diff --git a/.env.example b/.env.example
index 501b10a..cc5a88b 100644
--- a/.env.example
+++ b/.env.example
@@ -4,6 +4,7 @@ PASSWORD=your_idrac_password
 FAN_SPEED_STEPS=50:20,55:25,60:30,65:40
 CONTROL_INTERVAL_SECONDS=120
 ERROR_INTERVAL_SECONDS=120
+IPMI_FAILURE_BACKOFF_SECONDS=300
 IPMI_RETRY_COUNT=5
 IPMI_RETRY_DELAY_SECONDS=20
 IPMI_TIMEOUT_SECONDS=60
diff --git a/CHANGES.md b/CHANGES.md
index dc5556c..e8a40d6 100644
--- a/CHANGES.md
+++ b/CHANGES.md
@@ -12,6 +12,7 @@
 - Docker 运行镜像切换到 Debian slim，只安装 `python3`、`ipmitool` 和时区数据，在降低体积的同时保留更好的IPMI兼容性。
 - 新增 `FAN_SPEED_STEPS` 环境变量，允许用户通过 `.env` 自定义温度阈值和风扇转速档位。
 - 新增轮询间隔、IPMI重试、命令超时和 raw 风扇占空比查询开关，默认减少 iDRAC8 的会话压力。
+- 单轮 IPMI 重试全部失败时改为冷却跳过，并避免重复发送手动模式切换 raw 命令。
 
 ## Previous Improvements
 
diff --git a/README.md b/README.md
index f17d821..372e88b 100644
--- a/README.md
+++ b/README.md
@@ -75,6 +75,7 @@ PASSWORD=your_idrac_password
 FAN_SPEED_STEPS=50:20,55:25,60:30,65:40
 CONTROL_INTERVAL_SECONDS=120
 ERROR_INTERVAL_SECONDS=120
+IPMI_FAILURE_BACKOFF_SECONDS=300
 IPMI_RETRY_COUNT=5
 IPMI_RETRY_DELAY_SECONDS=20
 IPMI_TIMEOUT_SECONDS=60
@@ -120,6 +121,7 @@ python3 start.py
 | `FAN_SPEED_STEPS` | No | Temperature-to-speed rules. Default: `50:20,55:25,60:30,65:40`. / 温度和风扇转速规则，默认值：`50:20,55:25,60:30,65:40`。 |
 | `CONTROL_INTERVAL_SECONDS` | No | Normal control interval. Default: `120`. / 正常控制间隔，默认 `120` 秒。 |
 | `ERROR_INTERVAL_SECONDS` | No | Wait time after a failed control cycle. Default: same as `CONTROL_INTERVAL_SECONDS`. / 控制周期失败后的等待时间，默认等于正常控制间隔。 |
+| `IPMI_FAILURE_BACKOFF_SECONDS` | No | Cooldown after all IPMI retries fail. Default: `300`. / 单轮 IPMI 重试全部失败后的冷却时间，默认 `300` 秒。 |
 | `IPMI_RETRY_COUNT` | No | Retry count for each IPMI command. Default: `5`. / 单条 IPMI 命令重试次数，默认 `5`。 |
 | `IPMI_RETRY_DELAY_SECONDS` | No | Wait time between IPMI retries. Default: `20`. / IPMI 重试间隔，默认 `20` 秒。 |
 | `IPMI_TIMEOUT_SECONDS` | No | Subprocess timeout for each IPMI command. Default: `60`. / 单次 IPMI 命令超时时间，默认 `60` 秒。 |
@@ -182,7 +184,9 @@ ipmitool -H 192.168.1.100 -I lanplus -U root -P your_idrac_password sdr
 
 Common issues / 常见问题：
 
-- `Unable to establish IPMI v2 / RMCP+ session`: iDRAC IPMI service may be busy or unstable. Check network latency, duplicate monitoring scripts, and consider resetting iDRAC with `mc reset cold`.
+- `Unable to establish IPMI v2 / RMCP+ session`: iDRAC IPMI service may be busy or unstable. Occasional retries are expected on some iDRAC8 systems. If all retries fail, the controller skips that cycle and waits for `IPMI_FAILURE_BACKOFF_SECONDS`.
+- `Unable to establish IPMI v2 / RMCP+ session`：iDRAC IPMI 服务可能繁忙或不稳定。部分 iDRAC8 偶发重试是正常现象；如果单轮重试全部失败，控制器会跳过本轮并按 `IPMI_FAILURE_BACKOFF_SECONDS` 冷却等待。
+- Frequent IPMI session failures / 频繁 IPMI 会话失败：check network latency, duplicate monitoring scripts or duplicate containers, and consider resetting iDRAC with `mc reset cold`.
 - Connection failed / 连接失败：确认容器主机能访问 iDRAC 管理 IP。
 - Authentication failed / 认证失败：确认用户名、密码和 IPMI 权限。
 - Permission denied / 权限不足：建议使用专用 iDRAC 用户，并授予 IPMI 控制权限。
diff --git a/controller/client.py b/controller/client.py
index 0808505..f97c1dd 100644
--- a/controller/client.py
+++ b/controller/client.py
@@ -42,6 +42,7 @@ class FanController:
         self.use_raw_fan_duty = use_raw_fan_duty
         self.last_set_speed = None  # 记录最后设置的风扇速度
         self.is_auto_mode = False   # 记录当前是否为自动模式
+        self.is_manual_mode = False  # 记录当前是否已切换到手动模式
 
     # 解析温控规则配置，格式为 "50:20,55:25,60:30,65:40"
     def parse_fan_speed_steps(self, steps: str) -> tuple:
@@ -92,7 +93,12 @@ class FanController:
     # 设置手动风扇速度
     def set_fan_speed(self, speed: int):
         logger.info(f'设置风扇速度: {speed}%')
-        self.ipmi.set_fan_speed(speed)
+        if not self.is_manual_mode:
+            # 首次进入手动风扇控制时才切换模式，避免每次调速都多发一次raw命令
+            self.ipmi.switch_fan_mode(auto=False)
+            self.is_manual_mode = True
+
+        self.ipmi.set_fan_speed(speed, ensure_manual=False)
 
     # 根据最高温度计算目标风扇转速
     def get_required_fan_speed(self, temperature: int) -> int:
@@ -122,6 +128,7 @@ class FanController:
                 logger.info(f'切换风扇为自动模式')
                 self.ipmi.switch_fan_mode(auto=True)
                 self.is_auto_mode = True
+                self.is_manual_mode = False
                 self.last_set_speed = None  # 重置手动设置的速度
             else:
                 logger.info(f'当前已是自动模式，无需操作')
@@ -132,6 +139,7 @@ class FanController:
                 logger.info(f'从自动模式切换到手动模式')
                 self.ipmi.switch_fan_mode(auto=False)
                 self.is_auto_mode = False
+                self.is_manual_mode = True
 
             # 获取当前风扇转速
             current_speed = self.ipmi.get_fan_duty_cycle(sensor_data, use_raw=self.use_raw_fan_duty)
diff --git a/controller/ipmi.py b/controller/ipmi.py
index f227a8e..27fd51d 100644
--- a/controller/ipmi.py
+++ b/controller/ipmi.py
@@ -198,10 +198,11 @@ class IpmiTool:
         return self.run_cmd(cmd=auto_cmd) if auto else self.run_cmd(cmd=manual_cmd)
 
     # 设置手动风扇速度百分比
-    def set_fan_speed(self, speed: int):
+    def set_fan_speed(self, speed: int, ensure_manual: bool = True):
         """
         设置风扇速度
         :param speed:
+        :param ensure_manual:
         :return:
         """
         if speed < 10 or speed > 100:
@@ -209,6 +210,8 @@ class IpmiTool:
                 'speed must be between 10 and 100'
             )
 
-        self.switch_fan_mode(auto=False)
+        if ensure_manual:
+            self.switch_fan_mode(auto=False)
+
         base_cmd = 'raw 0x30 0x30 0x02 0xff'
         return self.run_cmd(cmd=f'{base_cmd} {hex(speed)}')
diff --git a/start.py b/start.py
index 430b671..2e2bc5d 100644
--- a/start.py
+++ b/start.py
@@ -1,6 +1,5 @@
 import os
 import time
-import traceback
 
 from controller.client import FanController
 from controller.logger import logger
@@ -51,6 +50,7 @@ if __name__ == '__main__':
         fan_speed_steps = os.getenv('FAN_SPEED_RULES')
     control_interval = get_int_env('CONTROL_INTERVAL_SECONDS', 120)
     error_interval = get_int_env('ERROR_INTERVAL_SECONDS', control_interval)
+    ipmi_failure_backoff = get_int_env('IPMI_FAILURE_BACKOFF_SECONDS', 300)
     ipmi_retry_count = get_int_env('IPMI_RETRY_COUNT', 5)
     ipmi_retry_delay = get_int_env('IPMI_RETRY_DELAY_SECONDS', 20)
     ipmi_timeout = get_int_env('IPMI_TIMEOUT_SECONDS', 60)
@@ -81,9 +81,13 @@ if __name__ == '__main__':
             # 执行一次温度读取和风扇控制周期
             client.run()
             time.sleep(control_interval)
-        except Exception as err:
-            logger.error(
-                f'运行控制器失败 {err}. {traceback.format_exc()}'
+        except RuntimeError as err:
+            logger.warning(
+                f'本轮IPMI控制失败，跳过本轮并等待 {ipmi_failure_backoff} 秒: {err}'
             )
+            # 连续会话失败时给iDRAC更长恢复窗口，避免马上进入下一轮重试
+            time.sleep(ipmi_failure_backoff)
+        except Exception as err:
+            logger.error(f'运行控制器失败 {err}', exc_info=True)
             # iDRAC会话异常时等待下一轮，避免连续请求压垮IPMI服务
             time.sleep(error_interval)
diff --git a/tests/test_fan_controller.py b/tests/test_fan_controller.py
index 8a3ed59..7eb70c2 100644
--- a/tests/test_fan_controller.py
+++ b/tests/test_fan_controller.py
@@ -1,4 +1,5 @@
 import unittest
+from unittest.mock import Mock
 
 from controller.client import FanController
 
@@ -54,6 +55,16 @@ class FanControllerConfigTest(unittest.TestCase):
 
         self.assertFalse(controller.use_raw_fan_duty)
 
+    def test_set_fan_speed_switches_manual_mode_only_once(self):
+        controller = self.make_controller()
+        controller.ipmi = Mock()
+
+        controller.set_fan_speed(30)
+        controller.set_fan_speed(40)
+
+        controller.ipmi.switch_fan_mode.assert_called_once_with(auto=False)
+        self.assertEqual(controller.ipmi.set_fan_speed.call_count, 2)
+
 
 if __name__ == '__main__':
     unittest.main()