feat: 重连不再等待并精简连接日志,失败才记断开与退避

- 连接曾稳定存活(>= _STABLE_SECONDS)而被服务端要求重连时立刻重连、
  不等待也不记日志:这期间事件由网关按 RESUME 补发,等待只是徒增盲区
- 仅在连接未能稳定存活(握手失败/看门狗超时/限流/抖动)时才记
  "WebSocket 已断开"并按指数退避重试
- 去掉常规的「WebSocket 已连接/已断开」info,只保留「发送 RESUME」等
  有信息量的行,消除每小时一次的日志噪音
- 新增 tests/test_reconnect.py:健康断开零等待零日志、失败退避翻倍、
  限流 60s、抖动不重置退避
This commit is contained in:
snowy-yang committed 2026-09-21 18:25:16 +08:00
1 parent 1552c5207b
commit b37a53f6e0
2 files changed
+119 -8

No files matched your search

+26 -8
View File
@@ -72,22 +72,42 @@ class WebsocketConnecter(Connecter):
self._outbound: asyncio.Queue[Payload] = asyncio.Queue()
async def run(self) -> None:
"""重连主循环。
健康连接被服务端要求重连(如满 1 小时的连接限寿)时不等待、也不记
连接/断开日志,直接立刻重连——这期间的事件由网关按 RESUME 补发,
盲区最小;只有连接未能稳定存活(握手失败、看门狗超时、限流、抖动)
才记断开并按退避等待,`_STABLE_SECONDS` 保证抖动的服务端不会被 1s
间隔反复探测。
"""
backoff: float = 1
delay: float = 0 # 0 表示立刻重连(首次重试不等待)
while True:
if delay:
logger.info(f"{delay}s 后重连")
await asyncio.sleep(delay)
try:
lived = await self._connect()
if lived >= self._STABLE_SECONDS:
backoff = 1
except RateLimitError:
logger.warning("gateway 触发频率限制")
backoff = 60
delay = backoff = 60
continue
except TimeoutError:
logger.warning("心跳超时,触发看门狗重连")
lived = 0.0
except (aiohttp.ClientError, RuntimeError) as e:
logger.warning(f"WebSocket 连接异常: {e}")
logger.info(f"{backoff}s 后重连")
await asyncio.sleep(backoff)
backoff = min(backoff * 2, 60)
lived = 0.0
if lived >= self._STABLE_SECONDS:
# 连接曾稳定存活:属正常断开(服务端要求重连),立刻重连
backoff = 1
delay = 0
else:
# 没能稳定存活:需要延迟重试,此时才记断开
logger.warning("WebSocket 已断开")
delay = backoff
backoff = min(backoff * 2, 60)
async def _connect(self) -> float:
"""建立一次连接直到断开,返回存活秒数(供 run 判断是否重置退避)。"""
@@ -107,7 +127,6 @@ class WebsocketConnecter(Connecter):
raise RuntimeError(f"获取 gateway 失败: {data}")
async with self._http.ws_connect(gateway, headers=headers) as ws:
logger.info("WebSocket 已连接")
# HELLO 才给出心跳间隔;receive 解析到后经 Future 交给心跳任务
interval: asyncio.Future[float] = asyncio.get_running_loop().create_future()
receive = asyncio.create_task(self.receive_helper(ws, interval))
@@ -121,7 +140,6 @@ class WebsocketConnecter(Connecter):
# 取回被取消任务的结果,避免 "Task exception was never retrieved",
# 也确保它们在下一轮重连前真正停下
await asyncio.gather(send, heartbeat, return_exceptions=True)
logger.info("WebSocket 已断开")
return monotonic() - started
async def receive_helper(
+93
View File
@@ -0,0 +1,93 @@
"""重连循环:健康断开立刻重连(不记连接/断开),失败才退避并记断开。"""
import asyncio
from typing import Any
import pytest
from loguru import logger
from qqbotsdk.connecter import RateLimitError, WebsocketConnecter
pytestmark = pytest.mark.asyncio
def _make(monkeypatch: pytest.MonkeyPatch, seq: list[Any]) -> WebsocketConnecter:
"""构造 connecter,并把 _connect 换成按 seq 依次返回(float=存活秒数,
Exception=连接失败)。seq 用尽后抛 IndexError 中止 run() 的无限循环。"""
conn = WebsocketConnecter.__new__(WebsocketConnecter)
remaining = list(seq)
async def fake_connect() -> float:
item = remaining.pop(0) # 用尽则 IndexError 逃出 run()
if isinstance(item, Exception):
raise item
return float(item)
monkeypatch.setattr(conn, "_connect", fake_connect)
return conn
async def _run_capture(
monkeypatch: pytest.MonkeyPatch, seq: list[Any]
) -> tuple[list[float], list[str]]:
"""跑 run() 直到 seq 用尽,返回 (sleep 延迟序列, 日志消息列表)。"""
sleeps: list[float] = []
logs: list[str] = []
real_sleep = asyncio.sleep
async def fake_sleep(delay: float) -> None:
sleeps.append(delay)
await real_sleep(0) # 让出控制权,避免紧循环
monkeypatch.setattr(asyncio, "sleep", fake_sleep)
sink_id = logger.add(lambda m: logs.append(m.record["message"]), level="INFO")
try:
conn = _make(monkeypatch, seq)
with pytest.raises(IndexError):
await conn.run()
finally:
logger.remove(sink_id)
return sleeps, logs
async def test_healthy_disconnect_reconnects_without_delay_or_log(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""健康连接(存活 >= 稳定阈值)断开后立刻重连:不 sleep、不记连接/断开。"""
sleeps, logs = await _run_capture(monkeypatch, [3600.0, 3600.0])
assert sleeps == [] # 两次都立刻重连
assert not any("已连接" in m or "已断开" in m for m in logs)
async def test_failed_reconnect_backs_off_and_logs_disconnect(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""连不上时记断开并按退避等待,退避逐次翻倍。"""
sleeps, logs = await _run_capture(
monkeypatch, [3600.0, TimeoutError(), 0.0, 0.0]
)
# 健康断开 → 无 sleep;失败后依次 1s、2s、4s
assert sleeps == [1, 2, 4]
assert sum("已断开" in m for m in logs) == 3
async def test_rate_limit_backs_off_hard(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""gateway 限流:退避 60s,且不重复记断开。"""
sleeps, logs = await _run_capture(monkeypatch, [RateLimitError(), 3600.0])
assert sleeps == [60]
assert any("频率限制" in m for m in logs)
assert not any("已断开" in m for m in logs)
async def test_flapping_connection_does_not_reset_backoff(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""连接建立后立刻掉线(存活 < 阈值)视为抖动:退避持续拉长,不归零。"""
sleeps, _ = await _run_capture(monkeypatch, [0.5, 0.5, 0.5])
assert sleeps == [1, 2, 4] # 每次都等得更久(若归零会是 1,1,1)