更新：

backend/rpa_engine/douyin_im/rate_limit.py

@@ -0,0 +1,117 @@
+"""全局出站发送限流器。
+
+目的：托管账号数量多（如 1000 个）时，避免自动回复/手动发送在同一时刻
+大量并发打向抖音接口，占满服务器出站带宽或触发平台级风控。
+
+机制（两层，先到先过）：
+  1. 并发上限（Semaphore）：同时在途的发送请求数不超过 KEFU_SEND_MAX_CONCURRENCY；
+  2. 令牌桶（QPS）：平均发送速率不超过 KEFU_SEND_RATE_PER_SEC，允许 KEFU_SEND_BURST 的突发。
+
+限流器按事件循环惰性创建（backend 主循环一个实例；独立脚本各自一个），
+所有账号共享同一个实例，因此是跨账号的全局闸门。
+
+环境变量：
+  KEFU_SEND_MAX_CONCURRENCY  同时在途发送数上限，默认 20，<=0 表示不限
+  KEFU_SEND_RATE_PER_SEC     平均每秒发送数上限，默认 10，<=0 表示不限
+  KEFU_SEND_BURST            令牌桶容量（突发上限），默认与并发上限相同
+"""
+from __future__ import annotations
+
+import asyncio
+import logging
+import os
+import time
+
+logger = logging.getLogger("douyin_im.rate_limit")
+
+
+def _env_float(name: str, default: float) -> float:
+    try:
+        return float(os.getenv(name, "") or default)
+    except ValueError:
+        return default
+
+
+class GlobalSendLimiter:
+    """并发上限 + 令牌桶。用法：async with limiter: await 发送。"""
+
+    def __init__(
+        self,
+        max_concurrency: int = 20,
+        rate_per_sec: float = 10.0,
+        burst: float | None = None,
+    ):
+        self.max_concurrency = int(max_concurrency)
+        self.rate_per_sec = float(rate_per_sec)
+        self.burst = float(burst if burst is not None else max(1, max_concurrency))
+        self._sem = (
+            asyncio.Semaphore(self.max_concurrency) if self.max_concurrency > 0 else None
+        )
+        self._tokens = self.burst
+        self._last_refill = time.monotonic()
+        self._token_lock = asyncio.Lock()
+        # 统计：便于日志观察限流是否生效
+        self._waited_total = 0.0
+        self._acquired_count = 0
+
+    async def _take_token(self) -> None:
+        if self.rate_per_sec <= 0:
+            return
+        async with self._token_lock:
+            while True:
+                now = time.monotonic()
+                self._tokens = min(
+                    self.burst, self._tokens + (now - self._last_refill) * self.rate_per_sec
+                )
+                self._last_refill = now
+                if self._tokens >= 1.0:
+                    self._tokens -= 1.0
+                    return
+                await asyncio.sleep((1.0 - self._tokens) / self.rate_per_sec)
+
+    async def __aenter__(self) -> "GlobalSendLimiter":
+        start = time.monotonic()
+        if self._sem is not None:
+            await self._sem.acquire()
+        try:
+            await self._take_token()
+        except BaseException:
+            if self._sem is not None:
+                self._sem.release()
+            raise
+        waited = time.monotonic() - start
+        self._waited_total += waited
+        self._acquired_count += 1
+        if waited > 1.0:
+            logger.info(
+                "send throttled: waited %.1fs (in-flight cap=%s, rate=%s/s)",
+                waited, self.max_concurrency or "∞", self.rate_per_sec or "∞",
+            )
+        return self
+
+    async def __aexit__(self, *exc) -> None:
+        if self._sem is not None:
+            self._sem.release()
+
+
+# 每个事件循环一个实例（backend 主循环即全局唯一；脚本自用循环互不影响）
+_limiters: dict[int, GlobalSendLimiter] = {}
+
+
+def get_send_limiter() -> GlobalSendLimiter:
+    loop = asyncio.get_running_loop()
+    key = id(loop)
+    limiter = _limiters.get(key)
+    if limiter is None:
+        max_conc = int(_env_float("KEFU_SEND_MAX_CONCURRENCY", 20))
+        rate = _env_float("KEFU_SEND_RATE_PER_SEC", 10.0)
+        burst = _env_float("KEFU_SEND_BURST", max(1, max_conc))
+        limiter = GlobalSendLimiter(max_conc, rate, burst)
+        _limiters[key] = limiter
+        logger.info(
+            "global send limiter ready: concurrency=%s, rate=%s/s, burst=%s",
+            max_conc if max_conc > 0 else "unlimited",
+            rate if rate > 0 else "unlimited",
+            burst,
+        )
+    return limiter