后端双协议: 新增 anthropic.go(Complete/Stream/工具转换) + client.go 工厂按 Protocol 路由; llm.Client 加 Stream; llm_chat 三方法收敛复用 llm.Client; resolveProtocol 按 provider 路由; Agent CLI 注入协议。编码套餐(GLM-5.2/Kimi/MiMo/Anthropic)走 Anthropic /v1/messages。 SSH凭证修复: applyCredential 正确解析凭证 JSON 的 private_key(旧实现把整个 JSON 当私钥 → ssh.ParsePrivateKey 失败 → ssh-key 登录不通); AssignCredential/批量导入/单台 CRUD 三处统一; 单台补齐 credential_id 字段(原被忽略)。 容量预测: 新增 AI 深度解读(buildCapacityInsightPrompt + PredictiveCapacityTab 按钮, 复用 aiApi.chat)。 其他: DeepSeek 模型名更新为 deepseek-v4-flash/pro; 智能助手去掉重复模型配置按钮; LLMConfig 精简生成参数(Claude Code 式)。 Co-Authored-By: Claude <noreply@anthropic.com>
170 lines
5.6 KiB
Go
170 lines
5.6 KiB
Go
package controller
|
||
|
||
import (
|
||
"net/http"
|
||
"strings"
|
||
"time"
|
||
|
||
"ezops/model"
|
||
|
||
"github.com/gin-gonic/gin"
|
||
"gorm.io/gorm"
|
||
)
|
||
|
||
// LLMHandler 大语言模型配置 + 对话代理 // Logan.Li
|
||
type LLMHandler struct {
|
||
db *gorm.DB
|
||
client *http.Client
|
||
}
|
||
|
||
// NewLLMHandler 创建 LLM 处理器 // Logan.Li
|
||
func NewLLMHandler(db *gorm.DB) *LLMHandler {
|
||
return &LLMHandler{
|
||
db: db,
|
||
// 超时控制:LLM 推理可能较慢,给 120s
|
||
client: &http.Client{Timeout: 120 * time.Second},
|
||
}
|
||
}
|
||
|
||
// maskAPIKey 脱敏 API Key(仅保留后 4 位) // Logan.Li
|
||
func maskAPIKey(key string) string {
|
||
if len(key) <= 4 {
|
||
return strings.Repeat("*", len(key))
|
||
}
|
||
return strings.Repeat("*", len(key)-4) + key[len(key)-4:]
|
||
}
|
||
|
||
// getActiveConfig 读取启用的配置(脱敏后) // Logan.Li
|
||
func (h *LLMHandler) getActiveConfig(maskKey bool) (model.LLMConfig, error) {
|
||
var cfg model.LLMConfig
|
||
if err := h.db.First(&cfg).Error; err != nil {
|
||
return cfg, err
|
||
}
|
||
if maskKey {
|
||
cfg.APIKey = maskAPIKey(cfg.APIKey)
|
||
}
|
||
return cfg, nil
|
||
}
|
||
|
||
// GetLLMConfig 获取 LLM 配置(API Key 脱敏) // Logan.Li
|
||
// GET /api/v1/aiops/llm-config
|
||
func (h *LLMHandler) GetLLMConfig(c *gin.Context) {
|
||
cfg, err := h.getActiveConfig(true)
|
||
if err != nil {
|
||
// 未配置返回空对象(前端渲染空表单)
|
||
c.JSON(http.StatusOK, model.Response{Code: 200, Data: gin.H{
|
||
"provider": "", "model": "", "api_key": "",
|
||
"endpoint": "", "temperature": 0.7, "max_tokens": 2000,
|
||
"system_prompt": "", "enabled": false, "configured": false,
|
||
}})
|
||
return
|
||
}
|
||
c.JSON(http.StatusOK, model.Response{Code: 200, Data: gin.H{
|
||
"id": cfg.ID,
|
||
"provider": cfg.Provider,
|
||
"model": cfg.Model,
|
||
"api_key": cfg.APIKey,
|
||
"endpoint": cfg.Endpoint,
|
||
"temperature": cfg.Temperature,
|
||
"max_tokens": cfg.MaxTokens,
|
||
"system_prompt": cfg.SystemPrompt,
|
||
"enabled": cfg.Enabled,
|
||
"configured": true,
|
||
}})
|
||
}
|
||
|
||
// SaveLLMConfig 保存 LLM 配置(upsert 单行) // Logan.Li
|
||
// PUT /api/v1/aiops/llm-config
|
||
func (h *LLMHandler) SaveLLMConfig(c *gin.Context) {
|
||
var req model.SaveLLMConfigRequest
|
||
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
|
||
}
|
||
var cfg model.LLMConfig
|
||
h.db.First(&cfg)
|
||
// 若前端回传脱敏值(含 ****),保留原 Key
|
||
if strings.Contains(req.APIKey, "****") && cfg.ID != 0 {
|
||
req.APIKey = cfg.APIKey
|
||
}
|
||
cfg.Provider = req.Provider
|
||
cfg.Model = req.Model
|
||
cfg.APIKey = req.APIKey
|
||
cfg.Endpoint = req.Endpoint
|
||
cfg.Temperature = req.Temperature
|
||
cfg.MaxTokens = req.MaxTokens
|
||
cfg.SystemPrompt = req.SystemPrompt
|
||
cfg.Enabled = req.Enabled
|
||
if cfg.ID == 0 {
|
||
if err := h.db.Create(&cfg).Error; err != nil {
|
||
c.JSON(http.StatusInternalServerError, model.Response{Code: 500, Message: "保存失败: " + err.Error()})
|
||
return
|
||
}
|
||
} else {
|
||
if err := h.db.Save(&cfg).Error; err != nil {
|
||
c.JSON(http.StatusInternalServerError, model.Response{Code: 500, Message: "保存失败: " + err.Error()})
|
||
return
|
||
}
|
||
}
|
||
// 返回脱敏后的配置
|
||
cfg.APIKey = maskAPIKey(cfg.APIKey)
|
||
c.JSON(http.StatusOK, model.Response{Code: 200, Message: "保存成功", Data: gin.H{
|
||
"id": cfg.ID,
|
||
"provider": cfg.Provider,
|
||
"model": cfg.Model,
|
||
"api_key": cfg.APIKey,
|
||
"endpoint": cfg.Endpoint,
|
||
"temperature": cfg.Temperature,
|
||
"max_tokens": cfg.MaxTokens,
|
||
"system_prompt": cfg.SystemPrompt,
|
||
"enabled": cfg.Enabled,
|
||
"configured": true,
|
||
}})
|
||
}
|
||
|
||
// resolveEndpoint 根据提供商补全默认 endpoint // Logan.Li
|
||
func resolveEndpoint(provider, endpoint string) string {
|
||
if endpoint != "" {
|
||
return strings.TrimRight(endpoint, "/")
|
||
}
|
||
switch provider {
|
||
case "deepseek":
|
||
return "https://api.deepseek.com/v1"
|
||
case "glm": // 智谱 GLM 通用 OpenAI 兼容端点(编码套餐改用 https://open.bigmodel.cn/api/coding/paas/v4) // Logan.Li
|
||
return "https://open.bigmodel.cn/api/paas/v4"
|
||
case "qwen":
|
||
return "https://dashscope.aliyuncs.com/compatible-mode/v1"
|
||
case "moonshot":
|
||
return "https://api.moonshot.cn/v1"
|
||
case "kimi": // Kimi 编码套餐 OpenAI 兼容端点(注意:该端点有客户端白名单,通用调用可能被拒) // Logan.Li
|
||
return "https://api.kimi.com/coding/v1"
|
||
case "mimo": // 小米 MiMo OpenAI 兼容端点 // Logan.Li
|
||
return "https://api.mimo.mi.com/v1"
|
||
case "anthropic": // Anthropic 官方 Messages 协议(anthropic.go 客户端拼 /v1/messages) // Logan.Li
|
||
return "https://api.anthropic.com"
|
||
case "glm-coding": // 智谱 GLM 编码套餐 Anthropic 协议端点(模型 GLM-5.2) // Logan.Li
|
||
return "https://open.bigmodel.cn/api/anthropic"
|
||
case "ollama":
|
||
return "http://localhost:11434/v1"
|
||
default: // openai / custom
|
||
return "https://api.openai.com/v1"
|
||
}
|
||
}
|
||
|
||
// resolveProtocol 根据提供商推断接入协议(决定走 OpenAI 还是 Anthropic 客户端) // Logan.Li
|
||
// 编码套餐(智谱 GLM 编码 / Anthropic 官方)走 Anthropic Messages 协议;其余走 OpenAI 兼容协议。
|
||
func resolveProtocol(provider string) string {
|
||
if provider == "anthropic" || provider == "glm-coding" {
|
||
return "anthropic"
|
||
}
|
||
return "openai"
|
||
}
|
||
|
||
// defaultSystemPrompt AIOps 运维助手默认系统提示词 // Logan.Li
|
||
const defaultSystemPrompt = `你是 AIOps 智能运维助手,专注于企业级运维场景。你可以:
|
||
1. 分析服务器、容器、K8s 集群的性能瓶颈与异常
|
||
2. 解读告警事件并给出根因分析与处置建议
|
||
3. 生成 Shell / Ansible / Python 运维脚本
|
||
4. 提供容量规划、成本优化、高可用架构建议
|
||
回答请简洁、专业、可执行,优先给出具体命令或步骤。使用中文回答。`
|