Files
aiops/backend/controller/agent_cli.go
T
Logan 1acb0c49af feat: Agent CLI 与终端审计回放全面优化
- Agent CLI: 多轮对话、SSH 长连接、流式推理、Evolver 按用户隔离、历史侧栏
- 终端审计: 回放播放器、命令书签、筛选分页导出、命令输出统计、录制落盘修复
- 修复 session_commands 表迁移、Kill 竞态导致录制丢失、无录制会话回放提示
- 审计历史页只读(移除删除),移除回放弹框录制搜索
2026-07-01 12:13:53 +08:00

479 lines
15 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
// agent_cli.go - Agent CLI WebSocket 控制器 // Logan.Li
// 驱动 pkg/agentcli 引擎:关联主机(model.Server 凭据)→ 构建引擎 → ReAct 循环,
// 通过 WebSocket 实时推送思考/工具调用/输出/结果,高危命令经前端确认,过程持久化到 DB。
package controller
import (
"context"
"encoding/json"
"fmt"
"net/http"
"sync"
"time"
"github.com/gin-gonic/gin"
"github.com/gorilla/websocket"
"gorm.io/gorm"
"ezops/middleware"
"ezops/model"
"ezops/pkg/agentcli"
"ezops/pkg/agentcli/core"
"ezops/pkg/agentcli/llm"
"ezops/pkg/agentcli/safety"
"ezops/pkg/agentcli/tools"
"ezops/pkg/remoteexec"
)
// AgentCLIHandler Agent CLI 控制器 // Logan.Li
type AgentCLIHandler struct {
db *gorm.DB
}
// NewAgentCLIHandler 创建控制器 // Logan.Li
func NewAgentCLIHandler(db *gorm.DB) *AgentCLIHandler { return &AgentCLIHandler{db: db} }
// wsEvent WebSocket 事件信封(后端→前端) // Logan.Li
type wsEvent struct {
Type string `json:"type"`
Round int `json:"round,omitempty"`
SessionID uint `json:"session_id,omitempty"`
Thought string `json:"thought,omitempty"`
ToolName string `json:"tool_name,omitempty"`
ToolArgs map[string]interface{} `json:"tool_args,omitempty"`
Risk string `json:"risk,omitempty"`
Output string `json:"output,omitempty"`
Success bool `json:"success,omitempty"`
Duration string `json:"duration,omitempty"`
Tokens int `json:"tokens,omitempty"`
Answer string `json:"answer,omitempty"`
Message string `json:"message,omitempty"`
ConfirmID string `json:"confirm_id,omitempty"`
Command string `json:"command,omitempty"`
}
// wsControl 前端→后端控制消息 // Logan.Li
type wsControl struct {
Type string `json:"type"` // start/continue/confirm/deny/abort
Query string `json:"query,omitempty"` // start 时任务描述
SafetyMode string `json:"safety_mode,omitempty"` // strict/balanced/permissive
ConfirmID string `json:"confirm_id,omitempty"` // confirm/deny 时
}
// writeWSCloseError 在 agentConn 建立前,直接通过 conn 推送一条 error 事件再关闭连接 // Logan.Li
// 用于 WS 升级后的预检失败(SSH 凭据/LLM 配置),让前端能拿到具体原因而非"一直转圈"
func writeWSCloseError(conn *websocket.Conn, msg string) {
_ = conn.WriteJSON(wsEvent{Type: "error", Message: msg})
// 留出短暂窗口让浏览器接收 error 帧,再发送 close 帧:否则 close 帧紧随其后、叠加
// defer conn.Close() 触发的 TCP 关闭,可能令 error 帧被丢弃,前端表现就是"正在连接主机"
// 永久转圈、看不到真实原因 // Logan.Li
time.Sleep(120 * time.Millisecond)
_ = conn.WriteMessage(websocket.CloseMessage,
websocket.FormatCloseMessage(websocket.CloseNormalClosure, msg))
}
// agentConn 单个 WS 连接的运行态 // Logan.Li
type agentConn struct {
conn *websocket.Conn
db *gorm.DB
userID uint
server model.Server
session *model.AgentCLISession
taskCtx context.Context
cancel context.CancelFunc
totalTokens int
tokenMu sync.Mutex
writeMu sync.Mutex
confirms map[string]chan bool
cfMu sync.Mutex
convHistory []llm.Message // 多轮追问对话历史(不含 system) // Logan.Li
running bool
runningMu sync.Mutex
}
// Run WebSocket 入口:GET /ws/agent-cli?id=<serverID>&token=<jwt> // Logan.Li
func (h *AgentCLIHandler) Run(c *gin.Context) {
serverID := c.Query("id")
if serverID == "" {
c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "缺少主机 id"})
return
}
var server model.Server
if err := h.db.First(&server, serverID).Error; err != nil {
c.JSON(http.StatusNotFound, model.Response{Code: 404, Message: "主机不存在"})
return
}
// 先升级 WebSocket:SSH 凭据 / LLM 配置预检放到升级之后,失败时通过 WS error 事件
// 把真实原因推给前端,避免握手 400 导致前端"一直转圈"看不到原因 // Logan.Li
conn, err := upgrader.Upgrade(c.Writer, c.Request, nil)
if err != nil {
return
}
defer conn.Close()
// SSH 凭据预检
sshCfg, err := buildSSHConfig(h.db, server)
if err != nil {
writeWSCloseError(conn, "SSH 连接失败: "+err.Error())
return
}
// LLM 配置预检
var llmCfg model.LLMConfig
if err := h.db.First(&llmCfg).Error; err != nil || !llmCfg.Enabled {
writeWSCloseError(conn, "LLM 服务未配置或未启用,请先在「大模型设置」中配置")
return
}
ac := &agentConn{
conn: conn, db: h.db,
userID: middleware.GetUserID(c), server: server,
taskCtx: context.Background(),
confirms: make(map[string]chan bool),
}
// 推送主机上下文
ac.write(wsEvent{Type: "session", SessionID: 0,
Message: fmt.Sprintf("已连接主机 %s(%s),LLM: %s。请描述要执行的任务。", server.Hostname, server.IP, llmCfg.Model)})
// 读循环:处理前端控制消息
for {
_, data, err := conn.ReadMessage()
if err != nil {
ac.abort()
return
}
var ctrl wsControl
if json.Unmarshal(data, &ctrl) != nil {
continue
}
switch ctrl.Type {
case "start":
if ctrl.Query == "" {
ac.write(wsEvent{Type: "error", Message: "任务描述不能为空"})
continue
}
ac.startTask(sshCfg, ctrl.Query, ctrl.SafetyMode)
case "continue":
if ctrl.Query == "" {
ac.write(wsEvent{Type: "error", Message: "追问内容不能为空"})
continue
}
ac.continueTask(sshCfg, ctrl.Query, ctrl.SafetyMode)
case "confirm":
ac.resolveConfirm(ctrl.ConfirmID, true)
case "deny":
ac.resolveConfirm(ctrl.ConfirmID, false)
case "abort":
ac.abort()
}
}
}
// parseSafetyMode 解析前端传入的安全模式 // Logan.Li
func parseSafetyMode(mode string) safety.Mode {
switch mode {
case "strict":
return safety.ModeStrict
case "permissive":
return safety.ModePermissive
default:
return safety.ModeBalanced
}
}
// startTask 启动新任务(新会话,清空对话历史) // Logan.Li
func (ac *agentConn) startTask(sshCfg remoteexec.SSHConfig, query, safetyMode string) {
ac.abort()
var llmCfg model.LLMConfig
if err := ac.db.First(&llmCfg).Error; err != nil || !llmCfg.Enabled {
ac.write(wsEvent{Type: "error", Message: "LLM 服务未配置或未启用,请先在「大模型设置」中配置"})
return
}
title := query
if r := []rune(title); len(r) > 60 {
title = string(r[:60]) + "..."
}
sess := &model.AgentCLISession{
UserID: ac.userID, ServerID: ac.server.ID,
ServerName: ac.server.Hostname, ServerHost: ac.server.IP,
Title: title, Status: "running",
Provider: llmCfg.Provider, Model: llmCfg.Model,
}
if err := ac.db.Create(sess).Error; err != nil {
ac.write(wsEvent{Type: "error", Message: "创建会话失败: " + err.Error()})
return
}
ac.session = sess
ac.convHistory = nil
ac.setTotalTokens(0)
ac.runTask(sshCfg, sess, query, safetyMode, llmCfg, false)
}
// continueTask 在同一会话内追问(保留上下文) // Logan.Li
func (ac *agentConn) continueTask(sshCfg remoteexec.SSHConfig, query, safetyMode string) {
if ac.isRunning() {
ac.write(wsEvent{Type: "error", Message: "任务进行中,请等待完成或停止"})
return
}
if ac.session == nil {
ac.write(wsEvent{Type: "error", Message: "请先开始新任务"})
return
}
if ac.session.Status == "running" {
ac.write(wsEvent{Type: "error", Message: "当前会话仍在执行中"})
return
}
var llmCfg model.LLMConfig
if err := ac.db.First(&llmCfg).Error; err != nil || !llmCfg.Enabled {
ac.write(wsEvent{Type: "error", Message: "LLM 服务未配置或未启用,请先在「大模型设置」中配置"})
return
}
ac.runTask(sshCfg, ac.session, query, safetyMode, llmCfg, true)
}
// runTask 执行一轮 ReAct(新建或追问共用) // Logan.Li
func (ac *agentConn) runTask(sshCfg remoteexec.SSHConfig, sess *model.AgentCLISession, query, safetyMode string, llmCfg model.LLMConfig, isContinue bool) {
if ac.isRunning() {
ac.write(wsEvent{Type: "error", Message: "任务进行中,请等待完成或停止"})
return
}
prior := ac.convHistory
priorTokens := ac.getTotalTokens()
roundOffset := 0
if !isContinue {
prior = nil
priorTokens = 0
} else {
roundOffset = ac.lastRound()
}
coreCfg := core.DefaultConfig()
coreCfg.RoundOffset = roundOffset
engine, err := agentcli.NewEngine(agentcli.EngineConfig{
LLM: buildLLMConfigFromModel(llmCfg),
SSH: sshCfg,
SafetyMode: parseSafetyMode(safetyMode),
MemoryDir: fmt.Sprintf("data/agentcli/user_%d", ac.userID),
Core: coreCfg,
})
if err != nil {
ac.finish(sess, "error", "引擎初始化失败: "+err.Error(), ac.lastRound(), 0)
return
}
engine.SetCallback(ac.onEvent)
engine.SetConfirmFn(ac.confirmFn)
ctx, cancel := context.WithCancel(context.Background())
ac.taskCtx = ctx
ac.cancel = cancel
ac.setRunning(true)
ac.db.Model(sess).Update("status", "running")
ac.persistMsg(sess.ID, 0, "user", "user", query, "", "", "", "", false, "", 0)
serverInfo := fmt.Sprintf("主机名: %s | IP: %s:%d | 用户: %s | OS: %s",
ac.server.Hostname, ac.server.IP, ac.server.Port, ac.server.Username, ac.server.OS)
msg := "任务已开始"
if isContinue {
msg = "继续追问"
}
ac.write(wsEvent{Type: "session", SessionID: sess.ID, Message: msg})
go func() {
defer func() {
ac.setRunning(false)
engine.Close()
}()
start := time.Now()
result, err := engine.Run(ctx, query, serverInfo, prior, priorTokens)
if result != nil && len(result.History) > 0 {
ac.convHistory = result.History
ac.setTotalTokens(result.TotalTokens)
}
dur := time.Since(start)
if ctx.Err() != nil {
ac.finish(sess, "aborted", "任务已中断", ac.lastRound(), dur)
} else if err != nil {
ac.finish(sess, "error", err.Error(), ac.lastRound(), dur)
} else {
ac.finish(sess, "done", "", ac.lastRound(), dur)
}
}()
}
func (ac *agentConn) isRunning() bool {
ac.runningMu.Lock()
defer ac.runningMu.Unlock()
return ac.running
}
func (ac *agentConn) setRunning(v bool) {
ac.runningMu.Lock()
ac.running = v
ac.runningMu.Unlock()
}
// setTotalTokens 更新当前任务 token 计数 // Logan.Li
func (ac *agentConn) setTotalTokens(n int) {
ac.tokenMu.Lock()
ac.totalTokens = n
ac.tokenMu.Unlock()
}
func (ac *agentConn) getTotalTokens() int {
ac.tokenMu.Lock()
defer ac.tokenMu.Unlock()
return ac.totalTokens
}
// onEvent 引擎事件回调:推 WS + 持久化 // Logan.Li
func (ac *agentConn) onEvent(e core.Event) {
if e.Type == core.EventProgress && e.Tokens > 0 {
ac.setTotalTokens(e.Tokens)
}
ev := wsEvent{
Type: string(e.Type), Round: e.Round, Thought: e.Thought,
ToolName: e.ToolName, ToolArgs: e.ToolArgs, Risk: e.Risk,
Output: e.Output, Success: e.Success, Duration: e.Duration,
Tokens: e.Tokens, Answer: e.Answer, Message: e.Message,
}
if ac.session != nil {
ev.SessionID = ac.session.ID
}
ac.write(ev)
// 持久化(progress / thought_delta 太频繁不入库) // Logan.Li
if ac.session == nil {
return
}
sid := ac.session.ID
switch e.Type {
case core.EventThought:
ac.persistMsg(sid, e.Round, "thought", "assistant", e.Thought, "", "", "", "", false, "", 0)
case core.EventToolCall:
argsJSON, _ := json.Marshal(e.ToolArgs)
ac.persistMsg(sid, e.Round, "tool_call", "tool", "", e.ToolName, string(argsJSON), "", e.Risk, false, "", 0)
case core.EventToolResult:
ac.persistMsg(sid, e.Round, "tool_result", "tool", e.Output, e.ToolName, "", "", "", e.Success, e.Duration, 0)
case core.EventDone:
ac.setTotalTokens(e.Tokens)
ac.persistMsg(sid, e.Round, "done", "assistant", e.Answer, "", "", "", "", true, "", e.Tokens)
case core.EventError:
ac.persistMsg(sid, e.Round, "error", "assistant", e.Message, "", "", "", "", false, "", 0)
}
}
// confirmFn 高危命令确认回调:推 confirm 事件并阻塞等前端回传 // Logan.Li
func (ac *agentConn) confirmFn(toolName string, args map[string]interface{}, risk tools.RiskLevel) (bool, error) {
confirmID := fmt.Sprintf("cf-%d", time.Now().UnixNano())
cmd, _ := args["command"].(string)
ac.write(wsEvent{Type: "confirm", ConfirmID: confirmID, ToolName: toolName, Command: cmd, Risk: risk.Display(), ToolArgs: args})
ch := make(chan bool, 1)
ac.cfMu.Lock()
ac.confirms[confirmID] = ch
ac.cfMu.Unlock()
select {
case approved := <-ch:
return approved, nil
case <-ac.taskCtx.Done():
ac.delConfirm(confirmID)
return false, ac.taskCtx.Err()
case <-time.After(5 * time.Minute):
ac.delConfirm(confirmID)
return false, fmt.Errorf("确认超时")
}
}
// resolveConfirm 前端回传确认结果 // Logan.Li
func (ac *agentConn) resolveConfirm(confirmID string, approved bool) {
ac.cfMu.Lock()
ch, ok := ac.confirms[confirmID]
if ok {
delete(ac.confirms, confirmID)
}
ac.cfMu.Unlock()
if ok {
ch <- approved
}
}
func (ac *agentConn) delConfirm(id string) {
ac.cfMu.Lock()
delete(ac.confirms, id)
ac.cfMu.Unlock()
}
// abort 中断当前任务并唤醒所有等待中的确认 // Logan.Li
func (ac *agentConn) abort() {
if ac.cancel != nil {
ac.cancel()
ac.cancel = nil
}
ac.cfMu.Lock()
for id, ch := range ac.confirms {
select {
case ch <- false:
default:
}
delete(ac.confirms, id)
}
ac.cfMu.Unlock()
}
// lastRound 估算已完成轮次(从 DB 取该会话最大 round) // Logan.Li
func (ac *agentConn) lastRound() int {
if ac.session == nil {
return 0
}
var maxRound int
ac.db.Model(&model.AgentCLIMessage{}).Where("session_id = ?", ac.session.ID).
Select("COALESCE(MAX(round),0)").Scan(&maxRound)
return maxRound
}
// finish 结束本轮执行并更新会话 DB // Logan.Li
func (ac *agentConn) finish(sess *model.AgentCLISession, status, errMsg string, rounds int, dur time.Duration) {
ac.setRunning(false)
tokens := ac.getTotalTokens()
if status == "done" || status == "aborted" || status == "error" {
ac.write(wsEvent{Type: status, SessionID: sess.ID, Message: errMsg, Round: rounds, Tokens: tokens})
}
updates := map[string]interface{}{
"status": status, "rounds": rounds, "duration": dur.Milliseconds(),
"total_tokens": tokens,
}
if errMsg != "" {
updates["error"] = errMsg
}
ac.db.Model(sess).Updates(updates)
}
// write 线程安全写 WS 消息 // Logan.Li
func (ac *agentConn) write(ev wsEvent) {
ac.writeMu.Lock()
defer ac.writeMu.Unlock()
b, _ := json.Marshal(ev)
_ = ac.conn.WriteMessage(websocket.TextMessage, b)
}
// persistMsg 持久化一条消息 // Logan.Li
func (ac *agentConn) persistMsg(sid uint, round int, typ, role, content, toolName, args, output, risk string, success bool, dur string, tokens int) {
msg := &model.AgentCLIMessage{
SessionID: sid, Round: round, Type: typ, Role: role, Content: content,
ToolName: toolName, ToolArgs: args, ToolOutput: output, Risk: risk,
Success: success, Duration: dur, Tokens: tokens,
}
if err := ac.db.Create(msg).Error; err != nil {
// 持久化失败不阻断主流程
_ = err
}
}