sycommon-python-lib 0.1.21__py3-none-any.whl → 0.1.46__py3-none-any.whl

sycommon/rabbitmq/rabbitmq_client.py

@@ -1,30 +1,33 @@
+from aio_pika import Channel
+from typing import Optional
 import asyncio
 import json
-from typing import Callable, Coroutine, Optional, Dict, Any, Union, Set
+from typing import Callable, Coroutine, Dict, Any, Union
 from aio_pika import Message, DeliveryMode, ExchangeType
 from aio_pika.abc import (
-    AbstractChannel,
     AbstractExchange,
     AbstractQueue,
     AbstractIncomingMessage,
     ConsumerTag,
+    AbstractRobustConnection,
 )
-from aiormq.exceptions import ChannelInvalidStateError, ConnectionClosed
-
+from sycommon.rabbitmq.rabbitmq_pool import RabbitMQConnectionPool
 from sycommon.logging.kafka_log import SYLogger
 from sycommon.models.mqmsg_model import MQMsgModel
-from sycommon.rabbitmq.rabbitmq_pool import RabbitMQConnectionPool
 
-# 最大重试次数限制
-MAX_RETRY_COUNT = 3
 
 logger = SYLogger
 
 
 class RabbitMQClient:
     """
-    RabbitMQ客户端（基于连接池），支持集群多节点配置
-    提供自动故障转移、连接恢复和消息可靠性保障
+    RabbitMQ 客户端（支持消息发布、消费、自动重连、异常重试）
+    核心特性：
+    1. 基于连接池复用资源，性能优化
+    2. 连接/通道失效时自动重建，高可用（限制并发重连）
+    3. 消息发布支持重试+mandatory机制+超时控制，确保路由有效
+    4. 消费支持手动ACK/NACK
+    5. 兼容JSON/字符串/字典消息格式
     """
 
     def __init__(
@@ -38,642 +41,432 @@ class RabbitMQClient:
         auto_delete: bool = False,
         auto_parse_json: bool = True,
         create_if_not_exists: bool = True,
-        connection_timeout: int = 10,
-        rpc_timeout: int = 10,
-        reconnection_delay: int = 1,
-        max_reconnection_attempts: int = 5,
         prefetch_count: int = 2,
-        consumption_stall_threshold: int = 10
+        **kwargs,
     ):
-        """
-        初始化RabbitMQ客户端（依赖连接池）
-
-        :param connection_pool: 连接池实例
-        """
-        # 连接池依赖
+        # 依赖注入：连接池（必须已初始化）
         self.connection_pool = connection_pool
+        if not self.connection_pool._initialized:
+            raise RuntimeError("连接池未初始化，请先调用 connection_pool.init_pools()")
 
-        # 交换器和队列参数
-        self.exchange_name = exchange_name
-        self.exchange_type = ExchangeType(exchange_type)
-        self.queue_name = queue_name
-        self.routing_key = routing_key
-        self.durable = durable
-        self.auto_delete = auto_delete
-
-        # 行为控制参数
-        self.auto_parse_json = auto_parse_json
-        self.create_if_not_exists = create_if_not_exists
-        self.connection_timeout = connection_timeout
-        self.rpc_timeout = rpc_timeout
-        self.prefetch_count = prefetch_count
-
-        # 重连参数
-        self.reconnection_delay = reconnection_delay
-        self.max_reconnection_attempts = max_reconnection_attempts
-
-        # 消息处理参数
-        self.consumption_stall_threshold = consumption_stall_threshold
-
-        # 通道和资源对象（从池获取）
-        self.channel: Optional[AbstractChannel] = None
-        self.exchange: Optional[AbstractExchange] = None
-        self.queue: Optional[AbstractQueue] = None
-
-        # 状态跟踪
-        self.actual_queue_name: Optional[str] = None
-        self._exchange_exists = False
-        self._queue_exists = False
-        self._queue_bound = False
-        self._is_consuming = False
-        self._closed = False
+        # 交换机配置
+        self.exchange_name = exchange_name.strip()
+        try:
+            self.exchange_type = ExchangeType(exchange_type.lower())
+        except ValueError:
+            SYLogger.warning(f"无效的exchange_type: {exchange_type}，默认使用'topic'")
+            self.exchange_type = ExchangeType.topic
+
+        # 队列配置
+        self.queue_name = queue_name.strip() if queue_name else None
+        self.routing_key = routing_key.strip() if routing_key else "#"
+        self.durable = durable  # 消息/队列持久化
+        self.auto_delete = auto_delete  # 无消费者时自动删除队列/交换机
+        self.auto_parse_json = auto_parse_json  # 自动解析JSON消息体
+        self.create_if_not_exists = create_if_not_exists  # 不存在则创建交换机/队列
+
+        # 消费配置
+        self.prefetch_count = max(1, prefetch_count)  # 每次预取消息数（避免消息堆积）
+
+        # 内部状态（资源+连接）
+        self._channel: Optional[Channel] = None
+        self._channel_conn: Optional[AbstractRobustConnection] = None  # 通道所属连接
+        self._exchange: Optional[AbstractExchange] = None
+        self._queue: Optional[AbstractQueue] = None
         self._consumer_tag: Optional[ConsumerTag] = None
-        self._last_activity_timestamp = asyncio.get_event_loop().time()
-        self._last_message_processed = asyncio.get_event_loop().time()
-
-        # 任务和处理器
-        self.message_handler: Optional[Callable[
-            [Union[Dict[str, Any], str], AbstractIncomingMessage],
-            Coroutine[Any, Any, None]
-        ]] = None
-        self._consuming_task: Optional[asyncio.Task] = None
-        self._reconnect_task: Optional[asyncio.Task] = None
-        self._keepalive_task: Optional[asyncio.Task] = None
-        self._monitor_task: Optional[asyncio.Task] = None
-
-        # 消息处理跟踪
-        self._processing_message_ids: Set[str] = set()
+        self._message_handler: Optional[Callable[[
+            MQMsgModel, AbstractIncomingMessage], Coroutine[Any, Any, None]]] = None
+        self._closed = False
 
-    @property
-    def is_connected(self) -> bool:
-        """检查当前通道是否有效"""
-        return (not self._closed and
-                self.channel is not None and
-                not self.channel.is_closed and
-                self.exchange is not None)
-
-    def _update_activity_timestamp(self) -> None:
-        """更新最后活动时间戳"""
-        self._last_activity_timestamp = asyncio.get_event_loop().time()
-
-    def _update_message_processed_timestamp(self) -> None:
-        """更新最后消息处理时间戳"""
-        self._last_message_processed = asyncio.get_event_loop().time()
-
-    async def _get_channel(self) -> AbstractChannel:
-        """从通道池获取通道（使用上下文管理器）"""
-        if not self.connection_pool.channel_pool:
-            raise Exception("连接池未初始化，请先调用init_pools")
-
-        # 使用async with获取通道，并通过变量返回
-        async with self.connection_pool.channel_pool.acquire() as channel:
-            return channel
-
-    async def _check_exchange_exists(self, channel: AbstractChannel) -> bool:
-        """检查交换机是否存在"""
-        try:
-            # 使用被动模式检查交换机
-            await asyncio.wait_for(
-                channel.declare_exchange(
-                    name=self.exchange_name,
-                    type=self.exchange_type,
-                    passive=True
-                ),
-                timeout=self.rpc_timeout
-            )
-            return True
-        except Exception:
-            return False
+        # 线程安全锁
+        self._consume_lock = asyncio.Lock()
+        self._connect_lock = asyncio.Lock()
+        # 跟踪连接关闭回调（用于后续移除）
+        self._conn_close_callback: Optional[Callable] = None
+        # 控制重连频率的信号量（限制并发重连数，默认1个）
+        self._reconnect_semaphore = asyncio.Semaphore(1)
+        # 固定重连间隔15秒（全局统一）
+        self._RECONNECT_INTERVAL = 15
+        # 重连任务锁（确保同一时间只有一个重连任务）
+        self._reconnect_task_lock = asyncio.Lock()
+        # 跟踪当前重连任务（避免重复创建）
+        self._current_reconnect_task: Optional[asyncio.Task] = None
+        # 连接失败计数器（用于告警）
+        self._reconnect_fail_count = 0
+        # 连接失败告警阈值
+        self._reconnect_alert_threshold = 5
 
-    async def _check_queue_exists(self, channel: AbstractChannel) -> bool:
-        """检查队列是否存在"""
-        if not self.queue_name:
+    @property
+    async def is_connected(self) -> bool:
+        """异步检查客户端连接状态（属性，不可调用）"""
+        if self._closed:
             return False
         try:
-            # 使用被动模式检查队列
-            await asyncio.wait_for(
-                channel.declare_queue(
-                    name=self.queue_name,
-                    passive=True
-                ),
-                timeout=self.rpc_timeout
+            # 校验通道+连接+核心资源都有效
+            return (
+                self._channel and not self._channel.is_closed
+                and self._channel_conn and not self._channel_conn.is_closed
+                and self._exchange is not None
+                and (not self.queue_name or self._queue is not None)
             )
-            return True
-        except Exception:
+        except Exception as e:
+            SYLogger.warning(f"检查连接状态失败: {str(e)}")
             return False
 
-    async def _bind_queue(self, channel: AbstractChannel, queue: AbstractQueue, exchange: AbstractExchange) -> bool:
-        """将队列绑定到交换机"""
-        bind_routing_key = self.routing_key if self.routing_key else '#'
+    async def connect(self) -> None:
+        if self._closed:
+            raise RuntimeError("客户端已关闭，无法重新连接")
 
-        for attempt in range(MAX_RETRY_COUNT + 1):
-            try:
-                await asyncio.wait_for(
-                    queue.bind(
-                        exchange,
-                        routing_key=bind_routing_key
-                    ),
-                    timeout=self.rpc_timeout
-                )
-                logger.info(
-                    f"队列 '{self.queue_name}' 已绑定到交换机 '{self.exchange_name}'，路由键: {bind_routing_key}")
-                return True
-            except Exception as e:
-                logger.warning(
-                    f"队列绑定失败（第{attempt+1}次尝试）: {str(e)}")
-            if attempt < MAX_RETRY_COUNT:
-                await asyncio.sleep(1)
-        return False
-
-    async def connect(self, force_reconnect: bool = False, declare_queue: bool = True) -> None:
-        """
-        从连接池获取资源并初始化（交换机、队列）
-        """
-        logger.debug(
-            f"连接参数 - force_reconnect={force_reconnect}, "
-            f"declare_queue={declare_queue}, create_if_not_exists={self.create_if_not_exists}"
-        )
-
-        # 如果已连接且不强制重连，则直接返回
-        if self.is_connected and not force_reconnect:
-            return
-
-        # 取消正在进行的重连任务
-        if self._reconnect_task and not self._reconnect_task.done():
-            self._reconnect_task.cancel()
-
-        # 重置状态
-        self._exchange_exists = False
-        self._queue_exists = False
-        self._queue_bound = False
-
-        retries = 0
-        last_exception = None
+        async with self._connect_lock:
+            # 释放旧资源（保留原有回调清理逻辑）
+            if self._channel and self._channel_conn:
+                try:
+                    if self._conn_close_callback and self._channel_conn:
+                        self._channel_conn.close_callbacks.discard(
+                            self._conn_close_callback)
+                    await self.connection_pool.release_channel(self._channel, self._channel_conn)
+                except Exception as e:
+                    SYLogger.warning(f"释放旧通道失败: {str(e)}")
+            self._channel = None
+            self._channel_conn = None
+            self._exchange = None
+            self._queue = None
+            self._conn_close_callback = None
 
-        while retries < self.max_reconnection_attempts:
             try:
-                # 从池获取通道
-                self.channel = await self._get_channel()
-                await self.channel.set_qos(prefetch_count=self.prefetch_count)
-
-                # 处理交换机
-                exchange_exists = await self._check_exchange_exists(self.channel)
-                if not exchange_exists:
-                    if self.create_if_not_exists:
-                        # 创建交换机
-                        self.exchange = await asyncio.wait_for(
-                            self.channel.declare_exchange(
-                                name=self.exchange_name,
-                                type=self.exchange_type,
-                                durable=self.durable,
-                                auto_delete=self.auto_delete
-                            ),
-                            timeout=self.rpc_timeout
-                        )
-                        logger.info(f"已创建交换机 '{self.exchange_name}'")
-                    else:
-                        raise Exception(
-                            f"交换机 '{self.exchange_name}' 不存在且不允许自动创建")
-                else:
-                    # 获取已有交换机
-                    self.exchange = await self.channel.get_exchange(self.exchange_name)
-                    logger.info(f"使用已存在的交换机 '{self.exchange_name}'")
-
-                # 处理队列
-                if declare_queue and self.queue_name:
-                    queue_exists = await self._check_queue_exists(self.channel)
-
-                    if not queue_exists:
-                        if not self.create_if_not_exists:
-                            raise Exception(
-                                f"队列 '{self.queue_name}' 不存在且不允许自动创建")
-
-                        # 创建队列
-                        self.queue = await asyncio.wait_for(
-                            self.channel.declare_queue(
-                                name=self.queue_name,
-                                durable=self.durable,
-                                auto_delete=self.auto_delete,
-                                exclusive=False
-                            ),
-                            timeout=self.rpc_timeout
-                        )
-                        self.actual_queue_name = self.queue_name
-                        logger.info(f"已创建队列 '{self.queue_name}'")
-                    else:
-                        # 获取已有队列
-                        self.queue = await self.channel.get_queue(self.queue_name)
-                        self.actual_queue_name = self.queue_name
-                        logger.info(f"使用已存在的队列 '{self.queue_name}'")
-
+                # 从连接池获取通道+连接（连接池返回的是 RobustChannel）
+                self._channel, self._channel_conn = await self.connection_pool.acquire_channel()
+
+                def on_conn_closed(conn: AbstractRobustConnection, exc: Optional[BaseException]):
+                    """连接关闭回调：触发固定间隔重连"""
+                    SYLogger.warning(
+                        f"客户端连接关闭: {conn!r}，原因: {exc}", exc_info=exc)
+                    self._reconnect_fail_count += 1
+                    # 超过阈值告警
+                    if self._reconnect_fail_count >= self._reconnect_alert_threshold:
+                        SYLogger.error(
+                            f"连接失败次数已达阈值({self._reconnect_alert_threshold})，请检查MQ服务状态")
+                    if not self._closed:
+                        asyncio.create_task(self._safe_reconnect())
+
+                self._conn_close_callback = on_conn_closed
+                if self._channel_conn:
+                    self._channel_conn.close_callbacks.add(
+                        self._conn_close_callback)
+
+                # 2. 设置预取计数（限流）
+                await self._channel.set_qos(prefetch_count=self.prefetch_count)
+                SYLogger.debug(f"设置预取计数: {self.prefetch_count}")
+
+                # 3. 低版本 RobustChannel 说明：默认启用异步发布确认，无显式确认方法
+                SYLogger.debug(
+                    "基于 RobustChannel 异步发布确认（低版本 aio-pika 不支持显式确认方法）")
+
+                # 4. 声明交换机
+                self._exchange = await self._channel.declare_exchange(
+                    name=self.exchange_name,
+                    type=self.exchange_type,
+                    durable=self.durable,
+                    auto_delete=self.auto_delete,
+                    passive=not self.create_if_not_exists,  # passive=True时，不存在则报错
+                )
+                SYLogger.info(
+                    f"交换机初始化成功: {self.exchange_name}（类型: {self.exchange_type.value}）")
+
+                # 5. 声明队列（如果配置了队列名）
+                if self.queue_name:
+                    self._queue = await self._channel.declare_queue(
+                        name=self.queue_name,
+                        durable=self.durable,
+                        auto_delete=self.auto_delete,
+                        passive=not self.create_if_not_exists,
+                    )
                     # 绑定队列到交换机
-                    if self.queue and self.exchange:
-                        bound = await self._bind_queue(self.channel, self.queue, self.exchange)
-                        if not bound:
-                            raise Exception(f"队列 '{self.queue_name}' 绑定到交换机失败")
-                else:
-                    # 不声明队列时的状态处理
-                    self.queue = None
-                    self.actual_queue_name = None
-                    logger.debug(f"跳过队列 '{self.queue_name}' 的声明和绑定")
-
-                # 验证连接状态
-                if not self.is_connected:
-                    raise Exception("连接验证失败，状态异常")
-
-                # 如果之前在消费，重新开始消费
-                if self._is_consuming and self.message_handler:
-                    await self.start_consuming()
-
-                # 启动连接监控和保活任务
-                self._start_monitoring()
-                self._start_keepalive()
-
-                self._update_activity_timestamp()
-                logger.info(f"RabbitMQ客户端初始化成功 (队列: {self.actual_queue_name})")
-                return
+                    await self._queue.bind(
+                        exchange=self._exchange,
+                        routing_key=self.routing_key,
+                    )
+                    SYLogger.info(
+                        f"队列初始化成功: {self.queue_name} "
+                        f"（绑定交换机: {self.exchange_name}, routing_key: {self.routing_key}）"
+                    )
 
+                # 重连成功，重置失败计数器
+                self._reconnect_fail_count = 0
+                SYLogger.info("客户端连接初始化完成")
             except Exception as e:
-                last_exception = e
-                logger.warning(f"资源初始化失败: {str(e)}，重试中...")
-                # 释放当前通道（放回池并重新获取）
-                self.channel = None
-                retries += 1
-                if retries < self.max_reconnection_attempts:
-                    await asyncio.sleep(self.reconnection_delay)
-
-        logger.error(f"最终初始化失败: {str(last_exception)}")
-        raise Exception(
-            f"经过{self.max_reconnection_attempts}次重试后仍无法初始化客户端。最后错误: {str(last_exception)}")
-
-    def _start_monitoring(self) -> None:
-        """启动连接和消费监控任务"""
-        if self._closed or (self._monitor_task and not self._monitor_task.done()):
-            return
-
-        async def monitor():
-            while not self._closed and self.channel:
-                try:
-                    # 检查通道状态
-                    if self.channel.is_closed:
-                        logger.warning("检测到通道已关闭，尝试重建")
-                        await self._recreate_channel()
-                        continue
-
-                    # 检查消费停滞
-                    if self._is_consuming:
-                        current_time = asyncio.get_event_loop().time()
-                        if current_time - self._last_message_processed > self.consumption_stall_threshold:
-                            if self._processing_message_ids:
-                                logger.warning(
-                                    f"消费停滞，但有 {len(self._processing_message_ids)} 个消息正在处理，暂不重启")
-                            else:
-                                # 确实无消息处理，再重启
-                                await self.stop_consuming()
-                                await asyncio.sleep(1)
-                                await self.start_consuming()
-                except Exception as e:
-                    logger.error(f"监控任务出错: {str(e)}")
-                    await asyncio.sleep(1)
-
-                await asyncio.sleep(30)
-
-        self._monitor_task = asyncio.create_task(monitor())
+                SYLogger.error(f"客户端连接失败: {str(e)}", exc_info=True)
+                # 清理异常状态
+                if self._conn_close_callback and self._channel_conn:
+                    self._channel_conn.close_callbacks.discard(
+                        self._conn_close_callback)
+                if self._channel and self._channel_conn:
+                    try:
+                        await self.connection_pool.release_channel(self._channel, self._channel_conn)
+                    except:
+                        pass
+                self._channel = None
+                self._channel_conn = None
+                # 触发重连（固定间隔）
+                if not self._closed:
+                    asyncio.create_task(self._safe_reconnect())
+                raise
+
+    async def _safe_reconnect(self):
+        """安全重连：信号量控制并发+固定15秒间隔，避免短时间大量重连"""
+        # 1. 信号量控制：限制同时进行的重连任务数（默认1个）
+        async with self._reconnect_semaphore:
+            # 2. 检查是否已有重连任务在运行（双重保障）
+            if self._current_reconnect_task and not self._current_reconnect_task.done():
+                SYLogger.debug("已有重连任务在运行，跳过重复触发")
+                return
 
-    async def _recreate_channel(self) -> None:
-        try:
-            # 无需手动释放，上下文管理器会自动处理
-            self.channel = await self._get_channel()
-            await self.channel.set_qos(prefetch_count=self.prefetch_count)
+            async with self._reconnect_task_lock:
+                if self._closed or await self.is_connected:
+                    SYLogger.debug("客户端已关闭或已连接，取消重连")
+                    return
 
-            # 重新获取交换机
-            self.exchange = await self.channel.get_exchange(self.exchange_name)
+                # 3. 固定15秒重连间隔（避免频繁重试）
+                SYLogger.info(f"将在15秒后尝试重连...")
+                await asyncio.sleep(self._RECONNECT_INTERVAL)
 
-            # 重新绑定队列
-            if self.queue_name:
-                self.queue = await self.channel.get_queue(self.queue_name)
-                if self.queue and self.exchange:
-                    await self._bind_queue(self.channel, self.queue, self.exchange)
+                if self._closed or await self.is_connected:
+                    SYLogger.debug("重连等待期间客户端状态变化，取消重连")
+                    return
 
-            # 重新开始消费
-            if self._is_consuming and self.message_handler:
-                await self.start_consuming()
+                try:
+                    # 4. 执行重连
+                    SYLogger.info("开始重连RabbitMQ客户端...")
+                    self._current_reconnect_task = asyncio.create_task(
+                        self.connect())
+                    await self._current_reconnect_task
+                except Exception as e:
+                    SYLogger.warning(f"重连失败: {str(e)}")
+                    # 重连失败后，不主动触发下一次（等待连接关闭回调再次触发，避免死循环）
+                finally:
+                    self._current_reconnect_task = None
 
-            logger.info("通道已重建并恢复服务")
-            self._update_activity_timestamp()
-        except Exception as e:
-            logger.error(f"通道重建失败: {str(e)}，触发重连")
-            await self.connect(force_reconnect=True)
-
-    def _start_keepalive(self) -> None:
-        """启动连接保活任务"""
-        if self._closed or (self._keepalive_task and not self._keepalive_task.done()):
-            return
-
-        async def keepalive():
-            while not self._closed and self.is_connected:
-                current_time = asyncio.get_event_loop().time()
-                # 检查是否超过指定时间无活动
-                if current_time - self._last_activity_timestamp > self.connection_pool.heartbeat * 1.5:
-                    logger.debug(
-                        f"连接 {self.connection_pool.heartbeat*1.5}s 无活动，执行保活检查")
-                    try:
-                        if self.channel.is_closed:
-                            logger.warning("连接已关闭，触发重连")
-                            await self.connect(force_reconnect=True)
+    async def set_message_handler(
+        self,
+        handler: Callable[[MQMsgModel, AbstractIncomingMessage], Coroutine[Any, Any, None]],
+    ) -> None:
+        """设置消息处理器（必须是协程函数）"""
+        if not asyncio.iscoroutinefunction(handler):
+            raise TypeError("消息处理器必须是协程函数（使用 async def 定义）")
+
+        async with self._consume_lock:
+            self._message_handler = handler
+            SYLogger.info("消息处理器设置成功")
+
+    async def start_consuming(self) -> Optional[ConsumerTag]:
+        """启动消息消费（支持自动重连）"""
+        if self._closed:
+            raise RuntimeError("客户端已关闭，无法启动消费")
+
+        async with self._consume_lock:
+            # 1. 校验前置条件
+            if not self._message_handler:
+                raise RuntimeError("未设置消息处理器，请先调用 set_message_handler()")
+            if not await self.is_connected:
+                await self.connect()
+            if not self._queue:
+                raise RuntimeError("未配置队列名，无法启动消费")
+
+            # 2. 定义消费回调（包含异常处理和重连逻辑）
+            async def consume_callback(message: AbstractIncomingMessage):
+                try:
+                    # 解析消息体
+                    if self.auto_parse_json:
+                        try:
+                            body_dict = json.loads(
+                                message.body.decode("utf-8"))
+                            msg_obj = MQMsgModel(**body_dict)
+                        except json.JSONDecodeError as e:
+                            SYLogger.error(
+                                f"JSON消息解析失败: {str(e)}，消息体: {message.body[:100]}...")
+                            await message.nack(requeue=False)  # 解析失败，不重入队
                             return
-
-                        # 轻量级操作保持连接活跃
-                        await asyncio.wait_for(
-                            self.channel.declare_exchange(
-                                name=self.exchange_name,
-                                type=self.exchange_type,
-                                passive=True
-                            ),
-                            timeout=5
+                    else:
+                        msg_obj = MQMsgModel(
+                            body=message.body.decode("utf-8"),
+                            routing_key=message.routing_key,
+                            delivery_tag=message.delivery_tag,
                         )
-                        self._update_activity_timestamp()
-                    except Exception as e:
-                        logger.warning(f"保活检查失败: {str(e)}，触发重连")
-                        await self.connect(force_reconnect=True)
-
-                await asyncio.sleep(self.connection_pool.heartbeat / 2)
-
-        self._keepalive_task = asyncio.create_task(keepalive())
 
-    async def _schedule_reconnect(self) -> None:
-        """安排重新连接"""
-        if self._reconnect_task and not self._reconnect_task.done():
-            return
-
-        logger.info(f"将在 {self.reconnection_delay} 秒后尝试重新连接...")
-
-        async def reconnect():
-            try:
-                await asyncio.sleep(self.reconnection_delay)
-                if not self._closed:
-                    await self.connect(force_reconnect=True)
-            except Exception as e:
-                logger.error(f"重连任务失败: {str(e)}")
-                if not self._closed:
-                    await self._schedule_reconnect()
+                    # 调用消息处理器（必须await，避免协程未等待警告）
+                    await self._message_handler(msg_obj, message)
 
-        self._reconnect_task = asyncio.create_task(reconnect())
+                    # 手动ACK（消息处理成功）
+                    await message.ack()
+                    SYLogger.debug(
+                        f"消息处理成功，delivery_tag: {message.delivery_tag}")
 
-    async def close(self) -> None:
-        """关闭客户端并释放资源"""
-        self._closed = True
-        self._is_consuming = False
+                except Exception as e:
+                    SYLogger.error(
+                        f"消息处理失败，delivery_tag: {message.delivery_tag}",
+                        exc_info=True
+                    )
+                    # 处理失败逻辑：首次失败重入队，再次失败丢弃
+                    if message.redelivered:
+                        SYLogger.warning(
+                            f"消息已重入队过，本次拒绝入队: {message.delivery_tag}")
+                        await message.reject(requeue=False)
+                    else:
+                        SYLogger.warning(f"消息重入队: {message.delivery_tag}")
+                        await message.nack(requeue=True)
+
+                    # 检查连接状态，失效则触发重连
+                    if not await self.is_connected:
+                        SYLogger.warning("连接已失效，触发客户端重连")
+                        asyncio.create_task(self._safe_reconnect())
+
+            # 3. 启动消费
+            self._consumer_tag = await self._queue.consume(consume_callback)
+            SYLogger.info(
+                f"开始消费队列: {self._queue.name}，consumer_tag: {self._consumer_tag}"
+            )
+            return self._consumer_tag
 
-        # 取消所有任务
-        for task in [self._keepalive_task, self._reconnect_task,
-                     self._consuming_task, self._monitor_task]:
-            if task and not task.done():
-                task.cancel()
+    async def stop_consuming(self) -> None:
+        """停止消息消费"""
+        async with self._consume_lock:
+            if self._consumer_tag and self._queue and not self._queue.is_closed:
                 try:
-                    await task
-                except asyncio.CancelledError:
-                    pass
-
-        # 重置状态
-        self.channel = None
-        self.exchange = None
-        self.queue = None
-        self._consumer_tag = None
-        self._processing_message_ids.clear()
-
-        logger.info("RabbitMQ客户端已关闭")
+                    await self._queue.cancel(self._consumer_tag)
+                    SYLogger.info(f"停止消费成功，consumer_tag: {self._consumer_tag}")
+                except Exception as e:
+                    SYLogger.error(f"停止消费失败: {str(e)}", exc_info=True)
+                finally:
+                    self._consumer_tag = None
 
     async def publish(
         self,
-        message_body: Union[str, Dict[str, Any]],
-        routing_key: Optional[str] = None,
-        content_type: str = "application/json",
+        message_body: Union[str, Dict[str, Any], MQMsgModel],
         headers: Optional[Dict[str, Any]] = None,
-        delivery_mode: DeliveryMode = DeliveryMode.PERSISTENT
+        content_type: str = "application/json",
+        delivery_mode: DeliveryMode = DeliveryMode.PERSISTENT,
+        retry_count: int = 3,
     ) -> None:
-        """发布消息（从池获取通道，自动重试）"""
-        if not self.is_connected:
-            logger.warning("连接已关闭，尝试重连后发布消息")
-            await self.connect(force_reconnect=True)
-
-        if not self.channel or not self.exchange:
-            raise Exception("RabbitMQ连接未初始化")
-
-        # 处理消息体
-        if isinstance(message_body, dict):
-            message_body_str = json.dumps(message_body, ensure_ascii=False)
-            if content_type == "text/plain":
-                content_type = "application/json"
-        else:
-            message_body_str = str(message_body)
-
-        # 创建消息对象
-        message = Message(
-            body=message_body_str.encode(),
-            content_type=content_type,
-            headers=headers or {},
-            delivery_mode=delivery_mode
-        )
-
-        # 发布消息（带重试机制）
-        retry_count = 0
-        max_retries = 2
-        while retry_count < max_retries:
-            try:
-                # 从池获取新通道用于发布（避免长时间占用消费通道）
-                async with self.connection_pool.channel_pool.acquire() as publish_channel:
-                    exchange = await publish_channel.get_exchange(self.exchange_name)
-                    confirmed = await exchange.publish(
-                        message,
-                        routing_key=routing_key or self.routing_key or '#',
-                        mandatory=True,
-                        timeout=5.0
-                    )
-                    if not confirmed:
-                        raise Exception("消息未被服务器确认接收")
-
-                self._update_activity_timestamp()
-                logger.debug(f"消息已发布到交换机 '{self.exchange_name}'")
-                return
-            except (ConnectionClosed, ChannelInvalidStateError):
-                retry_count += 1
-                logger.warning(f"连接已关闭，尝试重连后重新发布 (重试次数: {retry_count})")
-                await self.connect(force_reconnect=True)
-            except Exception as e:
-                retry_count += 1
-                logger.error(f"消息发布失败 (重试次数: {retry_count}): {str(e)}")
-                if retry_count < max_retries:
-                    await asyncio.sleep(1)
-
-        raise Exception(f"消息发布失败，经过{retry_count}次重试仍未成功")
-
-    # 以下方法（消息消费相关）逻辑与原有保持一致，仅适配连接池
-    def set_message_handler(self, handler):
-        self.message_handler = handler
-
-    async def start_consuming(self) -> ConsumerTag:
-        if self._is_consuming:
-            logger.debug("已经在消费中，返回现有consumer_tag")
-            if self._consumer_tag:
-                return self._consumer_tag
-            # 如果_is_consuming为True但无consumer_tag，重置状态并重新启动
-            logger.warning("检测到消费状态异常（无consumer_tag），重置状态后重试")
-            self._is_consuming = False  # 重置状态
-
-        if not self.is_connected:
-            await self.connect()
-
-        if not self.queue:
-            raise Exception("队列未初始化，无法开始消费")
-
-        if not self.message_handler:
-            raise Exception("未设置消息处理函数")
-
-        self._is_consuming = True
-        logger.info(f"开始消费队列: {self.actual_queue_name}")
+        """
+        发布消息（支持自动重试、mandatory路由校验、5秒超时控制）
+        :param message_body: 消息体（字符串/字典/MQMsgModel）
+        :param headers: 消息头（可选）
+        :param content_type: 内容类型（默认application/json）
+        :param delivery_mode: 投递模式（PERSISTENT=持久化，TRANSIENT=非持久化）
+        :param retry_count: 重试次数（默认3次）
+        """
+        if self._closed:
+            raise RuntimeError("客户端已关闭，无法发布消息")
 
+        # 处理消息体序列化
         try:
-            # 调用队列的consume方法，确保获取到consumer_tag
-            self._consumer_tag = await self.queue.consume(
-                self._message_wrapper,
-                no_ack=False  # 手动确认消息
-            )
-
-            # 确保consumer_tag有效
-            if not self._consumer_tag:
-                raise Exception("未能获取到有效的consumer_tag")
-
-            logger.info(
-                f"消费者已启动，队列: {self.actual_queue_name}, tag: {self._consumer_tag}")
-            return self._consumer_tag
+            if isinstance(message_body, MQMsgModel):
+                body = json.dumps(message_body.to_dict(),
+                                  ensure_ascii=False).encode("utf-8")
+            elif isinstance(message_body, dict):
+                body = json.dumps(
+                    message_body, ensure_ascii=False).encode("utf-8")
+            elif isinstance(message_body, str):
+                body = message_body.encode("utf-8")
+            else:
+                raise TypeError(f"不支持的消息体类型: {type(message_body)}")
         except Exception as e:
-            self._is_consuming = False  # 异常时重置状态
-            logger.error(f"启动消费失败: {str(e)}", exc_info=True)
+            SYLogger.error(f"消息体序列化失败: {str(e)}", exc_info=True)
             raise
 
-    async def _safe_cancel_consumer(self) -> bool:
-        if not self._consumer_tag or not self.queue or not self.channel:
-            return True
-
-        try:
-            await asyncio.wait_for(
-                self.queue.cancel(self._consumer_tag),
-                timeout=self.rpc_timeout
-            )
-            logger.info(f"消费者 {self._consumer_tag} 已取消")
-            return True
-        except Exception as e:
-            logger.error(f"取消消费者异常: {str(e)}")
-            return False
-
-    async def stop_consuming(self) -> None:
-        if not self._is_consuming:
-            return
-
-        self._is_consuming = False
-
-        if self._consumer_tag and self.queue:
-            await self._safe_cancel_consumer()
-
-        # 等待所有正在处理的消息完成
-        if self._processing_message_ids:
-            logger.info(
-                f"等待 {len(self._processing_message_ids)} 个正在处理的消息完成...")
-            while self._processing_message_ids and not self._closed:
-                await asyncio.sleep(0.1)
-
-        # 清理状态
-        self._consumer_tag = None
-        self._processing_message_ids.clear()
-
-        logger.info(f"已停止消费队列: {self.actual_queue_name}")
-
-    async def _parse_message(self, message: AbstractIncomingMessage) -> Union[Dict[str, Any], str]:
-        try:
-            body_str = message.body.decode('utf-8')
-            self._update_activity_timestamp()
-
-            if self.auto_parse_json:
-                return json.loads(body_str)
-            return body_str
-        except json.JSONDecodeError:
-            logger.warning(f"消息解析JSON失败，返回原始字符串")
-            return body_str
-        except Exception as e:
-            logger.error(f"消息解析出错: {str(e)}")
-            return message.body.decode('utf-8')
-
-    async def _message_wrapper(self, message: AbstractIncomingMessage) -> None:
-        if not self.message_handler or not self._is_consuming:
-            logger.warning("未设置消息处理器或已停止消费")
-            # await message.ack()
-            return
-
-        message_id = message.message_id or str(id(message))
-        if message_id in self._processing_message_ids:
-            logger.warning(f"检测到重复处理的消息ID: {message_id}，直接确认")
-            await message.ack()
-            return
-
-        self._processing_message_ids.add(message_id)
-
-        try:
-            logger.debug(f"收到队列 {self.actual_queue_name} 的消息: {message_id}")
-
-            parsed_data = await self._parse_message(message)
-            await self.message_handler(MQMsgModel(** parsed_data), message)
-
-            await message.ack()
-            self._update_activity_timestamp()
-            self._update_message_processed_timestamp()
-            logger.debug(f"消息 {message_id} 处理完成并确认")
+        # 构建消息对象
+        message = Message(
+            body=body,
+            headers=headers or {},
+            content_type=content_type,
+            delivery_mode=delivery_mode,
+        )
 
-        except Exception as e:
-            current_headers = message.headers or {}
-            retry_count = current_headers.get('x-retry-count', 0)
-            retry_count += 1
+        # 发布重试逻辑
+        for retry in range(retry_count):
+            try:
+                # 确保连接有效
+                if not await self.is_connected:
+                    SYLogger.warning(f"发布消息前连接失效，触发重连（retry: {retry}）")
+                    await self.connect()
+
+                # 核心：发布消息（添加 mandatory=True 和 timeout=5.0）
+                publish_result = await self._exchange.publish(
+                    message=message,
+                    routing_key=self.routing_key or self.queue_name or "#",
+                    mandatory=True,  # 强制路由到至少一个队列，否则返回None
+                    timeout=5.0      # 5秒超时控制，避免无限阻塞
+                )
 
-            logger.error(
-                f"消息 {message_id} 处理出错（第{retry_count}次重试）: {str(e)}",
-                exc_info=True
-            )
+                # 处理 mandatory=True 结果：未路由到队列返回 None，直接抛出异常
+                if publish_result is None:
+                    raise RuntimeError(
+                        f"消息未找到匹配的队列（routing_key: {self.routing_key}），mandatory=True 触发失败"
+                    )
 
-            if retry_count >= MAX_RETRY_COUNT:
-                logger.error(
-                    f"消息 {message_id} 已达到最大重试次数({MAX_RETRY_COUNT}次)，标记为失败")
-                await message.ack()
-                self._update_activity_timestamp()
+                # 低版本 RobustChannel 异步确认，无需显式等待，仅日志说明
+                SYLogger.info(
+                    f"消息发布成功（retry: {retry}），routing_key: {self.routing_key}，"
+                    f"delivery_mode: {delivery_mode.value}，mandatory: True，timeout: 5.0s"
+                )
                 return
+            except asyncio.TimeoutError:
+                SYLogger.error(
+                    f"消息发布超时（retry: {retry}/{retry_count-1}），超时时间: 5.0s"
+                )
+            except RuntimeError as e:
+                # 捕获 mandatory 未路由等业务异常
+                SYLogger.error(
+                    f"消息发布业务失败（retry: {retry}/{retry_count-1}）: {str(e)}"
+                )
+            except Exception as e:
+                SYLogger.error(
+                    f"消息发布失败（retry: {retry}/{retry_count-1}）: {str(e)}",
+                    exc_info=True
+                )
+                # 清理失效状态，下次重试重连
+                self._exchange = None
+            # 指数退避重试间隔（0.5s, 1s, 2s...）
+            await asyncio.sleep(0.5 * (2 ** retry))
+
+        # 所有重试失败，抛出最终异常
+        raise RuntimeError(
+            f"消息发布失败（已重试{retry_count}次），routing_key: {self.routing_key}，"
+            f"mandatory: True，timeout: 5.0s"
+        )
 
-            new_headers = current_headers.copy()
-            new_headers['x-retry-count'] = retry_count
+    async def close(self) -> None:
+        """关闭客户端（移除回调+释放资源）"""
+        self._closed = True
+        SYLogger.info("开始关闭RabbitMQ客户端...")
 
-            new_message = Message(
-                body=message.body,
-                content_type=message.content_type,
-                headers=new_headers,
-                delivery_mode=message.delivery_mode
-            )
+        # 停止重连任务
+        if self._current_reconnect_task and not self._current_reconnect_task.done():
+            self._current_reconnect_task.cancel()
+            try:
+                await self._current_reconnect_task
+            except asyncio.CancelledError:
+                SYLogger.debug("重连任务已取消")
 
-            await message.reject(requeue=False)
+        # 1. 停止消费
+        await self.stop_consuming()
 
-            if self.exchange:
-                await self.exchange.publish(
-                    new_message,
-                    routing_key=self.routing_key or '#',
-                    mandatory=True,
-                    timeout=5.0
-                )
-                self._update_activity_timestamp()
-                logger.info(f"消息 {message_id} 已重新发布，当前重试次数: {retry_count}")
-        finally:
-            if message_id in self._processing_message_ids:
-                self._processing_message_ids.remove(message_id)
-
-    async def __aenter__(self):
-        await self.connect()
-        return self
-
-    async def __aexit__(self, exc_type, exc, tb):
-        await self.close()
+        # 2. 释放通道到连接池
+        async with self._connect_lock:
+            if self._channel and self._channel_conn:
+                try:
+                    # 移除连接关闭回调
+                    if self._conn_close_callback:
+                        self._channel_conn.close_callbacks.discard(
+                            self._conn_close_callback)
+                    await self.connection_pool.release_channel(self._channel, self._channel_conn)
+                    SYLogger.info("通道释放成功")
+                except Exception as e:
+                    SYLogger.error(f"通道释放失败: {str(e)}", exc_info=True)
+            self._channel = None
+            self._channel_conn = None
+            self._exchange = None
+            self._queue = None
+            self._message_handler = None
+
+        SYLogger.info("RabbitMQ客户端已完全关闭")