF NOT EXISTS model_gateway_model_error_memory (
+                                                                id              BIGSERIAL PRIMARY KEY,
+                                                                tenant_id       BIGINT DEFAULT 0,
+                                                                creator         VARCHAR(64) DEFAULT '',
+    created_at      TIMESTAMPTZ DEFAULT now(),
+    updater         VARCHAR(64) DEFAULT '',
+    updated_at      TIMESTAMPTZ DEFAULT now(),
+    deleted_at      TIMESTAMPTZ DEFAULT NULL,
+    memory_key      CHAR(64) NOT NULL,
+    upstream        VARCHAR(512) NOT NULL DEFAULT '',
+    error_code      VARCHAR(128) NOT NULL DEFAULT '',
+    msg_fingerprint CHAR(32) NOT NULL DEFAULT '',
+    retryable       BOOLEAN NOT NULL DEFAULT false,
+    reason          VARCHAR(512) NOT NULL DEFAULT '',
+    analyzed_by     VARCHAR(128) NOT NULL DEFAULT ''
+    );
+CREATE UNIQUE INDEX IF NOT EXISTS uk_error_memory_memory_key
+    ON model_gateway_model_error_memory (memory_key)
+    WHERE deleted_at IS NULL;