fix: 设备异常状态清除（清除异常/一键清除，检查与更新原子化防竞态）

core/device_worker.py

@@ -115,6 +115,36 @@ def _remove_worker(serial):
         _WORKERS.pop(serial, None)
 
 
+def clear_worker_error(serial):
+    """清除设备错误/失败状态，恢复为 idle。运行中/连接中返回 False 不清理。
+
+    检查与更新在同一把锁内完成，避免与任务启动写 connecting/running 的竞态。
+    """
+    with _WORKERS_LOCK:
+        w = _WORKERS.get(serial)
+        if w is None:
+            return True  # 本就没有状态记录，视作已清理
+        if w.get("status") in ("running", "connecting"):
+            return False
+        w.update(status="idle", last_error="", last_warning="", task_job="",
+                 current_action="", attempt=0, progress={}, end_time=0,
+                 last_heartbeat=time.time())
+    return True
+
+
+def clear_all_worker_errors(exclude=()):
+    """批量清除所有 error/failed 状态（exclude 里的跳过）。返回清除数。"""
+    exclude = set(exclude)
+    with _WORKERS_LOCK:
+        targets = [s for s, w in _WORKERS.items()
+                   if s not in exclude and w.get("status") in ("error", "failed")]
+    cleared = 0
+    for s in targets:
+        if clear_worker_error(s):
+            cleared += 1
+    return cleared
+
+
 def get_worker_heartbeat(serial):
     """返回 worker 最近一次心跳时间，没有返回 0。"""
     with _WORKERS_LOCK:

static/admin/monitor.js

@@ -130,6 +130,19 @@ async function stopSelectedDevices(){
   setTimeout(loadMonitor,500);
 }
 
+// 清除单台设备的异常状态（error/failed → 未启动）
+async function clearDeviceError(serial){
+  if(!confirm('确认清除设备 '+serial+' 的异常状态？'))return;
+  const r=await apiPost('/api/device/clear_error',{serial:serial});
+  handleResult(r,()=>loadMonitor());
+}
+// 一键清除所有异常/失败设备的错误状态（运行中的任务不受影响）
+async function clearAllErrors(){
+  if(!confirm('确认清除所有设备的异常状态？（运行中/等待重试的设备会跳过）'))return;
+  const r=await apiPost('/api/device/clear_all_errors');
+  handleResult(r,()=>loadMonitor());
+}
+
 // ================== 设备画面查看 ==================
 let _shotTimer=null;       // 自动刷新定时器
 let _shotSerial='';        // 当前查看的 serial
@@ -456,10 +469,15 @@ function renderDeviceRow(dev){
   const endInfo=(dev.worker_status==='running'&&dev.end_time)?'<div style="font-size:11px;color:#94a3b8">预计结束 '+new Date(dev.end_time*1000).toLocaleTimeString()+'</div>':'';
   // 选择器健康警告（某步骤连续未找到元素，可能 App 改版选择器失效）
   const warnInfo=dev.last_warning?'<div style="font-size:11px;color:#f59e0b" title="'+esc(dev.last_warning)+'">⚠ '+esc(dev.last_warning)+'</div>':'';
-  const stopBtn=(dev.worker_status==='running'||dev.worker_status==='connecting')?'<button class="btn btn-danger btn-xs" onclick="apiAction(\'/api/stop_device\',{serial:\''+esc(dev.serial)+'\'})">停止</button>':'';
+  // 设备操作按钮需要"设备控制"权限（无权限只保留截图查看）
+  const canDev=_can('devices');
+  const stopBtn=canDev&&(dev.worker_status==='running'||dev.worker_status==='connecting')?'<button class="btn btn-danger btn-xs" onclick="apiAction(\'/api/stop_device\',{serial:\''+esc(dev.serial)+'\'})">停止</button>':'';
   // 截图按钮（任何状态都可看，离线时会报错）
   const shotBtn='<button class="btn btn-xs" onclick="openScreenshot(\''+esc(dev.serial)+'\')" title="查看设备画面（不影响任务）">截图</button>';
-  const btns='<div style="display:flex;gap:4px;flex-wrap:wrap">'+shotBtn+stopBtn+'</div>';
+  // 异常/失败设备提供"清除异常"（把状态恢复为未启动）
+  const clearBtn=canDev&&(dev.worker_status==='error'||dev.worker_status==='failed')
+    ?'<button class="btn btn-xs" onclick="clearDeviceError(\''+esc(dev.serial)+'\')" title="清除该设备的异常状态，恢复为未启动">清除异常</button>':'';
+  const btns='<div style="display:flex;gap:4px;flex-wrap:wrap">'+shotBtn+clearBtn+stopBtn+'</div>';
   const devName=dev.device_name||dev.model||'';
   const devLabel=devName?'<strong>'+esc(dev.serial)+'</strong><br><span style="font-size:11px;color:#94a3b8">'+esc(devName)+'</span>':'<strong>'+esc(dev.serial)+'</strong>';
   // 复选框列

templates/admin/monitor.html

@@ -289,10 +289,11 @@ select.form-control{cursor:pointer}
 
   <div class="section-title">设备列表</div>
   <div class="toolbar">
-    <button class="btn btn-danger" onclick="apiAction('/api/stop_all')">停止全部</button>
+    <button class="btn btn-danger" onclick="apiAction('/api/stop_all')" data-perm="devices">停止全部</button>
-    <button class="btn btn-danger" id="btn-stop-selected" onclick="stopSelectedDevices()" title="停止所有勾选的运行中设备">停止选中</button>
+    <button class="btn btn-danger" id="btn-stop-selected" onclick="stopSelectedDevices()" title="停止所有勾选的运行中设备" data-perm="devices">停止选中</button>
-    <button class="btn" onclick="apiAction('/api/release')">强制释放占用</button>
+    <button class="btn btn-danger" onclick="clearAllErrors()" title="清除所有异常/失败设备的错误状态（运行中/等待重试的跳过）" data-perm="devices">清除全部异常</button>
-    <button class="btn btn-primary" id="btn-scan-fg" onclick="scanForeground()">扫描前台App</button>
+    <button class="btn" onclick="apiAction('/api/release')" data-perm="devices">强制释放占用</button>
+    <button class="btn btn-primary" id="btn-scan-fg" onclick="scanForeground()" data-perm="devices">扫描前台App</button>
     <button class="btn" onclick="loadMonitor()">刷新</button>
     <input class="form-control list-search" id="search-devices" placeholder="搜索设备/型号/App/任务...">
     <span class="hint"><span id="fg-status"></span> · 每 5 秒自动刷新 · <span id="srvTime"></span> · <span id="sel-count">未选中</span></span>

web_server.py

@@ -28,12 +28,13 @@ from flask_login import (LoginManager, login_user, logout_user,
 
 from core.stf_client import STFClient
 from core.task_manager import TaskManager
-from core.device_worker import get_all_worker_status
+from core.device_worker import clear_worker_error, clear_all_worker_errors
 from core.logger import get_logger, _LOG_DIR, _MODULE_FILES
 from core.models import db, init_db, User, DeviceGroup, TaskJob, CustomAction
 from core.apk_manager import ApkManager
-from core.adb_helper import screenshot, list_installed_apps
+from core.adb_helper import screenshot, list_installed_apps, adb_connect_light, _ADB_LOCK, _adb
 from core import uiauto_helper
+from config import ADB_PATH, STF_SSH_TARGET, STF_DOCKER_CONTAINER
 from tasks import list_task_types, get_task_class
 
 _log = get_logger("web")
@@ -439,19 +440,45 @@ def api_stop_device():
 
 
 @app.route("/api/stop_all", methods=["POST"])
-@login_required
+@perm_required(PERM_DEVICES)
 def api_stop_all():
     stopped = mgr.stop_all()
     return jsonify({"ok": True, "stopped": stopped})
 
 
 @app.route("/api/release", methods=["POST"])
-@login_required
+@perm_required(PERM_DEVICES)
 def api_release():
     released, failed = stf.release_all_mine()
     return jsonify({"ok": True, "released": released, "failed": failed})
 
 
+@app.route("/api/device/clear_error", methods=["POST"])
+@perm_required(PERM_DEVICES)
+def api_device_clear_error():
+    """清除单台设备的异常状态（error/failed → idle），供设备列表"清除异常"按钮使用。"""
+    serial = (request.json or {}).get("serial", "")
+    if not serial:
+        return jsonify({"ok": False, "error": "缺少 serial"}), 400
+    if serial in mgr.get_running():
+        return jsonify({"ok": False, "error": "设备正在运行或等待重试，无法清除"}), 400
+    if not clear_worker_error(serial):
+        return jsonify({"ok": False, "error": "设备正在运行任务，无法清除"}), 400
+    _log.info(f"清除设备 {serial} 的异常状态")
+    return jsonify({"ok": True, "msg": f"已清除 {serial} 的异常状态"})
+
+
+@app.route("/api/device/clear_all_errors", methods=["POST"])
+@perm_required(PERM_DEVICES)
+def api_device_clear_all_errors():
+    """一键清除所有异常/失败设备（跳过正在运行/等待重试的）。"""
+    running = set(mgr.get_running())
+    cleared = clear_all_worker_errors(exclude=running)
+    _log.info(f"一键清除异常：共清除 {cleared} 台")
+    return jsonify({"ok": True, "cleared": cleared,
+                    "msg": f"已清除 {cleared} 台设备的异常状态"})
+
+
 @app.route("/api/device/screenshot")
 @login_required
 def api_device_screenshot():