aiops/backend/controller/llm_chat.go

package controller

import (
	"encoding/json"
	"fmt"
	"net/http"
	"strings"
	"time"

	"ezops/middleware"
	"ezops/model"
	"ezops/pkg/agentcli/llm"

	"github.com/gin-gonic/gin"
)

// toLLMMessages model.ChatMessage[] → llm.Message[]（智能助手纯对话，仅填 Role/Content） // Logan.Li
func toLLMMessages(msgs []model.ChatMessage) []llm.Message {
	out := make([]llm.Message, len(msgs))
	for i, m := range msgs {
		out[i] = llm.Message{Role: m.Role, Content: m.Content}
	}
	return out
}

// Chat AI 对话代理（统一走 llm.Client，按配置协议自动适配 OpenAI/Anthropic） // Logan.Li
// POST /api/v1/aiops/chat
func (h *LLMHandler) Chat(c *gin.Context) {
	var cfg model.LLMConfig
	if err := h.db.First(&cfg).Error; err != nil || !cfg.Enabled {
		c.JSON(http.StatusOK, model.Response{Code: 400, Message: "LLM 服务未配置或未启用，请在「模型配置」中完成设置"})
		return
	}

	var req model.ChatRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请求参数错误: " + err.Error()})
		return
	}

	// 组装系统提示词
	systemPrompt := cfg.SystemPrompt
	if strings.TrimSpace(systemPrompt) == "" {
		systemPrompt = defaultSystemPrompt
	}
	messages := append([]model.ChatMessage{{Role: "system", Content: systemPrompt}}, req.Messages...)

	resp, err := buildLLMClientFromConfig(cfg).Complete(c.Request.Context(), &llm.CompletionRequest{
		Messages: toLLMMessages(messages),
	})
	if err != nil {
		c.JSON(http.StatusOK, model.Response{Code: 502, Message: "LLM 调用失败: " + err.Error()})
		return
	}
	replyContent := resp.Message.Content

	// 持久化对话历史（按用户隔离，成对写入 user + assistant） // Logan.Li
	// 仅在成功时入库；错误气泡为瞬时 UI 不持久化
	uid := middleware.GetUserID(c)
	if uid != 0 && len(req.Messages) > 0 {
		lastUser := req.Messages[len(req.Messages)-1]
		now := time.Now()
		sid := req.SessionID // 前端传入或后端生成（兼容旧版客户端）
		if sid == "" {
			sid = fmt.Sprintf("sess-%d", now.UnixMilli())
		}
		h.db.Create(&[]model.ChatHistory{
			{UserID: uid, SessionID: sid, Role: "user", Content: lastUser.Content, CreatedAt: now},
			{UserID: uid, SessionID: sid, Role: "assistant", Content: replyContent, Model: cfg.Model, CreatedAt: now.Add(time.Millisecond)},
		})
	}

	c.JSON(http.StatusOK, model.Response{Code: 200, Data: model.ChatResponse{
		Content:    replyContent,
		Model:      cfg.Model,
		TokensUsed: resp.Usage.TotalTokens,
	}})
}

// ChatStream AI 流式对话代理（SSE） // Logan.Li
// POST /api/v1/aiops/chat/stream
// 客户端用 fetch-stream 消费，每条 SSE data 为 {"content": "..."}；出错发 event: error；完成发 event: done。
// 上游协议差异（OpenAI delta / Anthropic content_block_delta）由 llm.Client.Stream 屏蔽，
// controller 只负责 SSE 转发 + 客户端断开检测 + 历史持久化。
func (h *LLMHandler) ChatStream(c *gin.Context) {
	var cfg model.LLMConfig
	if err := h.db.First(&cfg).Error; err != nil || !cfg.Enabled {
		writeSSEError(c, 400, "LLM 服务未配置或未启用，请在「模型配置」中完成设置")
		return
	}

	var req model.ChatRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		writeSSEError(c, 400, "请求参数错误: "+err.Error())
		return
	}
	if len(req.Messages) == 0 {
		writeSSEError(c, 400, "消息不能为空")
		return
	}

	// 设置 SSE 响应头 // Logan.Li
	c.Writer.Header().Set("Content-Type", "text/event-stream")
	c.Writer.Header().Set("Cache-Control", "no-cache")
	c.Writer.Header().Set("Connection", "keep-alive")
	c.Writer.Header().Set("X-Accel-Buffering", "no") // Nginx 关闭缓冲
	c.Writer.WriteHeader(http.StatusOK)
	c.Writer.Flush() // 立即 flush 让客户端确认连接建立

	systemPrompt := cfg.SystemPrompt
	if strings.TrimSpace(systemPrompt) == "" {
		systemPrompt = defaultSystemPrompt
	}
	messages := append([]model.ChatMessage{{Role: "system", Content: systemPrompt}}, req.Messages...)

	// 流式补全：逐 token 回调转发 SSE；ctx 取消（客户端断开）由 llm.Client 内部终止上游读取
	var contentBuilder strings.Builder
	_, streamErr := buildLLMClientFromConfig(cfg).Stream(c.Request.Context(), &llm.CompletionRequest{
		Messages: toLLMMessages(messages),
	}, func(delta string) {
		contentBuilder.WriteString(delta)
		writeSSEChunk(c, delta)
	})
	fullContent := contentBuilder.String()

	// 出错且无任何内容 → 发 error 事件
	if streamErr != nil && fullContent == "" {
		writeSSEError(c, 502, "LLM 流式响应失败: "+streamErr.Error())
		return
	}

	// 持久化 + 结束事件（仅在非空且客户端未断开时入库；错误/中断不入库避免污染历史） // Logan.Li
	clientGone := c.Request.Context().Err() != nil
	if fullContent != "" && !clientGone {
		uid := middleware.GetUserID(c)
		if uid != 0 {
			lastUser := req.Messages[len(req.Messages)-1]
			now := time.Now()
			sid := req.SessionID
			if sid == "" {
				sid = fmt.Sprintf("sess-%d", now.UnixMilli())
			}
			h.db.Create(&[]model.ChatHistory{
				{UserID: uid, SessionID: sid, Role: "user", Content: lastUser.Content, CreatedAt: now},
				{UserID: uid, SessionID: sid, Role: "assistant", Content: fullContent, Model: cfg.Model, CreatedAt: now.Add(time.Millisecond)},
			})
		}
		c.Writer.WriteString("event: done\ndata: {\"model\":\"" + cfg.Model + "\"}\n\n")
		c.Writer.Flush()
	} else if !clientGone {
		// 无内容时通知前端结束（避免前端无限等待）
		c.Writer.WriteString("event: done\ndata: {}\n\n")
		c.Writer.Flush()
	}
}

// writeSSEChunk 写入一条内容增量 // Logan.Li
func writeSSEChunk(c *gin.Context, content string) {
	// 转义为 JSON 字符串避免换行符破坏 SSE 协议
	b, _ := json.Marshal(map[string]string{"content": content})
	c.Writer.WriteString("data: " + string(b) + "\n\n")
	c.Writer.Flush()
}

// writeSSEError 发送一个 error 事件（写入后立即结束流） // Logan.Li
func writeSSEError(c *gin.Context, code int, message string) {
	// 若 header 尚未写入（连接前错误），需先设置 SSE 头
	if !c.Writer.Written() {
		c.Writer.Header().Set("Content-Type", "text/event-stream")
		c.Writer.Header().Set("Cache-Control", "no-cache")
		c.Writer.Header().Set("Connection", "keep-alive")
		c.Writer.WriteHeader(http.StatusOK)
	}
	b, _ := json.Marshal(map[string]any{"code": code, "message": message})
	c.Writer.WriteString("event: error\ndata: " + string(b) + "\n\n")
	c.Writer.Flush()
}

// llmConfigFromTestRequest 将测试请求转为运行时配置（脱敏 Key 回退已保存值） // Logan.Li
func (h *LLMHandler) llmConfigFromTestRequest(req model.SaveLLMConfigRequest) (model.LLMConfig, string) {
	cfg := model.LLMConfig{
		Provider:         req.Provider,
		Model:            req.Model,
		APIKey:           req.APIKey,
		Endpoint:         req.Endpoint,
		Temperature:      req.Temperature,
		MaxTokens:        req.MaxTokens,
		ReasoningEnabled: req.ReasoningEnabled,
	}
	if cfg.Temperature == 0 {
		cfg.Temperature = 0.7
	}
	if cfg.MaxTokens == 0 {
		cfg.MaxTokens = 8192
	}
	if strings.Contains(req.APIKey, "****") {
		var saved model.LLMConfig
		if err := h.db.First(&saved).Error; err != nil {
			if !isLocalLLMProvider(req.Provider) {
				return cfg, "请填写 API Key"
			}
		} else {
			cfg.APIKey = saved.APIKey
		}
	}
	if cfg.Provider == "" {
		return cfg, "请选择提供商"
	}
	if cfg.Model == "" {
		return cfg, "请填写模型名"
	}
	if cfg.APIKey == "" && !isLocalLLMProvider(cfg.Provider) {
		return cfg, "请填写 API Key"
	}
	return cfg, ""
}

// TestLLMConnection 测试 LLM 连接（支持草稿配置，保存前即可验证） // Logan.Li
// POST /api/v1/aiops/llm-config/test
func (h *LLMHandler) TestLLMConnection(c *gin.Context) {
	var req model.SaveLLMConfigRequest
	_ = c.ShouldBindJSON(&req) // 空 body 时回退已保存配置 // Logan.Li

	var cfg model.LLMConfig
	var errMsg string
	if req.Provider != "" {
		cfg, errMsg = h.llmConfigFromTestRequest(req)
	} else {
		if err := h.db.First(&cfg).Error; err != nil {
			c.JSON(http.StatusOK, model.Response{Code: 400, Message: "请先填写配置并测试连接"})
			return
		}
		if cfg.APIKey == "" && !isLocalLLMProvider(cfg.Provider) {
			c.JSON(http.StatusOK, model.Response{Code: 400, Message: "请填写 API Key"})
			return
		}
	}
	if errMsg != "" {
		c.JSON(http.StatusOK, model.Response{Code: 400, Message: errMsg})
		return
	}
	resp, err := buildLLMClientFromConfig(cfg).Complete(c.Request.Context(), &llm.CompletionRequest{
		Messages: []llm.Message{{Role: "user", Content: "你是什么模型？请简短回答。"}},
	})
	if err != nil {
		c.JSON(http.StatusOK, model.Response{Code: 502, Message: "连接失败: " + err.Error()})
		return
	}
	c.JSON(http.StatusOK, model.Response{Code: 200, Message: "连接成功", Data: gin.H{
		"answer":            resp.Message.Content,
		"model":             resp.Model,
		"tokens_used":       resp.Usage.TotalTokens,
		"provider":          cfg.Provider,
		"reasoning_enabled": cfg.ReasoningEnabled,
	}})
}