feat: add agent start code

This commit is contained in:
engigu
2025-12-28 11:42:58 +08:00
parent 6d7997c92a
commit 78c3707cce
19 changed files with 1491 additions and 501 deletions
+154 -81
View File
@@ -12,6 +12,8 @@ import (
"path/filepath"
"strings"
"time"
"gorm.io/gorm"
)
// AgentService Agent 服务
@@ -29,75 +31,171 @@ func generateToken() string {
return hex.EncodeToString(bytes)
}
// Register Agent 注册(进入待审核状态
func (s *AgentService) Register(req *models.AgentRegisterRequest, ip string) (*models.Agent, error) {
// 检查是否已存在同名待审核的 Agent
var existing models.Agent
if err := database.DB.Where("name = ? AND status = ?", req.Name, "pending").First(&existing).Error; err == nil {
// 更新现有记录
now := models.LocalTime(time.Now())
database.DB.Model(&existing).Updates(map[string]interface{}{
"hostname": req.Hostname,
"version": req.Version,
"ip": ip,
"last_seen": now,
})
return &existing, nil
// generateRegCode 生成令牌(64位,与认证 Token 相同
func generateRegCode() string {
bytes := make([]byte, 32)
rand.Read(bytes)
return hex.EncodeToString(bytes)
}
// ========== 注册码管理 ==========
// CreateRegCode 创建令牌(同时创建 Agent 记录)
func (s *AgentService) CreateRegCode(remark string, maxUses int, expiresAt *time.Time) (*models.AgentRegCode, error) {
var expires *models.LocalTime
if expiresAt != nil {
t := models.LocalTime(*expiresAt)
expires = &t
}
// 创建新的待审核 Agent
token := generateRegCode()
regCode := &models.AgentRegCode{
Code: token,
Remark: remark,
MaxUses: maxUses,
ExpiresAt: expires,
Enabled: true,
}
if err := database.DB.Create(regCode).Error; err != nil {
return nil, err
}
logger.Infof("[Agent] 创建令牌: %s (max_uses=%d)", token[:8]+"...", maxUses)
return regCode, nil
}
// ListRegCodes 获取注册码列表
func (s *AgentService) ListRegCodes() []models.AgentRegCode {
var codes []models.AgentRegCode
database.DB.Order("id DESC").Find(&codes)
return codes
}
// DeleteRegCode 删除注册码
func (s *AgentService) DeleteRegCode(id uint) error {
return database.DB.Delete(&models.AgentRegCode{}, id).Error
}
// ValidateRegCode 验证注册码
func (s *AgentService) ValidateRegCode(code string) (*models.AgentRegCode, error) {
var regCode models.AgentRegCode
if err := database.DB.Where("code = ?", code).First(&regCode).Error; err != nil {
return nil, &ServiceError{Message: "无效的注册码"}
}
if !regCode.Enabled {
return nil, &ServiceError{Message: "注册码已禁用"}
}
// 检查使用次数
if regCode.MaxUses > 0 && regCode.UsedCount >= regCode.MaxUses {
return nil, &ServiceError{Message: "注册码已达到使用上限"}
}
// 检查过期时间
if regCode.ExpiresAt != nil && time.Time(*regCode.ExpiresAt).Before(time.Now()) {
return nil, &ServiceError{Message: "注册码已过期"}
}
return &regCode, nil
}
// UseRegCode 使用注册码(增加使用计数)
func (s *AgentService) UseRegCode(id uint) {
database.DB.Model(&models.AgentRegCode{}).Where("id = ?", id).UpdateColumn("used_count", gorm.Expr("used_count + 1"))
}
// ========== Agent 注册 ==========
// RegisterByToken 通过令牌注册 Agent(首次 WebSocket 连接时调用)
// 返回: agent, isNewAgent, error
func (s *AgentService) RegisterByToken(token string, machineID string, ip string) (*models.Agent, bool, error) {
// 验证令牌
regCode, err := s.ValidateRegCode(token)
if err != nil {
return nil, false, err
}
// 如果提供了 machine_id,先检查是否已存在
if machineID != "" {
var existing models.Agent
if err := database.DB.Where("machine_id = ?", machineID).First(&existing).Error; err == nil {
// 已存在,更新 token 和状态,复用已有 Agent
now := models.LocalTime(time.Now())
database.DB.Model(&existing).Updates(map[string]interface{}{
"token": token,
"ip": ip,
"status": "online",
"last_seen": now,
})
s.UseRegCode(regCode.ID)
logger.Infof("[Agent] Agent #%d 通过 machine_id 复用 (%s)", existing.ID, machineID[:8]+"...")
return &existing, false, nil
}
}
// 创建 Agent,使用令牌作为认证 Token
now := models.LocalTime(time.Now())
agent := &models.Agent{
Name: req.Name,
Hostname: req.Hostname,
Version: req.Version,
IP: ip,
Status: "pending",
LastSeen: &now,
Enabled: true,
Name: fmt.Sprintf("agent-%d", time.Now().Unix()),
Token: token,
MachineID: machineID,
IP: ip,
Status: "online",
LastSeen: &now,
Enabled: true,
}
if err := database.DB.Create(agent).Error; err != nil {
return nil, err
return nil, false, err
}
logger.Infof("[Agent] 新 Agent 注册: %s (%s)", req.Name, ip)
return agent, nil
s.UseRegCode(regCode.ID)
logger.Infof("[Agent] Agent 通过令牌注册: #%d (%s)", agent.ID, ip)
return agent, true, nil
}
// Approve 审核通过 Agent,生成 Token
func (s *AgentService) Approve(id uint) (*models.Agent, error) {
agent := s.GetByID(id)
if agent == nil {
return nil, &ServiceError{Message: "Agent 不存在"}
// Register Agent 注册(必须使用令牌)- 保留兼容旧版本
func (s *AgentService) Register(req *models.AgentRegisterRequest, ip string) (*models.Agent, string, error) {
// 必须提供令牌
if req.Token == "" {
return nil, "", &ServiceError{Message: "缺少注册令牌"}
}
if agent.Status != "pending" {
return nil, &ServiceError{Message: "Agent 状态不是待审核"}
regCode, err := s.ValidateRegCode(req.Token)
if err != nil {
return nil, "", err
}
token := generateToken()
// 检查是否已存在同名 Agent
var existing models.Agent
if err := database.DB.Where("name = ?", req.Name).First(&existing).Error; err == nil {
return nil, "", &ServiceError{Message: "Agent 名称已存在"}
}
// 创建新 Agent,使用令牌作为认证 Token
now := models.LocalTime(time.Now())
if err := database.DB.Model(agent).Updates(map[string]interface{}{
"token": token,
"status": "online",
"last_seen": now,
}).Error; err != nil {
return nil, err
agent := &models.Agent{
Name: req.Name,
Token: req.Token,
Hostname: req.Hostname,
Version: req.Version,
BuildTime: req.BuildTime,
IP: ip,
Status: "online",
LastSeen: &now,
Enabled: true,
}
agent.Token = token
agent.Status = "online"
agent.LastSeen = &now
if err := database.DB.Create(agent).Error; err != nil {
return nil, "", err
}
logger.Infof("[Agent] Agent 已审核通过: %s (#%d)", agent.Name, agent.ID)
return agent, nil
}
// Reject 拒绝 Agent
func (s *AgentService) Reject(id uint) error {
return database.DB.Delete(&models.Agent{}, id).Error
s.UseRegCode(regCode.ID)
logger.Infof("[Agent] Agent 注册成功: %s (%s)", req.Name, ip)
return agent, req.Token, nil
}
// Update 更新 Agent
@@ -109,7 +207,7 @@ func (s *AgentService) Update(id uint, name, description string, enabled bool) e
}).Error
}
// Delete 删除 Agent
// Delete 删除 Agent(物理删除)
func (s *AgentService) Delete(id uint) error {
// 检查是否有关联任务
var count int64
@@ -118,7 +216,7 @@ func (s *AgentService) Delete(id uint) error {
return &ServiceError{Message: "该 Agent 下还有关联任务,无法删除"}
}
return database.DB.Delete(&models.Agent{}, id).Error
return database.DB.Unscoped().Delete(&models.Agent{}, id).Error
}
// GetByID 根据 ID 获取 Agent
@@ -139,27 +237,16 @@ func (s *AgentService) GetByToken(token string) *models.Agent {
return &agent
}
// List 获取已审核的 Agent 列表
// List 获取 Agent 列表
func (s *AgentService) List() []models.Agent {
var agents []models.Agent
database.DB.Where("status != ?", "pending").Order("id DESC").Find(&agents)
database.DB.Order("id DESC").Find(&agents)
return agents
}
// ListPending 获取待审核的 Agent 列表
func (s *AgentService) ListPending() []models.Agent {
var agents []models.Agent
database.DB.Where("status = ?", "pending").Order("id DESC").Find(&agents)
return agents
}
// RegenerateToken 重新生成 Token
// RegenerateToken 重新生成 Token - 已废弃,保留空实现避免路由错误
func (s *AgentService) RegenerateToken(id uint) (string, error) {
newToken := generateToken()
if err := database.DB.Model(&models.Agent{}).Where("id = ?", id).Update("token", newToken).Error; err != nil {
return "", err
}
return newToken, nil
return "", &ServiceError{Message: "此功能已禁用"}
}
// Heartbeat Agent 心跳
@@ -209,20 +296,6 @@ func (s *AgentService) Heartbeat(token, ip, version, buildTime, hostname, osType
return agent, nil
}
// CheckPendingAgent 检查待审核 Agent 的状态(用于 Agent 轮询)
func (s *AgentService) CheckPendingAgent(name, ip string) (*models.Agent, error) {
var agent models.Agent
if err := database.DB.Where("name = ? AND ip = ?", name, ip).First(&agent).Error; err != nil {
return nil, &ServiceError{Message: "Agent 未注册"}
}
// 更新最后心跳时间
now := models.LocalTime(time.Now())
database.DB.Model(&agent).Update("last_seen", now)
return &agent, nil
}
// GetTasks 获取 Agent 的任务列表
func (s *AgentService) GetTasks(agentID uint) []models.AgentTask {
var tasks []models.Task
+293
View File
@@ -0,0 +1,293 @@
package services
import (
"baihu/internal/database"
"baihu/internal/logger"
"baihu/internal/models"
"encoding/json"
"sync"
"time"
"github.com/gorilla/websocket"
)
// AgentWSManager WebSocket 连接管理器
type AgentWSManager struct {
connections map[uint]*AgentConnection // agentID -> connection
ipConnections map[string]int // IP -> 连接数
ipLastAttempt map[string]time.Time // IP -> 最后连接尝试时间
ipFailCount map[string]int // IP -> 连续失败次数
mu sync.RWMutex
}
// 限流配置
const (
maxConnectionsPerIP = 10 // 每个 IP 最大连接数
minConnectInterval = 5 * time.Second // 同一 IP 最小连接间隔
maxFailCount = 5 // 最大连续失败次数
failBlockDuration = 5 * time.Minute // 失败后封禁时长
)
// AgentConnection Agent WebSocket 连接
type AgentConnection struct {
AgentID uint
IP string
Conn *websocket.Conn
Send chan []byte
LastPing time.Time
mu sync.Mutex
}
// WSMessage WebSocket 消息结构
type WSMessage struct {
Type string `json:"type"`
Data json.RawMessage `json:"data,omitempty"`
}
// 消息类型常量
const (
WSTypeHeartbeat = "heartbeat"
WSTypeHeartbeatAck = "heartbeat_ack"
WSTypeTasks = "tasks"
WSTypeTaskResult = "task_result"
WSTypeUpdate = "update"
WSTypeDisconnect = "disconnect"
WSTypeConnected = "connected" // 连接成功,包含注册状态
WSTypeDisabled = "disabled" // Agent 被禁用
WSTypeEnabled = "enabled" // Agent 被启用
WSTypeFetchTasks = "fetch_tasks" // Agent 请求任务列表
)
var agentWSManager *AgentWSManager
var agentWSOnce sync.Once
// GetAgentWSManager 获取单例
func GetAgentWSManager() *AgentWSManager {
agentWSOnce.Do(func() {
agentWSManager = &AgentWSManager{
connections: make(map[uint]*AgentConnection),
ipConnections: make(map[string]int),
ipLastAttempt: make(map[string]time.Time),
ipFailCount: make(map[string]int),
}
go agentWSManager.cleanupLoop()
})
return agentWSManager
}
// CheckRateLimit 检查 IP 限流,返回是否允许连接
func (m *AgentWSManager) CheckRateLimit(ip string) (bool, string) {
m.mu.Lock()
defer m.mu.Unlock()
now := time.Now()
// 检查是否被封禁(连续失败过多)
if failCount, exists := m.ipFailCount[ip]; exists && failCount >= maxFailCount {
if lastAttempt, ok := m.ipLastAttempt[ip]; ok {
if now.Sub(lastAttempt) < failBlockDuration {
remaining := failBlockDuration - now.Sub(lastAttempt)
return false, "连接失败次数过多,请 " + remaining.Round(time.Second).String() + " 后重试"
}
// 封禁时间已过,重置计数
delete(m.ipFailCount, ip)
}
}
// 检查连接频率
if lastAttempt, exists := m.ipLastAttempt[ip]; exists {
if now.Sub(lastAttempt) < minConnectInterval {
return false, "连接过于频繁,请稍后重试"
}
}
// 检查 IP 连接数
if count, exists := m.ipConnections[ip]; exists && count >= maxConnectionsPerIP {
return false, "该 IP 连接数已达上限"
}
m.ipLastAttempt[ip] = now
return true, ""
}
// RecordConnectFail 记录连接失败
func (m *AgentWSManager) RecordConnectFail(ip string) {
m.mu.Lock()
defer m.mu.Unlock()
m.ipFailCount[ip]++
m.ipLastAttempt[ip] = time.Now()
if m.ipFailCount[ip] >= maxFailCount {
logger.Warnf("[AgentWS] IP %s 连续失败 %d 次,已封禁 %v", ip, m.ipFailCount[ip], failBlockDuration)
}
}
// RecordConnectSuccess 记录连接成功,重置失败计数
func (m *AgentWSManager) RecordConnectSuccess(ip string) {
m.mu.Lock()
defer m.mu.Unlock()
delete(m.ipFailCount, ip)
}
// Register 注册连接
func (m *AgentWSManager) Register(agentID uint, conn *websocket.Conn, ip string) *AgentConnection {
m.mu.Lock()
defer m.mu.Unlock()
// 关闭旧连接
if old, exists := m.connections[agentID]; exists {
// 减少旧 IP 的连接计数
if old.IP != "" {
if count, ok := m.ipConnections[old.IP]; ok && count > 0 {
m.ipConnections[old.IP] = count - 1
}
}
old.Close()
}
ac := &AgentConnection{
AgentID: agentID,
IP: ip,
Conn: conn,
Send: make(chan []byte, 256),
LastPing: time.Now(),
}
m.connections[agentID] = ac
// 增加 IP 连接计数
m.ipConnections[ip]++
logger.Infof("[AgentWS] Agent #%d 已连接 (%s)", agentID, ip)
return ac
}
// Unregister 注销连接
func (m *AgentWSManager) Unregister(agentID uint) {
m.mu.Lock()
defer m.mu.Unlock()
if conn, exists := m.connections[agentID]; exists {
// 减少 IP 连接计数
if conn.IP != "" {
if count, ok := m.ipConnections[conn.IP]; ok && count > 0 {
m.ipConnections[conn.IP] = count - 1
}
}
conn.Close()
delete(m.connections, agentID)
logger.Infof("[AgentWS] Agent #%d 已断开", agentID)
}
}
// GetConnection 获取连接
func (m *AgentWSManager) GetConnection(agentID uint) *AgentConnection {
m.mu.RLock()
defer m.mu.RUnlock()
return m.connections[agentID]
}
// SendToAgent 发送消息给指定 Agent
func (m *AgentWSManager) SendToAgent(agentID uint, msgType string, data interface{}) error {
conn := m.GetConnection(agentID)
if conn == nil {
return nil // Agent 不在线
}
dataBytes, _ := json.Marshal(data)
msg := WSMessage{Type: msgType, Data: dataBytes}
msgBytes, _ := json.Marshal(msg)
select {
case conn.Send <- msgBytes:
return nil
default:
return nil // 缓冲区满,丢弃
}
}
// BroadcastTasks 广播任务更新给指定 Agent
func (m *AgentWSManager) BroadcastTasks(agentID uint) {
agentService := NewAgentService()
tasks := agentService.GetTasks(agentID)
m.SendToAgent(agentID, WSTypeTasks, map[string]interface{}{
"tasks": tasks,
})
}
// OnlineCount 在线 Agent 数量
func (m *AgentWSManager) OnlineCount() int {
m.mu.RLock()
defer m.mu.RUnlock()
return len(m.connections)
}
// cleanupLoop 清理超时连接
func (m *AgentWSManager) cleanupLoop() {
ticker := time.NewTicker(30 * time.Second)
defer ticker.Stop()
for range ticker.C {
m.mu.Lock()
now := time.Now()
// 清理超时连接
for agentID, conn := range m.connections {
if now.Sub(conn.LastPing) > 2*time.Minute {
// 减少 IP 连接计数
if conn.IP != "" {
if count, ok := m.ipConnections[conn.IP]; ok && count > 0 {
m.ipConnections[conn.IP] = count - 1
}
}
conn.Close()
delete(m.connections, agentID)
// 更新数据库状态
database.DB.Model(&models.Agent{}).Where("id = ?", agentID).Update("status", "offline")
logger.Infof("[AgentWS] Agent #%d 心跳超时,已断开", agentID)
}
}
// 清理过期的限流记录(超过 10 分钟未活动)
for ip, lastAttempt := range m.ipLastAttempt {
if now.Sub(lastAttempt) > 10*time.Minute {
delete(m.ipLastAttempt, ip)
delete(m.ipFailCount, ip)
// 只清理没有活跃连接的 IP 计数
if m.ipConnections[ip] == 0 {
delete(m.ipConnections, ip)
}
}
}
m.mu.Unlock()
}
}
// Close 关闭连接
func (c *AgentConnection) Close() {
c.mu.Lock()
defer c.mu.Unlock()
if c.Conn != nil {
c.Conn.Close()
c.Conn = nil
}
if c.Send != nil {
close(c.Send)
c.Send = nil
}
}
// WriteMessage 写入消息
func (c *AgentConnection) WriteMessage(data []byte) error {
c.mu.Lock()
defer c.mu.Unlock()
if c.Conn == nil {
return nil
}
c.Conn.SetWriteDeadline(time.Now().Add(10 * time.Second))
return c.Conn.WriteMessage(websocket.TextMessage, data)
}
// UpdatePing 更新心跳时间
func (c *AgentConnection) UpdatePing() {
c.LastPing = time.Now()
}
+4 -2
View File
@@ -11,7 +11,7 @@ func NewTaskService() *TaskService {
return &TaskService{}
}
func (ts *TaskService) CreateTask(name, command, schedule string, timeout int, workDir, cleanConfig, envs, taskType, config string) *models.Task {
func (ts *TaskService) CreateTask(name, command, schedule string, timeout int, workDir, cleanConfig, envs, taskType, config string, agentID *uint) *models.Task {
if taskType == "" {
taskType = "task"
}
@@ -25,6 +25,7 @@ func (ts *TaskService) CreateTask(name, command, schedule string, timeout int, w
WorkDir: workDir,
CleanConfig: cleanConfig,
Envs: envs,
AgentID: agentID,
Enabled: true,
}
database.DB.Create(task)
@@ -61,7 +62,7 @@ func (ts *TaskService) GetTaskByID(id int) *models.Task {
return &task
}
func (ts *TaskService) UpdateTask(id int, name, command, schedule string, timeout int, workDir, cleanConfig, envs string, enabled bool, taskType, config string) *models.Task {
func (ts *TaskService) UpdateTask(id int, name, command, schedule string, timeout int, workDir, cleanConfig, envs string, enabled bool, taskType, config string, agentID *uint) *models.Task {
var task models.Task
if err := database.DB.First(&task, id).Error; err != nil {
return nil
@@ -74,6 +75,7 @@ func (ts *TaskService) UpdateTask(id int, name, command, schedule string, timeou
task.CleanConfig = cleanConfig
task.Envs = envs
task.Enabled = enabled
task.AgentID = agentID
if taskType != "" {
task.Type = taskType
}