sycommon-python-lib 0.1.46__py3-none-any.whl → 0.1.57b1__py3-none-any.whl

sycommon/rabbitmq/rabbitmq_pool.py

@@ -1,19 +1,29 @@
 import asyncio
-from typing import Optional, List, Set, Iterator, Tuple, Dict, Callable
-from aio_pika import connect_robust, Channel, Message
+import random
+from typing import Optional, List, Dict, Callable, Tuple
+from aio_pika import connect_robust, RobustChannel, Message
 from aio_pika.abc import (
     AbstractRobustConnection, AbstractQueue, AbstractExchange, AbstractMessage
 )
-from aio_pika.exceptions import ChannelClosed
-import aiormq.exceptions
-
 from sycommon.logging.kafka_log import SYLogger
 
 logger = SYLogger
 
 
+class AsyncProperty:
+    """实现 await obj.attr 的支持"""
+
+    def __init__(self, method):
+        self.method = method
+
+    def __get__(self, obj, objtype=None):
+        if obj is None:
+            return self
+        return self.method(obj)
+
+
 class RabbitMQConnectionPool:
-    """单连接RabbitMQ通道池（核心特性：严格单连接+重连释放旧资源+新连接保留自动恢复+全场景加锁）"""
+    """单连接单通道RabbitMQ客户端 (严格执行“先清理后连接”策略)"""
 
     def __init__(
         self,
@@ -22,14 +32,12 @@ class RabbitMQConnectionPool:
         username: str,
         password: str,
         virtualhost: str = "/",
-        channel_pool_size: int = 1,
-        heartbeat: int = 30,
+        heartbeat: int = 15,
         app_name: str = "",
-        connection_timeout: int = 30,
-        reconnect_interval: int = 30,
+        connection_timeout: int = 15,
+        reconnect_interval: int = 5,
         prefetch_count: int = 2,
     ):
-        # 基础配置校验与初始化
         self.hosts = [host.strip() for host in hosts if host.strip()]
         if not self.hosts:
             raise ValueError("至少需要提供一个RabbitMQ主机地址")
@@ -43,544 +51,348 @@ class RabbitMQConnectionPool:
         self.connection_timeout = connection_timeout
         self.reconnect_interval = reconnect_interval
         self.prefetch_count = prefetch_count
-        self.channel_pool_size = max(1, channel_pool_size)  # 确保池大小不小于1
 
-        # 节点轮询（仅重连时切换）
-        self._host_iterator: Iterator[str] = self._create_host_iterator()
-        self._current_host: Optional[str] = None
+        self._current_host: str = random.choice(self.hosts)
+        logger.info(f"[INIT] 随机选择RabbitMQ主机: {self._current_host}")
 
-        # 核心资源（严格单连接+通道池，仅绑定当前活跃连接）
-        self._connection: Optional[AbstractRobustConnection] = None  # 唯一活跃连接
-        self._free_channels: List[Channel] = []  # 当前连接的空闲通道（带自动恢复）
-        self._used_channels: Set[Channel] = set()  # 当前连接的使用中通道（带自动恢复）
-        self._consumer_channels: Dict[str, Tuple[Channel,
-                                                 AbstractRobustConnection, Callable, bool, dict]] = {}  # 消费者通道跟踪
+        # 核心资源
+        self._connection: Optional[AbstractRobustConnection] = None
+        self._channel: Optional[RobustChannel] = None
+        self._consumer_channels: Dict[str, RobustChannel] = {}
 
-        # 状态控制（确保并发安全和单连接）
-        self._lock = asyncio.Lock()  # 全局唯一锁，保护所有共享状态
+        # 状态控制
+        self._lock = asyncio.Lock()
         self._initialized = False
         self._is_shutdown = False
-        self._reconnecting = False  # 避免并发重连
-        self._connection_version = 0  # 连接版本号（区分新旧连接/通道）
-
-    def _create_host_iterator(self) -> Iterator[str]:
-        """创建无限循环的节点轮询迭代器"""
-        while True:
-            for host in self.hosts:
-                yield host
-
-    async def _is_connection_valid(self) -> bool:
-        """原子化检查连接有效性（加锁保证无竞态）"""
-        async with self._lock:
-            return (
-                self._connection is not None
-                and not self._connection.is_closed
-                and not self._reconnecting
-            )
 
-    @property
+    @AsyncProperty
     async def is_alive(self) -> bool:
-        """对外暴露的连接存活状态（异步+原子化）"""
-        if self._is_shutdown:
-            return False
-        return await self._is_connection_valid()
-
-    async def _safe_close_old_resources(self):
-        """强制关闭所有旧资源（加锁保证原子性，重连前必调用）"""
+        """对外暴露的连接存活状态"""
         async with self._lock:
-            logger.info(f"开始释放旧资源（连接版本: {self._connection_version}）...")
+            if self._is_shutdown:
+                return False
+            if not self._initialized:
+                return False
+            if self._connection is None or self._connection.is_closed:
+                return False
+            if self._channel is None or self._channel.is_closed:
+                return False
+            return True
 
-            # 1. 关闭所有消费者通道（独立管理，终止旧自动恢复）
-            for queue_name, (channel, _, _, _, _) in self._consumer_channels.items():
-                try:
-                    if not channel.is_closed:
-                        await channel.close()
-                    logger.info(f"已关闭队列 {queue_name} 的旧消费者通道（自动恢复终止）")
-                except Exception as e:
-                    logger.warning(f"关闭消费者通道 {queue_name} 失败: {str(e)}")
-            self._consumer_channels.clear()
+    async def _cleanup_resources(self):
+        """
+        彻底清理旧资源
+        必须在持有 self._lock 的情况下调用
+        """
+        logger.info("🧹 [CLEANUP] 开始清理旧资源...")
 
-            # 2. 关闭所有普通通道（空闲+使用中，终止旧自动恢复）
-            all_channels = self._free_channels + list(self._used_channels)
-            for channel in all_channels:
-                try:
-                    if not channel.is_closed:
-                        await channel.close()
-                except Exception as e:
-                    logger.warning(f"关闭旧通道失败: {str(e)}")
-            self._free_channels.clear()
-            self._used_channels.clear()
+        # 1. 清理所有消费者通道
+        if self._consumer_channels:
+            channels_to_close = list(self._consumer_channels.values())
+            self._consumer_channels.clear()
 
-            # 3. 强制关闭旧连接（彻底终止旧连接的所有自动恢复）
-            if self._connection:
+            for ch in channels_to_close:
                 try:
-                    if not self._connection.is_closed:
-                        await self._connection.close()
-                    logger.info(
-                        f"已关闭旧连接: {self._current_host}:{self.port}（版本: {self._connection_version}）")
+                    if not ch.is_closed:
+                        await ch.close()
                 except Exception as e:
-                    logger.warning(f"关闭旧连接失败: {str(e)}")
-                self._connection = None  # 置空，确保单连接
-
-            logger.info("旧资源释放完成（所有旧自动恢复逻辑已终止）")
-
-    async def _create_single_connection(self) -> AbstractRobustConnection:
-        """创建唯一活跃连接（重连前已释放旧资源，确保单连接）"""
-        max_attempts = len(self.hosts) * 2  # 每个节点尝试2次
-        attempts = 0
-        last_error: Optional[Exception] = None
-
-        while attempts < max_attempts and not self._is_shutdown:
-            self._current_host = next(self._host_iterator)
-            conn_url = f"amqp://{self.username}:{self.password}@{self._current_host}:{self.port}/{self.virtualhost}"
+                    logger.warning(f"⚠️ [CLEANUP_CH] 关闭消费者通道失败: {e}")
 
+        # 2. 关闭主通道
+        if self._channel:
             try:
-                target_version = self._connection_version + 1
-                logger.info(
-                    f"尝试创建连接: {self._current_host}:{self.port} "
-                    f"（目标版本: {target_version}，{attempts+1}/{max_attempts}）"
-                )
-                # 创建连接（保留aio-pika原生自动恢复）
-                conn = await connect_robust(
-                    conn_url,
-                    properties={
-                        "connection_name": f"{self.app_name}_conn_v{target_version}",
-                        "product": self.app_name
-                    },
-                    heartbeat=self.heartbeat,
-                    timeout=self.connection_timeout,
-                    reconnect_interval=5,  # 单节点内部短间隔重连（原生自动恢复）
-                    max_reconnect_attempts=3,  # 单节点最大重试3次
-                )
-                logger.info(
-                    f"连接创建成功: {self._current_host}:{self.port}（版本: {target_version}）")
-                return conn
+                if not self._channel.is_closed:
+                    await self._channel.close()
             except Exception as e:
-                attempts += 1
-                last_error = e
-                logger.error(
-                    f"连接节点 {self._current_host}:{self.port} 失败（{attempts}/{max_attempts}）: {str(e)}",
-                    exc_info=True
-                )
-                await asyncio.sleep(min(5 * attempts, self.reconnect_interval))
-
-        raise ConnectionError(
-            f"所有节点创建连接失败（节点列表: {self.hosts}）"
-        ) from last_error
-
-    async def _init_channel_pool(self):
-        """初始化通道池（加锁保证原子性，绑定当前连接）"""
-        async with self._lock:
-            if self._is_shutdown:
-                raise RuntimeError("通道池已关闭，无法初始化")
+                logger.warning(f"⚠️ [CLEANUP_MAIN_CH] 关闭主通道失败: {e}")
+            finally:
+                self._channel = None
 
-            # 校验当前连接有效性
-            if not self._connection or self._connection.is_closed:
-                raise RuntimeError("无有效连接，无法初始化通道池")
-
-            self._free_channels.clear()
-            self._used_channels.clear()
-
-            # 创建指定数量的通道（保留原生自动恢复）
-            for i in range(self.channel_pool_size):
-                try:
-                    channel = await self._connection.channel()  # 新通道自带自动恢复
-                    await channel.set_qos(prefetch_count=self.prefetch_count)
-                    self._free_channels.append(channel)
-                except Exception as e:
-                    logger.error(
-                        f"创建通道失败（第{i+1}个，连接版本: {self._connection_version}）: {str(e)}",
-                        exc_info=True
-                    )
-                    continue
-
-            logger.info(
-                f"通道池初始化完成 - 连接: {self._current_host}:{self.port}（版本: {self._connection_version}）, "
-                f"可用通道数: {len(self._free_channels)}/{self.channel_pool_size}（均带自动恢复）"
-            )
+        # 3. 关闭连接
+        if self._connection:
+            try:
+                if not self._connection.is_closed:
+                    # close() 可能是同步的，也可能是异步的，aio_pika 中通常是异步的
+                    await self._connection.close()
+            except Exception as e:
+                logger.warning(f"⚠️ [CLEANUP_CONN] 关闭连接失败: {e}")
+            finally:
+                self._connection = None
 
-    async def _reconnect_if_needed(self) -> bool:
-        """连接失效时重连（加锁保护，严格单连接+释放旧资源）"""
-        # 快速判断，避免无效加锁
-        if self._is_shutdown or self._reconnecting:
-            return False
+        logger.info("✅ [CLEANUP] 资源清理完成")
 
-        self._reconnecting = True
+    async def _create_connection_impl(self, host: str) -> AbstractRobustConnection:
+        conn_url = (
+            f"amqp://{self.username}:{self.password}@{host}:{self.port}/"
+            f"{self.virtualhost}?name={self.app_name}&heartbeat={self.heartbeat}"
+            f"&reconnect_interval={self.reconnect_interval}&fail_fast=1"
+        )
+        logger.info(f"🔌 [CONNECT] 尝试连接节点: {host}")
         try:
-            logger.warning(f"连接失效（当前版本: {self._connection_version}），开始重连...")
-
-            # 1. 强制释放所有旧资源（加锁保证原子性）
-            await self._safe_close_old_resources()
-
-            # 2. 递增连接版本号（加锁保证原子性，区分新旧连接）
-            async with self._lock:
-                self._connection_version += 1
-                target_version = self._connection_version
-
-            # 3. 创建新连接（保留原生自动恢复）
-            new_conn = await self._create_single_connection()
-
-            # 4. 绑定新连接（加锁保证原子性）
-            async with self._lock:
-                self._connection = new_conn
-
-            # 5. 重新初始化通道池（新通道带自动恢复）
-            await self._init_channel_pool()
-
-            # 6. 恢复消费者通道（新通道带自动恢复）
-            await self._restore_consumer_channels()
-
-            logger.info(f"重连成功（新连接版本: {target_version}），所有通道均带自动恢复")
-            async with self._lock:
-                self._initialized = True  # 重连成功后标记为已初始化
-            return True
+            conn = await asyncio.wait_for(
+                connect_robust(conn_url),
+                timeout=self.connection_timeout + 5
+            )
+            logger.info(f"✅ [CONNECT_OK] 节点连接成功: {host}")
+            return conn
         except Exception as e:
-            logger.error(f"重连失败: {str(e)}", exc_info=True)
-            async with self._lock:
-                self._initialized = False
-            return False
-        finally:
-            self._reconnecting = False
-
-    async def _restore_consumer_channels(self):
-        """重连后恢复消费者通道（加锁保证原子性，新通道带自动恢复）"""
+            logger.error(f"❌ [CONNECT_FAIL] 节点 {host} 连接失败: {str(e)}")
+            raise ConnectionError(f"无法连接RabbitMQ {host}") from e
+
+    async def _ensure_main_channel(self) -> RobustChannel:
+        """
+        确保主通道有效
+        逻辑：
+        1. 检查连接状态
+        2. 如果断开 -> 清理 -> 轮询重试
+        3. 如果连接在但通道断开 -> 仅重建通道
+        """
         async with self._lock:
-            if not self._consumer_channels or not self._connection or self._connection.is_closed:
-                return
-            logger.info(
-                f"开始恢复 {len(self._consumer_channels)} 个消费者通道（连接版本: {self._connection_version}）")
+            if self._is_shutdown:
+                raise RuntimeError("客户端已关闭")
 
-            # 临时保存消费者配置（队列名、回调、auto_ack、参数）
-            consumer_configs = list(self._consumer_channels.items())
-            self._consumer_channels.clear()
+            # --- 阶段A：连接恢复逻辑 (如果连接断了) ---
+            if self._connection is None or self._connection.is_closed:
 
-        # 重新创建消费者（不加锁，避免阻塞其他操作）
-        for queue_name, (_, _, callback, auto_ack, kwargs) in consumer_configs:
-            try:
-                await self.consume_queue(queue_name, callback, auto_ack, **kwargs)
-            except Exception as e:
-                logger.error(
-                    f"恢复消费者队列 {queue_name} 失败: {str(e)}", exc_info=True)
+                # 1. 【强制】先彻底清理所有旧资源
+                await self._cleanup_resources()
 
-    async def _clean_invalid_channels(self):
-        """清理失效通道并补充（加锁保证原子性，仅处理当前连接）"""
-        async with self._lock:
-            if self._is_shutdown or self._reconnecting:
-                return
+                retry_hosts = self.hosts.copy()
+                random.shuffle(retry_hosts)
+                last_error = None
+                max_attempts = min(len(retry_hosts), 3)
 
-            # 1. 校验当前连接有效性
-            current_valid = (
-                self._connection is not None
-                and not self._connection.is_closed
-                and not self._reconnecting
-            )
-            if not current_valid:
-                # 连接失效，触发重连（不加锁，避免死锁）
-                asyncio.create_task(self._reconnect_if_needed())
-                return
+                # 2. 轮询尝试新连接
+                for _ in range(max_attempts):
+                    if not retry_hosts:
+                        break
 
-            # 2. 清理空闲通道（仅保留当前连接的有效通道）
-            valid_free = []
-            for chan in self._free_channels:
-                try:
-                    if not chan.is_closed and chan.connection == self._connection:
-                        valid_free.append(chan)
-                    else:
-                        logger.warning(f"清理失效空闲通道（连接版本不匹配或已关闭）")
-                except Exception:
-                    logger.warning(f"清理异常空闲通道")
-            self._free_channels = valid_free
+                    host = retry_hosts.pop()
+                    self._current_host = host
+                    temp_conn = None
 
-            # 3. 清理使用中通道（仅保留当前连接的有效通道）
-            valid_used = set()
-            for chan in self._used_channels:
-                try:
-                    if not chan.is_closed and chan.connection == self._connection:
-                        valid_used.add(chan)
-                    else:
-                        logger.warning(f"清理失效使用中通道（连接版本不匹配或已关闭）")
-                except Exception:
-                    logger.warning(f"清理异常使用中通道")
-            self._used_channels = valid_used
-
-            # 4. 补充通道到指定大小（新通道带自动恢复）
-            total_valid = len(self._free_channels) + len(self._used_channels)
-            missing = self.channel_pool_size - total_valid
-            if missing > 0:
-                logger.info(
-                    f"通道池缺少{missing}个通道，补充中（连接版本: {self._connection_version}）...")
-                for _ in range(missing):
                     try:
-                        channel = await self._connection.channel()  # 新通道带自动恢复
-                        await channel.set_qos(prefetch_count=self.prefetch_count)
-                        self._free_channels.append(channel)
-                    except Exception as e:
-                        logger.error(f"补充通道失败: {str(e)}", exc_info=True)
+                        temp_conn = await self._create_connection_impl(host)
+
+                        # 3. 只有在连接成功后，才更新 self._connection
+                        self._connection = temp_conn
+                        temp_conn = None  # 转移所有权
+                        self._initialized = True
+                        last_error = None
+                        logger.info(f"🔗 [RECONNECT_OK] 切换到节点: {host}")
                         break
 
+                    except Exception as e:
+                        logger.warning(f"⚠️ [RECONNECT_RETRY] 节点 {host} 不可用")
+                        if temp_conn is not None:
+                            # 尝试连接失败了，必须把这个“半成品”连接关掉
+                            try:
+                                await temp_conn.close()
+                            except Exception:
+                                pass
+                        last_error = e
+                        await asyncio.sleep(self.reconnect_interval)
+
+                # 4. 如果所有尝试都失败
+                if last_error:
+                    # 确保状态是干净的
+                    self._connection = None
+                    self._initialized = False
+                    logger.error("💥 [RECONNECT_FATAL] 所有节点重试失败")
+                    raise ConnectionError("所有 RabbitMQ 节点连接失败") from last_error
+
+            # --- 阶段B：通道恢复逻辑 (如果连接在但通道断了) ---
+            # 注意：这里不需要清理连接，只重置通道
+            if self._channel is None or self._channel.is_closed:
+                try:
+                    self._channel = await self._connection.channel()
+                    await self._channel.set_qos(prefetch_count=self.prefetch_count)
+                    logger.info(f"✅ [CHANNEL_OK] 主通道已恢复")
+                except Exception as e:
+                    # 如果连通道都创建不了，说明这个连接也是坏的，回滚到阶段A
+                    logger.error(f"❌ [CHANNEL_FAIL] 通道创建失败，标记连接无效: {e}")
+                    # 强制清理连接，触发下一次进入阶段A
+                    await self._cleanup_resources()
+                    raise
+
+            return self._channel
+
     async def init_pools(self):
-        """初始化通道池（加锁保证原子性，仅执行一次）"""
+        """初始化入口"""
         async with self._lock:
+            if self._is_shutdown:
+                raise RuntimeError("客户端已关闭")
             if self._initialized:
-                logger.warning("通道池已初始化，无需重复调用")
                 return
-            if self._is_shutdown:
-                raise RuntimeError("通道池已关闭，无法初始化")
 
-        try:
-            # 1. 创建新连接（保留原生自动恢复）
-            new_conn = await self._create_single_connection()
+        # 在 try 之前声明变量，确保 except 块能访问
+        conn_created_in_this_try = None
 
-            # 2. 初始化连接版本号和绑定连接（加锁保证原子性）
-            async with self._lock:
-                self._connection_version += 1
-                self._connection = new_conn
+        try:
+            # 锁外创建连接，减少锁持有时间
+            init_host = random.choice(self.hosts)
+            conn = await self._create_connection_impl(init_host)
 
-            # 3. 初始化通道池（新通道带自动恢复）
-            await self._init_channel_pool()
+            # 记录本次创建的连接
+            conn_created_in_this_try = conn
 
-            # 4. 标记为已初始化（加锁保证原子性）
             async with self._lock:
+                if self._is_shutdown:
+                    raise RuntimeError("客户端已关闭")
+
+                # 提交新资源
+                self._connection = conn
+                self._channel = await self._connection.channel()
+                await self._channel.set_qos(prefetch_count=self.prefetch_count)
                 self._initialized = True
 
-            logger.info("RabbitMQ单连接通道池初始化完成（所有通道均带自动恢复）")
-        except Exception as e:
-            logger.error(f"初始化失败: {str(e)}", exc_info=True)
-            await self._safe_close_old_resources()
-            raise
+                # 所有权转移成功，清空临时引用，防止 finally 重复关闭
+                conn_created_in_this_try = None
 
-    async def acquire_channel(self) -> Tuple[Channel, AbstractRobustConnection]:
-        """获取通道（加锁保证原子性，返回当前连接+带自动恢复的通道）"""
-        # 快速判断，避免无效加锁
-        async with self._lock:
-            if not self._initialized:
-                raise RuntimeError("通道池未初始化，请先调用init_pools()")
-            if self._is_shutdown:
-                raise RuntimeError("通道池已关闭，无法获取通道")
+                logger.info(f"🚀 [INIT_OK] 连接池初始化完成: {init_host}")
 
-        # 先清理失效通道（加锁保证原子性）
-        await self._clean_invalid_channels()
+        except Exception as e:
+            logger.error(f"💥 [INIT_FAIL] 初始化异常: {str(e)}")
 
-        async with self._lock:
-            # 双重校验连接有效性
-            current_valid = (
-                self._connection is not None
-                and not self._connection.is_closed
-                and not self._reconnecting
-            )
-            if not current_valid:
-                # 连接失效，触发重连（不加锁，避免死锁）
-                reconnect_success = await self._reconnect_if_needed()
-                if not reconnect_success:
-                    raise RuntimeError("连接失效且重连失败，无法获取通道")
-
-            # 优先从空闲池获取（带自动恢复的通道）
-            if self._free_channels:
-                channel = self._free_channels.pop()
-                self._used_channels.add(channel)
-                return channel, self._connection
-
-            # 通道池已满，创建临时通道（带自动恢复，用完关闭）
-            try:
-                channel = await self._connection.channel()  # 临时通道带自动恢复
-                await channel.set_qos(prefetch_count=self.prefetch_count)
-                self._used_channels.add(channel)
-                logger.warning(
-                    f"通道池已达上限（{self.channel_pool_size}），创建临时通道（带自动恢复，用完自动关闭）"
-                )
-                return channel, self._connection
-            except Exception as e:
-                logger.error(f"获取通道失败: {str(e)}", exc_info=True)
-                raise
+            # 这里现在可以合法访问 conn_created_in_this_try
+            if conn_created_in_this_try is not None:
+                try:
+                    await conn_created_in_this_try.close()
+                except Exception:
+                    pass
 
-    async def release_channel(self, channel: Channel, conn: AbstractRobustConnection):
-        """释放通道（加锁保证原子性，仅归还当前连接的有效通道）"""
-        # 快速判断，避免无效加锁
-        if not channel or not conn or self._is_shutdown:
-            return
+            if not self._is_shutdown:
+                await self.close()
+            raise
 
+    async def force_reconnect(self):
+        """
+        强制重连
+        严格执行：清理所有资源 -> 尝试建立新资源
+        """
         async with self._lock:
-            # 仅处理当前连接的通道（旧连接的通道直接关闭）
-            if conn != self._connection:
-                try:
-                    if not channel.is_closed:
-                        await channel.close()
-                    logger.warning(f"已关闭非当前连接的通道（版本不匹配，自动恢复终止）")
-                except Exception as e:
-                    logger.warning(f"关闭非当前连接通道失败: {str(e)}")
+            if self._is_shutdown:
                 return
 
-            # 通道不在使用中，直接返回
-            if channel not in self._used_channels:
-                return
+            logger.warning("🔄 [FORCE_RECONNECT] 开始强制重连...")
 
-            # 移除使用中标记
-            self._used_channels.remove(channel)
+            # 1. 【关键】标记未初始化，迫使 _ensure_main_channel 走清理流程
+            self._initialized = False
 
-            # 仅归还有效通道（当前连接有效+通道未关闭+池未满）
-            current_valid = (
-                self._connection is not None
-                and not self._connection.is_closed
-                and not self._reconnecting
-            )
-            if current_valid and not channel.is_closed and len(self._free_channels) < self.channel_pool_size:
-                self._free_channels.append(channel)
-            else:
-                # 无效通道直接关闭（终止自动恢复）
-                try:
-                    if not channel.is_closed:
-                        await channel.close()
-                except Exception as e:
-                    logger.warning(f"关闭通道失败: {str(e)}")
+            # 2. 【关键】立即清理旧资源 (在锁内)
+            await self._cleanup_resources()
 
-    async def declare_queue(self, queue_name: str, **kwargs) -> AbstractQueue:
-        """声明队列（使用池内通道，带自动恢复）"""
-        channel, conn = await self.acquire_channel()
-        try:
-            return await channel.declare_queue(queue_name, **kwargs)
-        finally:
-            await self.release_channel(channel, conn)
+            # 此时 self._connection 和 self._channel 均为 None
 
-    async def declare_exchange(self, exchange_name: str, exchange_type: str = "direct", **kwargs) -> AbstractExchange:
-        """声明交换机（使用池内通道，带自动恢复）"""
-        channel, conn = await self.acquire_channel()
+        # 3. 锁外触发恢复 (避免阻塞锁太久)
         try:
-            return await channel.declare_exchange(exchange_name, exchange_type, **kwargs)
-        finally:
-            await self.release_channel(channel, conn)
+            await self.acquire_channel()
+            logger.info("✅ [FORCE_RECONNECT_OK] 强制重连成功")
+        except Exception as e:
+            logger.error(f"❌ [FORCE_RECONNECT_FAIL] 强制重连失败: {e}")
+            raise
+
+    async def acquire_channel(self) -> Tuple[RobustChannel, AbstractRobustConnection]:
+        """获取主通道"""
+        if not self._initialized and not self._is_shutdown:
+            await self.init_pools()
+        return await self._ensure_main_channel(), self._connection
 
     async def publish_message(self, routing_key: str, message_body: bytes, exchange_name: str = "", **kwargs):
-        """发布消息（使用池内通道，带自动恢复）"""
-        channel, conn = await self.acquire_channel()
+        channel, _ = await self.acquire_channel()
         try:
             exchange = channel.default_exchange if not exchange_name else await channel.get_exchange(exchange_name)
             message = Message(body=message_body, **kwargs)
             await exchange.publish(message, routing_key=routing_key)
-            logger.debug(
-                f"消息发布成功 - 连接: {self._current_host}:{self.port}（版本: {self._connection_version}）, "
-                f"交换机: {exchange.name}, 路由键: {routing_key}"
-            )
         except Exception as e:
-            logger.error(f"发布消息失败: {str(e)}", exc_info=True)
-            # 发布失败触发重连（下次使用新通道）
-            asyncio.create_task(self._reconnect_if_needed())
+            logger.error(f"❌ [PUBLISH_FAIL] 发布失败: {str(e)}")
             raise
-        finally:
-            await self.release_channel(channel, conn)
 
     async def consume_queue(self, queue_name: str, callback: Callable[[AbstractMessage], asyncio.Future], auto_ack: bool = False, **kwargs):
-        """消费队列（使用独立通道，带自动恢复，支持多消费者）"""
-        # 快速判断，避免无效加锁
+        if not self._initialized:
+            await self.init_pools()
+
+        # 检查是否已存在
         async with self._lock:
-            if not self._initialized:
-                raise RuntimeError("通道池未初始化，请先调用init_pools()")
             if self._is_shutdown:
-                raise RuntimeError("通道池已关闭，无法启动消费")
+                raise RuntimeError("客户端已关闭")
+            if queue_name in self._consumer_channels:
+                logger.warning(f"⚠️ [CONSUMER_EXISTS] 队列 {queue_name} 已在消费中")
+                return
+            if not self._connection or self._connection.is_closed:
+                raise RuntimeError("连接不可用，无法启动消费")
 
-        # 先声明队列（确保队列存在）
+        # 声明队列 (使用主通道)
         await self.declare_queue(queue_name, **kwargs)
 
-        # 获取独立通道（消费者通道不放入普通池）
-        channel, conn = await self.acquire_channel()
+        try:
+            # 获取最新连接
+            _, conn = await self.acquire_channel()
 
-        # 注册消费者通道（加锁保证原子性）
-        async with self._lock:
-            self._consumer_channels[queue_name] = (
-                channel, conn, callback, auto_ack, kwargs)
+            # 创建消费者通道
+            consumer_channel = await conn.channel()
+            await consumer_channel.set_qos(prefetch_count=self.prefetch_count)
 
-        async def consume_callback_wrapper(message: AbstractMessage):
-            """消费回调包装（处理通道失效和自动恢复）"""
-            try:
-                # 加锁校验通道和连接状态（原子性）
-                async with self._lock:
-                    conn_matched = conn == self._connection
-                    channel_valid = not channel.is_closed
-                    current_conn_valid = (
-                        self._connection is not None
-                        and not self._connection.is_closed
-                        and not self._reconnecting
-                    )
-
-                # 通道/连接失效，触发重连恢复
-                if not conn_matched or not channel_valid or not current_conn_valid:
-                    logger.warning(
-                        f"消费者通道 {queue_name} 失效（连接版本不匹配/通道关闭），触发重连恢复")
-
-                    # 移除旧消费者记录（加锁保证原子性）
-                    async with self._lock:
-                        if self._consumer_channels.get(queue_name) == (channel, conn, callback, auto_ack, kwargs):
-                            del self._consumer_channels[queue_name]
-
-                    # 释放旧通道（加锁保证原子性）
-                    await self.release_channel(channel, conn)
-
-                    # 重新创建消费者（新通道带自动恢复）
-                    asyncio.create_task(self.consume_queue(
-                        queue_name, callback, auto_ack, **kwargs))
-
-                    # Nack消息（避免丢失）
-                    if not auto_ack:
-                        await message.nack(requeue=True)
+            async with self._lock:
+                # 再次检查，防止并发创建
+                if self._is_shutdown:
+                    await consumer_channel.close()
+                    return
+                if queue_name in self._consumer_channels:
+                    await consumer_channel.close()  # 其他协程已经创建了
                     return
 
-                # 执行业务回调
-                await callback(message)
-                if not auto_ack:
-                    await message.ack()
-            except ChannelClosed as e:
-                logger.error(f"消费者通道 {queue_name} 关闭: {str(e)}", exc_info=True)
-                if not auto_ack:
-                    await message.nack(requeue=True)
-                asyncio.create_task(self._reconnect_if_needed())
-            except aiormq.exceptions.ChannelInvalidStateError as e:
-                logger.error(
-                    f"消费者通道 {queue_name} 状态异常: {str(e)}", exc_info=True)
-                if not auto_ack:
-                    await message.nack(requeue=True)
-                asyncio.create_task(self.consume_queue(
-                    queue_name, callback, auto_ack, **kwargs))
-            except Exception as e:
-                logger.error(
-                    f"消费消息失败（队列: {queue_name}）: {str(e)}", exc_info=True)
-                if not auto_ack:
-                    await message.nack(requeue=True)
-
-        # 日志输出（加锁获取当前连接信息）
-        async with self._lock:
-            current_host = self._current_host
-            current_version = self._connection_version
+                self._consumer_channels[queue_name] = consumer_channel
 
-        logger.info(
-            f"开始消费队列: {queue_name} - 连接: {current_host}:{self.port}（版本: {current_version}）, "
-            f"通道带自动恢复"
-        )
+            async def consume_callback_wrapper(message: AbstractMessage):
+                try:
+                    await callback(message)
+                    if not auto_ack:
+                        await message.ack()
+                except Exception as e:
+                    logger.error(f"❌ [CALLBACK_ERR] {queue_name}: {e}")
+                    if not auto_ack:
+                        await message.nack(requeue=True)
 
-        try:
-            # 启动消费（使用带自动恢复的通道）
-            await channel.basic_consume(
-                queue_name,
-                consumer_callback=consume_callback_wrapper,
-                auto_ack=auto_ack,
-                **kwargs
+            await consumer_channel.basic_consume(
+                queue_name, consumer_callback=consume_callback_wrapper, auto_ack=auto_ack
             )
+            logger.info(f"🎧 [CONSUME_START] {queue_name}")
+
         except Exception as e:
-            logger.error(f"启动消费失败（队列: {queue_name}）: {str(e)}", exc_info=True)
-            # 清理异常的消费者记录和通道（加锁保证原子性）
-            await self.release_channel(channel, conn)
+            logger.error(f"💥 [CONSUME_ERR] {queue_name}: {e}")
+            # 失败时清理字典
             async with self._lock:
-                if self._consumer_channels.get(queue_name) == (channel, conn, callback, auto_ack, kwargs):
-                    del self._consumer_channels[queue_name]
+                if queue_name in self._consumer_channels:
+                    # 注意：这里清理的是字典里的引用，通道本身应该在 try 块里被关闭了吗？
+                    # 如果 consumer_channel 创建成功但 basic_consume 失败，需要手动关闭
+                    ch = self._consumer_channels.pop(queue_name, None)
+                    if ch:
+                        try:
+                            await ch.close()
+                        except:
+                            pass
             raise
 
     async def close(self):
-        """关闭通道池（加锁保证原子性，释放所有资源）"""
+        """资源销毁"""
         async with self._lock:
             if self._is_shutdown:
-                logger.warning("通道池已关闭，无需重复操作")
                 return
             self._is_shutdown = True
+            self._initialized = False
+
+        logger.info("🛑 [CLOSE] 开始关闭连接池...")
+
+        # 1. 清理所有资源
+        await self._cleanup_resources()
 
-        logger.info("开始关闭RabbitMQ单连接通道池...")
-        # 强制释放所有资源（包括自动恢复的通道）
-        await self._safe_close_old_resources()
-        logger.info("RabbitMQ单连接通道池已完全关闭（所有自动恢复逻辑已终止）")
+        logger.info("🏁 [CLOSE] 连接池已关闭")
+
+    async def declare_queue(self, queue_name: str, **kwargs) -> AbstractQueue:
+        channel, _ = await self.acquire_channel()
+        return await channel.declare_queue(queue_name, **kwargs)
+
+    async def declare_exchange(self, exchange_name: str, exchange_type: str = "direct", **kwargs) -> AbstractExchange:
+        channel, _ = await self.acquire_channel()
+        return await channel.declare_exchange(exchange_name, exchange_type, **kwargs)