Add dynamic sensor logic for fixed and psu presence/state checking in thermalctld

sonic-thermalctld/scripts/thermalctld

@@ -520,10 +520,10 @@ class TemperatureUpdater(logger.Logger):
         self.table = swsscommon.Table(state_db, TemperatureUpdater.TEMPER_INFO_TABLE_NAME)
         self.phy_entity_table = swsscommon.Table(state_db, PHYSICAL_ENTITY_INFO_TABLE)
         self.chassis_table = None
+        self.all_thermals = set()
 
         self.is_chassis_system = chassis.is_modular_chassis()
         if self.is_chassis_system:
-            self.module_thermals = set()
             my_slot = try_get(chassis.get_my_slot, INVALID_SLOT)
             if my_slot != INVALID_SLOT:
                 try:
@@ -566,30 +566,34 @@ class TemperatureUpdater(logger.Logger):
         :return:
         """
         self.log_debug("Start temperature updating")
+        available_thermals = set()
         for index, thermal in enumerate(self.chassis.get_all_thermals()):
             if self.task_stopping_event.is_set():
                 return
 
+            available_thermals.add((thermal, CHASSIS_INFO_KEY, index))
             self._refresh_temperature_status(CHASSIS_INFO_KEY, thermal, index)
 
         for psu_index, psu in enumerate(self.chassis.get_all_psus()):
             parent_name = 'PSU {}'.format(psu_index + 1)
-            for thermal_index, thermal in enumerate(psu.get_all_thermals()):
-                if self.task_stopping_event.is_set():
-                    return
+            if psu.get_presence() and psu.get_powergood_status():
+                for thermal_index, thermal in enumerate(psu.get_all_thermals()):
+                    if self.task_stopping_event.is_set():
+                        return
 
-                self._refresh_temperature_status(parent_name, thermal, thermal_index)
+                    available_thermals.add((thermal, parent_name, thermal_index))
+                    self._refresh_temperature_status(parent_name, thermal, thermal_index)
 
         for sfp_index, sfp in enumerate(self.chassis.get_all_sfps()):
             parent_name = 'SFP {}'.format(sfp_index + 1)
             for thermal_index, thermal in enumerate(sfp.get_all_thermals()):
                 if self.task_stopping_event.is_set():
                     return
 
+                available_thermals.add((thermal, parent_name, thermal_index))
                 self._refresh_temperature_status(parent_name, thermal, thermal_index)
 
         if self.is_chassis_system:
-            available_thermals = set()
             for module_index, module in enumerate(self.chassis.get_all_modules()):
                 module_name = try_get(module.get_name, 'Module {}'.format(module_index + 1))
 
@@ -610,19 +614,18 @@ class TemperatureUpdater(logger.Logger):
                         self._refresh_temperature_status(sfp_name, thermal, thermal_index)
 
                 for psu_index, psu in enumerate(module.get_all_psus()):
-                    psu_name = '{} PSU {}'.format(module_name, psu_index + 1)
-                    for thermal_index, thermal in enumerate(psu.get_all_thermals()):
-                        if self.task_stopping_event.is_set():
-                            return
-
-                        available_thermals.add((thermal, psu_name, thermal_index))
-                        self._refresh_temperature_status(psu_name, thermal, thermal_index)
-
-
-            thermals_to_remove = self.module_thermals - available_thermals
-            self.module_thermals = available_thermals
-            for thermal, parent_name, thermal_index in thermals_to_remove:
-                self._remove_thermal_from_db(thermal, parent_name, thermal_index)
+                    if psu.get_presence() and psu.get_powergood_status():
+                        psu_name = '{} PSU {}'.format(module_name, psu_index + 1)
+                        for thermal_index, thermal in enumerate(psu.get_all_thermals()):
+                            if self.task_stopping_event.is_set():
+                                return
+                            available_thermals.add((thermal, psu_name, thermal_index))
+                            self._refresh_temperature_status(psu_name, thermal, thermal_index)
+
+        thermals_to_remove = self.all_thermals - available_thermals
+        self.all_thermals = available_thermals
+        for thermal, parent_name, thermal_index in thermals_to_remove:
+            self._remove_thermal_from_db(thermal, parent_name, thermal_index)
 
         self.log_debug("End temperature updating")
 

sonic-thermalctld/tests/mock_platform.py

@@ -175,6 +175,9 @@ def get_serial(self):
 
     def get_status(self):
         return self._status
+
+    def get_powergood_status(self):
+        return self._status
 
     def set_status(self, status):
         self._status = status

sonic-thermalctld/tests/test_thermalctld.py

@@ -501,11 +501,15 @@ def test_update_module_thermals(self):
         chassis.set_modular_chassis(True)
         temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
         temperature_updater.update()
-        assert len(temperature_updater.module_thermals) == 3
+        assert len(temperature_updater.all_thermals) == 3
+
+        chassis._module_list[0]._psu_list[0].set_status(False)
+        temperature_updater.update()
+        assert len(temperature_updater.all_thermals) == 2
 
         chassis._module_list = []
         temperature_updater.update()
-        assert len(temperature_updater.module_thermals) == 0
+        assert len(temperature_updater.all_thermals) == 0
 
 
 # Modular chassis-related tests