更新：

backend/rpa_engine/playwright_worker.py

@@ -49,6 +49,36 @@ async def _launch_chromium(pw, args: list[str], headless: Optional[bool] = None)
     return await pw.chromium.launch(**launch_kwargs)
 
 
+# ---------------------------------------------------------------------------
+# 启动错峰门：批量启动大量账号时，把各 worker 的启动时刻按固定间隔排开，
+# 避免同一瞬间大量凭证校验/WS 建连/首轮拉取叠峰。
+# 空闲时单个账号启动无需等待；只有短时间内大量启动才会排队。
+# KEFU_WORKER_START_INTERVAL_SECONDS：相邻两个 worker 启动的最小间隔，默认 1.5s，<=0 关闭。
+# ---------------------------------------------------------------------------
+_start_gate = {"lock": None, "next_at": 0.0}
+
+
+async def _startup_stagger(account_id: int) -> None:
+    try:
+        interval = float(os.getenv("KEFU_WORKER_START_INTERVAL_SECONDS", "") or 1.5)
+    except ValueError:
+        interval = 1.5
+    if interval <= 0:
+        return
+    if _start_gate["lock"] is None:
+        _start_gate["lock"] = asyncio.Lock()
+    async with _start_gate["lock"]:
+        now = time.monotonic()
+        wait = max(0.0, _start_gate["next_at"] - now)
+        _start_gate["next_at"] = max(now, _start_gate["next_at"]) + interval
+    if wait > 0:
+        if wait > 5:
+            logger.info(
+                f"Account {account_id}: start queued, waiting {wait:.1f}s to smooth batch startup"
+            )
+        await asyncio.sleep(wait)
+
+
 def format_error(exc: BaseException) -> str:
     message = str(exc).strip()
     if "Target page, context or browser has been closed" in message:
@@ -995,6 +1025,9 @@ class DouyinWorker:
         logger.info(f"Starting worker loop for account {self.account_id}")
 
         try:
+            await _startup_stagger(self.account_id)
+            if self.stopping:
+                return
             storage_state = await self._load_storage_state()
             cookie_info = analyze_cookie(
                 json.dumps(storage_state, ensure_ascii=False) if storage_state else None