物联网软件

iot/EdgeAgent/updater/main.go

@@ -1,132 +1,189 @@
 package main
 
 import (
-	"flag"
 	"fmt"
 	"io"
 	"net/http"
-	"net/url"
 	"os"
 	"path/filepath"
+	"sentinel/pkg/config"
 	"sentinel/pkg/device"
 	"sentinel/pkg/log"
 	"sentinel/pkg/net"
-	"strconv"
+	"sentinel/pkg/platform"
 	"time"
 )
 
 func main() {
-	// 定义命令行参数
-	version := flag.String("version", "", "current version of main program")
+	log.Println("[daemon] 设备ID:", device.GetDeviceID())
 
-	flag.Parse()
+	log.Init(config.LOG_FILE_DIC)
 
-	if *version == "" {
-		// updater 视角：-1 表示“未知版本”，一定触发更新检测
-		*version = "0"
-		log.Println("[updater] --version not provided, fallback to -1")
-		fmt.Println("[updater] 主程序版本号:", *version)
-	}
+	starter := newMainProgramStarter()
+	// 启动主程序
+	initMainProcess(starter)
+	ticker := time.NewTicker(config.DAEMON_UPDATE_CHECK_INTERVAL_SECONDS * time.Second)
+	defer ticker.Stop()
 
-	deviceID := device.GetDeviceID()
-	fmt.Printf("[updater] 当前设备id: %s\n", deviceID)
-	versionInt, err := strconv.Atoi(*version)
-	if err != nil {
-		log.Println("[updater] invalid version:", *version)
-		versionInt = 0
-	}
-	if err := RunUpdate(deviceID, versionInt); err != nil {
-		log.Fatalf("[updater] 更新失败: %v", err)
-	}
-	fmt.Println("[updater] 更新程序结束")
+	go func() {
+		for range ticker.C {
+			//log.Println("[daemon] 定时检查主程序运行状态")
+			handleProcessCheck(starter)
+
+			cfg := config.LoadConfig()
+			if cfg.NeedUpdate {
+				log.Println("[daemon] 配置文件标记需要更新，触发更新")
+				cfg.NeedUpdate = false
+				if err := config.WriteLocalConfig(cfg); err != nil {
+					log.Println("[daemon] 写回配置文件失败:", err)
+				}
+				handleUpdate(starter)
+			}
+		}
+	}()
+	select {}
 }
-func RunUpdate(deviceID string, version int) error {
-	// 1. 检查更新
-	info, err := api.CheckUpdate(deviceID)
-	if err != nil {
-		fmt.Println("[updater] 请求错误，请检查网络")
-		return err
-	}
-	fmt.Println("[updater] 新版本：", info.Version)
-	fmt.Println("[updater] 新内容：", info.Notes)
-	fmt.Println("[updater] 下载地址：", info.DownloadURL)
 
+func initMainProcess(starter platform.MainProgramStarter) {
+	if err := startMainProgram(starter); err != nil {
+		log.Println("[daemon] 启动失败，回退到更新:", err)
+		handleUpdate(starter)
+	}
+}
+
+var lastState string
+var lastPID int
+
+func handleProcessCheck(starter platform.MainProgramStarter) {
+	cfg := config.LoadConfig()
+
+	// 1. PID 是否存在
+	running, err := starter.IsProcessRunning(cfg.PID)
+	if cfg.PID <= 0 || err != nil || !running {
+		if lastState != "NOT_RUNNING" {
+			log.Println("[daemon] 主程序未运行，尝试启动")
+			lastState = "NOT_RUNNING"
+			lastPID = 0
+		}
+		restartMain(starter)
+		return
+	}
+
+	// 2. 语义健康判断
+	switch cfg.ControlState {
+	case "DEGRADED":
+		if lastState != "DEGRADED" {
+			log.Println("[daemon] 主程序存在错误，等待主程序自行处理")
+			lastState = "DEGRADED"
+			lastPID = cfg.PID
+		}
+		return
+	case "CONTROL_LOST":
+		if lastState != "CONTROL_LOST" {
+			log.Println("[daemon] 主程序心跳超时，已失控，强制重启")
+			lastState = "CONTROL_LOST"
+			lastPID = cfg.PID
+		}
+		restartMain(starter)
+		return
+	default:
+		// 3. 健康
+		if lastState != "HEALTHY" || lastPID != cfg.PID {
+			log.Println("[daemon] 主程序健康运行，PID:", cfg.PID)
+			lastState = "HEALTHY"
+			lastPID = cfg.PID
+		}
+	}
+}
+
+func restartMain(starter platform.MainProgramStarter) {
+	cfg := config.LoadConfig()
+	if cfg.PID > 0 {
+		_ = starter.KillProcess(cfg.PID)
+		time.Sleep(300 * time.Millisecond)
+	}
+	_ = startMainProgram(starter)
+}
+
+func handleUpdate(starter platform.MainProgramStarter) {
+	cfg := config.LoadConfig()
+	info, err := api.CheckUpdate(device.GetDeviceID())
+	if err != nil {
+		log.Println("[daemon] 检查更新失败:", err)
+		return
+	}
+
+	if info.Version <= cfg.VersionCode {
+		log.Println("[daemon] 没有新版本，跳过")
+		return
+	}
+
+	log.Println("[daemon] 发现新版本 %d\n", info.Version)
+
+	// 停止主程序
+	if cfg.PID > 0 {
+		_ = starter.KillProcess(cfg.PID)
+		time.Sleep(500 * time.Millisecond)
+	}
+
+	if err := downloadAndReplace(starter, info.DownloadURL); err != nil {
+		log.Println("[daemon] 更新失败:", err)
+		return
+	}
+
+	if err := startMainProgram(starter); err != nil {
+		log.Println("[daemon] 更新后启动失败:", err)
+		return
+	}
+
+	log.Println("[daemon] 更新完成")
+}
+
+func startMainProgram(starter platform.MainProgramStarter) error {
+	selfPath, _ := os.Executable()
+	selfDir := filepath.Dir(selfPath)
+	targetExe := filepath.Join(selfDir, starter.GetMainName())
+
+	if _, err := os.Stat(targetExe); os.IsNotExist(err) {
+		log.Println("[daemon] 主程序文件不存在，触发更新")
+		handleUpdate(starter)
+		return nil // 先不启动，等更新完再启动
+	}
+
+	err := starter.Start(targetExe)
+	return err
+}
+
+func downloadAndReplace(
+	starter platform.MainProgramStarter,
+	downloadURL string,
+) error {
 	// 获取主程序路径
 	selfPath, err := os.Executable()
 	if err != nil {
-		return err
+		return fmt.Errorf("获取自身路径失败: %w", err)
 	}
 	selfDir := filepath.Dir(selfPath)
-
-	starter := newMainProgramStarter()
 	targetExe := filepath.Join(selfDir, starter.GetMainName())
-
-	// 2. 对比版本号，没有新版本则直接启动原程序
-	if info.Version <= version {
-		fmt.Println("[updater] 暂未发现新版本，启动原程序")
-		if err := starter.Start(targetExe); err != nil {
-			return err
-		}
-		os.Exit(0)
-	}
-
-	// 3. 有新版本则先备份到 ./tmp/old_app/
-	backupDir := filepath.Join(selfDir, "tmp", "old_app")
-	_ = os.MkdirAll(backupDir, 0755)
-	backupFile := filepath.Join(backupDir, "main_"+strconv.Itoa(version)+".bak")
-	if err := os.Rename(targetExe, backupFile); err != nil {
-		fmt.Println("[updater] 备份失败，但继续更新:", err)
-	}
-
-	// 4. 下载新版本到 ./tmp
-	tmpDir := filepath.Join(selfDir, "tmp")
-	_ = os.MkdirAll(tmpDir, 0755)
-
-	u, err := url.Parse(info.DownloadURL)
+	outFile, err := os.Create(targetExe)
 	if err != nil {
-		return err
+		return fmt.Errorf("创建目标文件失败: %w", err)
 	}
-	base := filepath.Base(u.Path)
-	ext := filepath.Ext(base)
-
-	tmpFile, err := os.CreateTemp(tmpDir, "main_*"+ext)
-	if err != nil {
-		return err
-	}
-	defer tmpFile.Close()
+	defer outFile.Close()
 
 	client := &http.Client{Timeout: 30 * time.Second}
-	resp, err := client.Get(info.DownloadURL)
+	resp, err := client.Get(downloadURL)
 	if err != nil {
-		return err
+		return fmt.Errorf("下载新程序失败: %w", err)
 	}
 	defer resp.Body.Close()
 
-	if _, err := io.Copy(tmpFile, resp.Body); err != nil {
-		return err
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("下载失败, HTTP状态码: %d", resp.StatusCode)
 	}
 
-	// 5. 重命名新文件到 ./main.exe
-	tmpFile.Close() // 关闭临时文件才能重命名
-	maxRetry := 20
-	for i := 0; i < maxRetry; i++ {
-		err := os.Rename(tmpFile.Name(), targetExe)
-		if err == nil {
-			break
-		}
-		fmt.Println("[updater] 文件被占用，等待 500ms 再尝试...")
-		time.Sleep(500 * time.Millisecond)
-		if i == maxRetry-1 {
-			return fmt.Errorf("替换失败: %w", err)
-		}
+	if _, err := io.Copy(outFile, resp.Body); err != nil {
+		return fmt.Errorf("写入目标文件失败: %w", err)
 	}
-
-	// 6. 启动主程序，同时完全退出自己
-	if err := starter.Start(targetExe); err != nil {
-		return err
-	}
-	fmt.Printf("[updater] 更新完成，新程序已启动，退出更新程序")
-	os.Exit(0)
-
 	return nil
 }

iot/EdgeAgent/updater/main_linux.go

@@ -6,6 +6,7 @@ package main
 import (
 	"os"
 	"os/exec"
+	"strconv"
 )
 
 type mainProgramStarter struct{}
@@ -25,3 +26,29 @@ func (s *mainProgramStarter) Start(targetExe string) error {
 	// Linux 下：先保持最简单，保证能跑
 	return cmd.Start()
 }
+
+// 判断进程是否在运行
+func (s *mainProgramStarter) IsProcessRunning(pid int) (bool, error) {
+	if pid <= 0 {
+		return false, nil
+	}
+
+	procPath := "/proc/" + strconv.Itoa(pid)
+	_, err := os.Stat(procPath)
+	if err == nil {
+		return true, nil
+	}
+
+	if os.IsNotExist(err) {
+		return false, nil
+	}
+
+	// 其他错误（例如权限问题）
+	return false, err
+}
+
+// 杀掉进程
+func (l *mainProgramStarter) KillProcess(pid int) error {
+	cmd := exec.Command("kill", "-9", strconv.Itoa(pid))
+	return cmd.Run()
+}

iot/EdgeAgent/updater/main_windows.go

@@ -6,7 +6,10 @@ package main
 import (
 	"os"
 	"os/exec"
+	"strconv"
 	"syscall"
+
+	"golang.org/x/sys/windows"
 )
 
 type mainProgramStarter struct{}
@@ -27,3 +30,34 @@ func (s *mainProgramStarter) Start(targetExe string) error {
 
 	return cmd.Start()
 }
+
+// 判断进程是否在运行
+func (s *mainProgramStarter) IsProcessRunning(pid int) (bool, error) {
+	if pid <= 0 {
+		return false, nil
+	}
+
+	// 打开进程句柄
+	handle, err := windows.OpenProcess(windows.PROCESS_QUERY_LIMITED_INFORMATION, false, uint32(pid))
+	if err != nil {
+		// 权限不足，但进程存在
+		if err == windows.ERROR_ACCESS_DENIED {
+			return true, nil
+		}
+		// 进程不存在
+		if err == windows.ERROR_INVALID_PARAMETER {
+			return false, nil
+		}
+		// 其他错误
+		return false, err
+	}
+	defer windows.CloseHandle(handle)
+
+	return true, nil
+}
+
+// 杀掉进程
+func (w *mainProgramStarter) KillProcess(pid int) error {
+	cmd := exec.Command("taskkill", "/PID", strconv.Itoa(pid), "/F")
+	return cmd.Run()
+}