进程检测能力调整

internal/process/manager.go

@@ -399,7 +399,6 @@ func (pm *ProcessManager) isFRPCProcess(pid int) (bool, string) {
 // 进程存活检测
 // ================================================================
 
-// isProcessAlive 检查 PID 对应的进程是否存活
 func (pm *ProcessManager) isProcessAlive(pid int) bool {
 	if pid <= 0 {
 		return false
@@ -421,6 +420,50 @@ func (pm *ProcessManager) isProcessAlive(pid int) bool {
 	return process.Signal(syscall.Signal(0)) == nil
 }
 
+// ================================================================
+// 进程清理
+// ================================================================
+
+func (pm *ProcessManager) killProcess(pid int) error {
+	if pid <= 0 {
+		return nil
+	}
+
+	if runtime.GOOS == "windows" {
+		cmd := exec.Command("taskkill", "/F", "/PID", strconv.Itoa(pid))
+		return cmd.Run()
+	}
+
+	process, err := os.FindProcess(pid)
+	if err != nil {
+		return err
+	}
+	return process.Kill()
+}
+
+func (pm *ProcessManager) cleanupOrphans() {
+	// 检查端口是否被占用
+	occupied, err := pm.CheckPort()
+	if err != nil || !occupied {
+		return
+	}
+
+	// 检查是否有 PID 文件
+	pid := pm.readPIDFile()
+	if pid > 0 && pm.isProcessAlive(pid) {
+		return
+	}
+
+	log.Printf("[WARN] 检测到孤儿 frpc 进程 (端口 %d 被占用但无有效 PID)，正在清理...", pm.adminPort)
+	if runtime.GOOS == "windows" {
+		exec.Command("taskkill", "/F", "/IM", "frpc.exe").Run()
+	} else {
+		exec.Command("pkill", "-f", "frpc").Run()
+	}
+	pm.deletePIDFile()
+	time.Sleep(1 * time.Second)
+}
+
 // ================================================================
 // PID 文件操作
 // ================================================================
@@ -532,6 +575,10 @@ func (pm *ProcessManager) Start(ctx context.Context) error {
 		return fmt.Errorf("获取锁失败: %w", err)
 	}
 	defer pm.Unlock()
+
+	// 清理孤儿进程
+	pm.cleanupOrphans()
+
 	return pm.startLocked(ctx)
 }
 
@@ -557,45 +604,28 @@ func (pm *ProcessManager) startLocked(ctx context.Context) error {
 		return fmt.Errorf("启动 frpc 失败: %w", err)
 	}
 
-	// ============================================================
-	// 组合状态检测：最多等待 5 秒，三条件全部满足才算成功
-	// ============================================================
+	// 立即写入 PID 文件
+	if err := pm.writePIDFile(cmd.Process.Pid); err != nil {
+		log.Printf("[WARN] 写入 PID 文件失败: %v", err)
+	}
+	log.Printf("[DEBUG] frpc 进程已启动，PID: %d", cmd.Process.Pid)
+
+	// 检测端口是否监听（最多 5 秒）
 	for attempt := 0; attempt < StartupMaxAttempts; attempt++ {
-		// 条件1: 检查 PID 文件
-		pid := pm.readPIDFile()
-		if pid <= 0 {
-			time.Sleep(StartupRetryDelay)
-			continue
-		}
-
-		// 条件2: 检查进程是否存活
-		if !pm.isProcessAlive(pid) {
-			time.Sleep(StartupRetryDelay)
-			continue
-		}
-
-		// 条件3: 检查端口是否可连接
 		occupied, err := pm.CheckPort()
 		if err == nil && occupied {
-			// 三项全部满足！
-			pm.writePIDFile(pid)
 			log.Printf("[INFO] frpc 启动成功 (PID: %d, 端口: %d)，耗时 %dms",
-				pid, pm.adminPort, attempt*int(StartupRetryDelay/time.Millisecond))
+				cmd.Process.Pid, pm.adminPort, attempt*int(StartupRetryDelay/time.Millisecond))
 			return nil
 		}
-
 		time.Sleep(StartupRetryDelay)
 	}
 
-	// 超时：输出诊断信息
-	pid := pm.readPIDFile()
-	if pid > 0 {
-		alive := pm.isProcessAlive(pid)
-		occupied, _ := pm.CheckPort()
-		return fmt.Errorf("frpc 启动超时: PID=%d, 进程存活=%v, 端口监听=%v", pid, alive, occupied)
-	}
-
-	return fmt.Errorf("frpc 启动超时: 未生成 PID 文件")
+	// 端口监听超时：杀掉启动的进程并清理
+	log.Printf("[WARN] frpc 启动超时 (PID: %d)，正在清理...", cmd.Process.Pid)
+	pm.killProcess(cmd.Process.Pid)
+	pm.deletePIDFile()
+	return fmt.Errorf("frpc 启动超时: 端口未监听 (PID: %d)", cmd.Process.Pid)
 }
 
 func (pm *ProcessManager) Stop(ctx context.Context) error {
@@ -652,8 +682,7 @@ func (pm *ProcessManager) stopLocked(_ context.Context) error {
 
 	// 端口未释放, 强制 kill
 	if runtime.GOOS == "windows" {
-		cmd := exec.Command("taskkill", "/F", "/IM", "frpc.exe")
-		cmd.Run()
+		exec.Command("taskkill", "/F", "/IM", "frpc.exe").Run()
 	} else {
 		proc.Kill()
 	}