[Mellanox] Adjust log level to avoid too many thermal logs (sonic-net#4631)

* Trigger thermal action log only if thermal condition changes
* test file existence before read file content
* fix error for set psu fan speed
* Remove logs because it print too frequently

Author: Junchao-Mellanox

platform/mellanox/mlnx-platform-api/sonic_platform/fan.py

@@ -241,6 +241,8 @@ def set_speed(self, speed):
         status = True
 
         if self.is_psu_fan:
+            if not self.get_presence():
+                return False
             from .thermal import logger
             try:
                 with open(self.psu_i2c_bus_path, 'r') as f:

platform/mellanox/mlnx-platform-api/sonic_platform/psu.py

@@ -112,6 +112,8 @@ def _read_generic_file(self, filename, len):
         """
         result = 0
         try:
+            if not os.path.exists(filename):
+                return result
             with open(filename, 'r') as fileobj:
                 result = int(fileobj.read().strip())
         except Exception as e:

platform/mellanox/mlnx-platform-api/sonic_platform/thermal.py

@@ -495,12 +495,15 @@ def set_thermal_algorithm_status(cls, status, force=True):
         We usually disable the algorithm when we want to set a fix speed. E.g, when 
         a fan unit is removed from system, we will set fan speed to 100% and disable 
         the algorithm to avoid it adjust the speed.
+
+        Returns:
+            True if thermal algorithm status changed.
         """
         if not cls.thermal_profile:
             raise Exception("Fail to get thermal profile for this switch")
 
         if not force and cls.thermal_algorithm_status == status:
-            return
+            return False
 
         cls.thermal_algorithm_status = status
         content = "enabled" if status else "disabled"
@@ -521,6 +524,7 @@ def set_thermal_algorithm_status(cls, status, force=True):
                 for index in range(count):
                     cls._write_generic_file(join(THERMAL_ZONE_GEARBOX_PATH.format(start + index), THERMAL_ZONE_MODE), content)
                     cls._write_generic_file(join(THERMAL_ZONE_GEARBOX_PATH.format(start + index), THERMAL_ZONE_POLICY), policy)
+        return True
 
     @classmethod
     def check_thermal_zone_temperature(cls):

platform/mellanox/mlnx-platform-api/sonic_platform/thermal_actions.py

@@ -66,9 +66,6 @@ def set_psu_fan_speed(cls, thermal_info_dict, speed):
                 for psu_fan in psu.get_all_fans():
                     psu_fan.set_speed(speed)
 
-            logger.log_info('Updated PSU FAN speed to {}%'.format(speed))
-
-
 
 @thermal_json_object('fan.all.check_and_set_speed')
 class CheckAndSetAllFanSpeedAction(SetAllFanSpeedAction):
@@ -131,14 +128,17 @@ def execute(self, thermal_info_dict):
         from .thermal import Thermal
         from .thermal_conditions import UpdateCoolingLevelToMinCondition
         from .fan import Fan
-        Thermal.set_thermal_algorithm_status(self.status, False)
-        if self.status:
-            # Check thermal zone temperature, if all thermal zone temperature
-            # back to normal, set it to minimum allowed speed to
-            # save power
-            UpdateCoolingLevelToMinAction.update_cooling_level_to_minimum(thermal_info_dict)
+        status_changed = Thermal.set_thermal_algorithm_status(self.status, False)
 
-        logger.log_info('Changed thermal algorithm status to {}'.format(self.status))
+        # Only update cooling level if thermal algorithm status changed
+        if status_changed:
+            if self.status:
+                # Check thermal zone temperature, if all thermal zone temperature
+                # back to normal, set it to minimum allowed speed to
+                # save power
+                UpdateCoolingLevelToMinAction.update_cooling_level_to_minimum(thermal_info_dict)
+
+            logger.log_info('Changed thermal algorithm status to {}'.format(self.status))
 
 
 class ChangeMinCoolingLevelAction(ThermalPolicyActionBase):
@@ -174,8 +174,6 @@ def execute(self, thermal_info_dict):
             Fan.set_cooling_level(Fan.min_cooling_level, current_cooling_level)
             UpdateCoolingLevelToMinAction.update_cooling_level_to_minimum(thermal_info_dict)
 
-        logger.log_info('Changed minimum cooling level to {}'.format(Fan.min_cooling_level))
-
 
 class UpdatePsuFanSpeedAction(ThermalPolicyActionBase):
     def execute(self, thermal_info_dict):