Pythonasyncio异步并发与多固定出口IP调度实战

一、为什么异步场景需要不同的出口管理策略

1.1 同步 vs 异步的核心差异

1.2 异步场景的核心挑战

单事件循环下管理多个出口，最大的挑战是状态一致性：多个协程同时 acquire/release 出口，如果用普通的 dict 管理分配状态，协程切换时机不对会导致同一个出口被超量分配。必须用 asyncio.Lock 或 asyncio.Semaphore 做并发保护。

第二个挑战是健康检查不阻塞：同步场景健康检查在后台线程跑，阻塞不影响主流程；异步场景下如果健康检查协程卡在 await 上，会拖慢整个事件循环的调度。需要给健康检查加超时和熔断。

二、基于源地址绑定的异步连接管理

2.1 为什么不用应用层转发配置

aiohttp 支持在应用层配置 HTTP 转发地址，但这有两个问题：一是每条请求都经过中间节点转发，多一跳延迟；二是某些对端服务会检测转发协议头（如 Via、X-Forwarded-For），判定为非直连。

更稳妥的方式是源地址绑定：在 TCP 层将连接绑定到指定的本地 IP 地址，对端看到的是该 IP 的直连请求，没有转发特征。

2.2 aiohttp TCPConnector 源地址绑定

import asyncio
import aiohttp
from dataclasses import dataclass, field
from typing import Optional
import time

@dataclass
class ExitConfig:
    """单个固定出口的配置"""
    name: str               # 出口标识
    source_ip: str          # 绑定的本地源 IP
    max_concurrent: int = 100   # 该出口最大并发连接数
    enabled: bool = True

@dataclass
class ExitState:
    """出口运行时状态"""
    config: ExitConfig
    connector: Optional[aiohttp.TCPConnector] = None
    healthy: bool = True
    active_count: int = 0       # 当前活跃连接数
    last_check: float = 0.0     # 上次健康检查时间
    fail_count: int = 0         # 连续失败次数
    total_requests: int = 0     # 总请求数
    total_errors: int = 0       # 总错误数

    @property
    def error_rate(self) -> float:
        if self.total_requests == 0:
            return 0.0
        return self.total_errors / self.total_requests

    @property
    def a vailable(self) -> bool:
        """是否可分配：健康 + 未达并发上限"""
        return (
            self.config.enabled
            and self.healthy
            and self.active_count < self.config.max_concurrent
        )

2.3 创建带源地址绑定的 Connector

async def create_connector(exit_config: ExitConfig) -> aiohttp.TCPConnector:
    """为指定出口创建绑定源 IP 的 TCPConnector"""
    connector = aiohttp.TCPConnector(
        local_addr=(exit_config.source_ip, 0),  # 绑定源 IP，端口随机
        limit=exit_config.max_concurrent,        # 连接池上限
        limit_per_host=50,                       # 单对端最大连接
        keepalive_timeout=30,                    # keepalive 超时
        enable_cleanup_closed=True,              # 关闭的连接及时清理
        ttl_dns_cache=300,                       # DNS 缓存 5 分钟
    )
    return connector

local_addr 参数是关键——它在 socket 层调用 bind() 绑定到指定本地 IP，所有通过该 Connector 发出的连接都从这个 IP 出去。用 ss -tnp 可以验证：

# 查看指定源 IP 的活跃连接
ss -tnp | grep "10.0.1.100"
# 输出示例：
# ESTAB  0  0  10.0.1.100:54321  93.184.216.34:443

三、多出口并发调度器

3.1 AsyncExitPool 核心类

class AsyncExitPool:
    """异步多固定出口 IP 调度池"""

    def __init__(self, exit_configs: list[ExitConfig]):
        self._exits: dict[str, ExitState] = {}
        self._lock = asyncio.Lock()
        self._init_exits(exit_configs)

    def _init_exits(self, configs: list[ExitConfig]):
        for cfg in configs:
            self._exits[cfg.name] = ExitState(config=cfg)

    async def start(self):
        """初始化所有出口的 Connector"""
        for state in self._exits.values():
            state.connector = await create_connector(state.config)

    async def acquire(self, preferred: Optional[str] = None) -> tuple[str, aiohttp.TCPConnector]:
        """
        获取一个可用出口的 Connector
        preferred: 优先使用的出口名称（None 表示自动选择负载最低的）
        返回: (出口名称, Connector)
        """
        async with self._lock:
            # 优先尝试指定出口
            if preferred and preferred in self._exits:
                state = self._exits[preferred]
                if state.a vailable:
                    state.active_count += 1
                    return preferred, state.connector

            # 自动选择：负载最低的健康出口
            candidates = [s for s in self._exits.values() if s.a vailable]
            if not candidates:
                raise RuntimeError("无可用出口")

            # 按 active_count / max_concurrent 比率排序，选负载最低的
            candidates.sort(key=lambda s: s.active_count / s.config.max_concurrent)
            chosen = candidates[0]
            chosen.active_count += 1
            return chosen.config.name, chosen.connector

    async def release(self, name: str, success: bool = True):
        """归还出口"""
        async with self._lock:
            state = self._exits.get(name)
            if state:
                state.active_count = max(0, state.active_count - 1)
                state.total_requests += 1
                if not success:
                    state.total_errors += 1
                    state.fail_count += 1
                    if state.fail_count >= 3:
                        state.healthy = False
                        print(f"[WARN] 出口 {name} 连续失败 {state.fail_count} 次，标记为不健康")
                else:
                    state.fail_count = 0  # 成功则重置失败计数

    async def get_stats(self) -> list[dict]:
        """获取所有出口状态"""
        async with self._lock:
            return [
                {
                    "name": s.config.name,
                    "source_ip": s.config.source_ip,
                    "healthy": s.healthy,
                    "active": s.active_count,
                    "max": s.config.max_concurrent,
                    "usage": f"{s.active_count}/{s.config.max_concurrent}",
                    "total_req": s.total_requests,
                    "error_rate": f"{s.error_rate:.2%}",
                    "fail_count": s.fail_count,
                }
                for s in self._exits.values()
            ]

3.2 使用示例

async def fetch_with_exit(pool: AsyncExitPool, url: str, preferred: str = None):
    """使用固定出口发起异步请求"""
    exit_name, connector = await pool.acquire(preferred)
    success = False
    try:
        timeout = aiohttp.ClientTimeout(total=10, connect=5)
        async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
            async with session.get(url) as resp:
                if resp.status == 200:
                    success = True
                    return await resp.text()
                else:
                    print(f"[WARN] {exit_name} -> {url} status={resp.status}")
                    return None
    except Exception as e:
        print(f"[ERROR] {exit_name} -> {url}: {e}")
        return None
    finally:
        await pool.release(exit_name, success)


async def main():
    configs = [
        ExitConfig(name="us-01", source_ip="10.0.1.100", max_concurrent=100),
        ExitConfig(name="us-02", source_ip="10.0.1.101", max_concurrent=100),
        ExitConfig(name="eu-01", source_ip="10.0.2.100", max_concurrent=80),
    ]
    pool = AsyncExitPool(configs)
    await pool.start()

    # 并发 500 个任务，自动分配到 3 个出口
    urls = [f"https://httpbin.org/delay/{i % 3}" for i in range(500)]
    tasks = [fetch_with_exit(pool, url) for url in urls]
    results = await asyncio.gather(*tasks, return_exceptions=True)

    # 查看出口使用情况
    stats = await pool.get_stats()
    for s in stats:
        print(s)

asyncio.run(main())

四、异步健康检查与故障转移

4.1 定时健康检查协程

class ExitHealthChecker:
    """异步出口健康检查器"""

    def __init__(self, pool: AsyncExitPool, check_url: str, interval: int = 30):
        self.pool = pool
        self.check_url = check_url   # 健康检查用的对端地址
        self.interval = interval     # 检查间隔（秒）
        self._task: Optional[asyncio.Task] = None

    async def _check_one(self, name: str, state: ExitState):
        """检查单个出口"""
        timeout = aiohttp.ClientTimeout(total=5, connect=3)
        try:
            async with aiohttp.ClientSession(connector=state.connector, timeout=timeout) as session:
                start = time.monotonic()
                async with session.get(self.check_url) as resp:
                    latency = (time.monotonic() - start) * 1000
                    if resp.status == 200:
                        async with self.pool._lock:
                            if not state.healthy:
                                print(f"[INFO] 出口 {name} 恢复健康 (latency={latency:.0f}ms)")
                            state.healthy = True
                            state.fail_count = 0
                            state.last_check = time.time()
                        return True
        except Exception:
            async with self.pool._lock:
                state.fail_count += 1
                if state.fail_count >= 3 and state.healthy:
                    state.healthy = False
                    print(f"[WARN] 出口 {name} 健康检查失败，标记不健康")
                state.last_check = time.time()
            return False

    async def _run(self):
        """定时检查循环"""
        while True:
            tasks = [
                self._check_one(name, state)
                for name, state in self.pool._exits.items()
                if state.config.enabled
            ]
            await asyncio.gather(*tasks, return_exceptions=True)
            await asyncio.sleep(self.interval)

    def start(self):
        """启动健康检查后台协程"""
        self._task = asyncio.create_task(self._run())

    def stop(self):
        if self._task:
            self._task.cancel()

4.2 健康检查指标

4.3 故障转移逻辑

当 acquire() 发现所有出口不可用时，有两种策略：

async def acquire_with_fallback(self, preferred: str = None) -> tuple[str, aiohttp.TCPConnector]:
    """带故障转移的获取出口"""
    try:
        return await asyncio.wait_for(self.acquire(preferred), timeout=3.0)
    except (RuntimeError, asyncio.TimeoutError):
        # 所有出口不可用，尝试重新激活被标记为不健康的出口
        async with self._lock:
            for state in self._exits.values():
                if not state.healthy and state.config.enabled:
                    print(f"[WARN] 全部出口不可用，尝试重新激活 {state.config.name}")
                    state.healthy = True
                    state.fail_count = 0
        # 重试一次
        return await asyncio.wait_for(self.acquire(preferred), timeout=3.0)

五、速率控制与出口轮换

5.1 令牌桶限速器

每个出口独立限速，避免单个 IP 请求过密触发对端限制：

class AsyncTokenBucket:
    """异步令牌桶：控制单个出口的请求速率"""

    def __init__(self, rate: float, capacity: int):
        self.rate = rate           # 令牌生成速率（个/秒）
        self.capacity = capacity   # 桶容量
        self._tokens = capacity    # 当前令牌数
        self._last_refill = time.monotonic()
        self._lock = asyncio.Lock()

    async def acquire(self, tokens: int = 1):
        """获取令牌，不足时等待"""
        async with self._lock:
            while self._tokens < tokens:
                # 计算需要等待的时间
                needed = tokens - self._tokens
                wait_time = needed / self.rate
                await asyncio.sleep(wait_time)
                self._refill()
            self._tokens -= tokens

    def _refill(self):
        now = time.monotonic()
        elapsed = now - self._last_refill
        self._tokens = min(self.capacity, self._tokens + elapsed * self.rate)
        self._last_refill = now

5.2 出口轮换策略

class ExitRotator:
    """出口轮换管理：按任务特征分配出口"""

    def __init__(self, pool: AsyncExitPool):
        self.pool = pool
        self._buckets: dict[str, AsyncTokenBucket] = {}

    def register_rate_limit(self, exit_name: str, rate: float, capacity: int):
        """为出口注册速率限制"""
        self._buckets[exit_name] = AsyncTokenBucket(rate, capacity)

    async def fetch(self, url: str, region: str = None) -> Optional[str]:
        """
        按区域偏好选择出口并限速
        region: 地区标识（如 us/eu），None 表示自动
        """
        # 按地区筛选可用出口
        preferred = f"{region}-01" if region else None

        exit_name, connector = await self.pool.acquire(preferred)

        # 获取该出口的令牌桶
        bucket = self._buckets.get(exit_name)
        if bucket:
            await bucket.acquire()  # 等待令牌

        success = False
        try:
            timeout = aiohttp.ClientTimeout(total=10)
            async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
                async with session.get(url) as resp:
                    if resp.status == 200:
                        success = True
                        return await resp.text()
            self._tokens -= tokens

    def _refill(self):
        now = time.monotonic()
        elapsed = now - self._last_refill
        self._tokens = min(self.capacity, self._tokens + elapsed * self.rate)
        self._last_refill = now

5.2 出口轮换策略

class ExitRotator:
    """出口轮换管理：按任务特征分配出口"""

    def __init__(self, pool: AsyncExitPool):
        self.pool = pool
        self._buckets: dict[str, AsyncTokenBucket] = {}

    def register_rate_limit(self, exit_name: str, rate: float, capacity: int):
        """为出口注册速率限制"""
        self._buckets[exit_name] = AsyncTokenBucket(rate, capacity)

    async def fetch(self, url: str, region: str = None) -> Optional[str]:
        """
        按区域偏好选择出口并限速
        region: 地区标识（如 us/eu），None 表示自动
        """
        # 按地区筛选可用出口
        preferred = f"{region}-01" if region else None

        exit_name, connector = await self.pool.acquire(preferred)

        # 获取该出口的令牌桶
        bucket = self._buckets.get(exit_name)
        if bucket:
            await bucket.acquire()  # 等待令牌

        success = False
        try:
            timeout = aiohttp.ClientTimeout(total=10)
            async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
                async with session.get(url) as resp:
                    if resp.status == 200:
                        success = True
                        return await resp.text()
        except Exception as e:
            print(f"[ERROR] {exit_name}: {e}")
        finally:
            await self.pool.release(exit_name, success)
        return None

5.3 速率控制参数参考

六、监控指标采集

6.1 Prometheus 格式指标输出

class ExitMetrics:
    """出口池指标采集器"""

    @staticmethod
    def to_prometheus(stats: list[dict]) -> str:
        lines = []
        for s in stats:
            labels = f'name="{s["name"]}",ip="{s["source_ip"]}"'
            lines.append(f'exit_healthy{{{labels}}} {1 if s["healthy"] else 0}')
            lines.append(f'exit_active_connections{{{labels}}} {s["active"]}')
            lines.append(f'exit_max_connections{{{labels}}} {s["max"]}')
            lines.append(f'exit_total_requests{{{labels}}} {s["total_req"]}')
            # error_rate 需要解析
            err = float(s["error_rate"].rstrip("%")) / 100
            lines.append(f'exit_error_rate{{{labels}}} {err}')
        return "n".join(lines) + "n"

6.2 暴露 HTTP 指标端点

from aiohttp import web

async def metrics_handler(request: web.Request):
    pool: AsyncExitPool = request.app["exit_pool"]
    stats = await pool.get_stats()
    text = ExitMetrics.to_prometheus(stats)
    return web.Response(text=text, content_type="text/plain")

async def start_metrics_server(pool: AsyncExitPool, port: int = 9100):
    app = web.Application()
    app["exit_pool"] = pool
    app.router.add_get("/metrics", metrics_handler)
    runner = web.AppRunner(app)
    await runner.setup()
    site = web.TCPSite(runner, "0.0.0.0", port)
    await site.start()
    print(f"[INFO] 指标端点启动: http://0.0.0.0:{port}/metrics")

6.3 监控指标表

七、四个常见误区

误区一：异步场景直接复用同步的 ExitPool。 同步场景里，线程锁（threading.Lock）确实能把状态保护住；可一旦放到异步环境里，这套做法就不对劲了。线程锁会把整个事件循环卡住——某个协程一旦拿到锁，其他协程就没机会被切换上来，结果本来并发的异步流程，硬生生退化成了串行。正确的做法是换成 asyncio.Lock，让协程在 await 时自然让出事件循环。

误区二：每个请求创建新的 Connector。 Connector 内部维护连接池，频繁创建/销毁会导致 TCP 连接无法复用，每次都走完整握手。正确做法是每个出口一个 Connector，生命周期与 ExitPool 一致，只在关闭时清理。

误区三：健康检查不加超时。 健康检查协程如果 await 一个卡住的对端，会占用事件循环的调度槽。虽然 asyncio 不会完全阻塞（其他协程还能跑），但积压太多超时协程会导致事件循环响应变慢。每个健康检查请求必须设 total=5, connect=3 的超时。

误区四：用 Semaphore 替代 active_count 管理。asyncio.Semaphore 的作用，只是把总并发数卡住；它并不会告诉你，现在到底有多少资源正在被占用，哪一个出口又更“忙”。真正要做按负载分配，就得靠 active_count 这种显式计数方式，把每个出口的压力看得清清楚楚，进而优先选择负载最低的出口，而不是停留在简单轮询这一步。

排查清单

# 1. 确认源地址绑定生效
ss -tnp | grep "10.0.1.100"
# 应看到大量 ESTAB 连接从该 IP 发出

# 2. 确认连接池复用（TIME_WAIT 少说明复用正常）
ss -s | grep "TIME-WAIT"

# 3. 查看出口使用均衡度
curl http://localhost:9100/metrics | grep exit_active

# 4. 确认事件循环没有阻塞（事件循环延迟 < 50ms）
# 在代码中插入：
# loop_start = time.monotonic()
# await asyncio.sleep(0)
# loop_delay = (time.monotonic() - loop_start) * 1000
# print(f"事件循环延迟: {loop_delay:.1f}ms")

# 5. 查看协程数量是否合理
# asyncio.all_tasks() 返回当前所有活跃协程
print(f"活跃协程数: {len(asyncio.all_tasks())}")

小结

异步场景下多固定出口 IP 管理的核心就三条：源地址绑定替代应用层转发（无转发特征、无额外跳数）、asyncio.Lock 保护状态一致性（协程级并发安全）、非阻塞健康检查与故障转移（不拖慢事件循环）。把 AsyncExitPool + 令牌桶限速 + Prometheus 指标这套搭起来，千级并发任务跨多个固定出口调度就能稳定跑。