Addition of DPU Chassis for thermalctld (#564)

sonic-thermalctld/scripts/thermalctld

@@ -26,7 +26,7 @@ SYSLOG_IDENTIFIER = 'thermalctld'
 NOT_AVAILABLE = 'N/A'
 CHASSIS_INFO_KEY = 'chassis 1'
 PHYSICAL_ENTITY_INFO_TABLE = 'PHYSICAL_ENTITY_INFO'
-INVALID_SLOT = -1
+INVALID_SLOT_OR_DPU = -1
 
 ERR_UNKNOWN = 1
 
@@ -523,9 +523,11 @@ class TemperatureUpdater(logger.Logger):
         self.all_thermals = set()
 
         self.is_chassis_system = chassis.is_modular_chassis()
-        if self.is_chassis_system:
-            my_slot = try_get(chassis.get_my_slot, INVALID_SLOT)
-            if my_slot != INVALID_SLOT:
+        self.is_smartswitch_dpu = chassis.is_smartswitch() and chassis.is_dpu()
+        self.is_chassis_upd_required = self.is_chassis_system or self.is_smartswitch_dpu
+        if self.is_chassis_upd_required:
+            my_slot = try_get(chassis.get_my_slot if self.is_chassis_system else chassis.get_dpu_id, INVALID_SLOT_OR_DPU)
+            if my_slot != INVALID_SLOT_OR_DPU:
                 try:
                     # Modular chassis does not have to have table CHASSIS_STATE_DB.
                     # So catch the exception here and ignore it.
@@ -540,7 +542,7 @@ class TemperatureUpdater(logger.Logger):
             table_keys = self.table.getKeys()
             for tk in table_keys:
                 self.table._del(tk)
-                if self.is_chassis_system and self.chassis_table is not None:
+                if self.is_chassis_upd_required and self.chassis_table is not None:
                     self.chassis_table._del(tk)
         if self.phy_entity_table:
             phy_entity_keys = self.phy_entity_table.getKeys()
@@ -593,6 +595,7 @@ class TemperatureUpdater(logger.Logger):
                 available_thermals.add((thermal, parent_name, thermal_index))
                 self._refresh_temperature_status(parent_name, thermal, thermal_index)
 
+        # As there are no modules present in DPU, this IF condition is not updated to consider DPU chassis
         if self.is_chassis_system:
             for module_index, module in enumerate(self.chassis.get_all_modules()):
                 module_name = try_get(module.get_name, 'Module {}'.format(module_index + 1))
@@ -702,7 +705,7 @@ class TemperatureUpdater(logger.Logger):
                 ])
 
             self.table.set(name, fvs)
-            if self.is_chassis_system and self.chassis_table is not None:
+            if self.is_chassis_upd_required and self.chassis_table is not None:
                 self.chassis_table.set(name, fvs)
         except Exception as e:
             self.log_warning('Failed to update thermal status for {} - {}'.format(name, repr(e)))

sonic-thermalctld/tests/mock_platform.py

@@ -335,7 +335,10 @@ def __init__(self):
         self._replaceable = False
 
         self._is_chassis_system = False
+        self._is_dpu = False
+        self._is_smartswitch = False
         self._my_slot = module_base.ModuleBase.MODULE_INVALID_SLOT
+        self._dpu_id = None
         self._thermal_manager = MockThermalManager()
 
     def make_absent_fan(self):
@@ -445,6 +448,26 @@ def get_position_in_parent(self):
     def is_replaceable(self):
         return self._replaceable
 
+    def is_dpu(self):
+        return self._is_dpu
+
+    def is_smartswitch(self):
+        return self._is_smartswitch
+
+    def set_smartswitch(self, is_true):
+        self._is_smartswitch = is_true
+
+    def set_dpu(self, is_true):
+        self._is_dpu = is_true
+
+    def set_dpu_id(self, dpu_id):
+        self._dpu_id = dpu_id
+
+    def get_dpu_id(self):
+        # The default behaviour is Not implemented Error
+        if not self._dpu_id:
+            raise NotImplementedError
+        return self._dpu_id
 
 class MockModule(module_base.ModuleBase):
     def __init__(self):

sonic-thermalctld/tests/test_thermalctld.py

@@ -508,6 +508,74 @@ def test_update_module_thermals(self):
         assert len(temperature_updater.all_thermals) == 0
 
 
+# DPU chassis-related tests
+def test_dpu_chassis_thermals():
+    chassis = MockChassis()
+    # Modular chassis (Not a dpu chassis) No Change in TemperatureUpdater Behaviour
+    chassis.set_modular_chassis(True)
+    chassis.set_my_slot(1)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    assert temperature_updater.chassis_table
+    # DPU chassis TemperatureUpdater without is_smartswitch False return - No update to CHASSIS_STATE_DB
+    chassis.set_modular_chassis(False)
+    chassis.set_dpu(True)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    assert not temperature_updater.chassis_table
+    # DPU chassis TemperatureUpdater without get_dpu_id implmenetation- No update to CHASSIS_STATE_DB
+    chassis.set_smartswitch(True)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    assert not temperature_updater.chassis_table
+    # DPU chassis TemperatureUpdater with get_dpu_id implemented - Update data to CHASSIS_STATE_DB
+    dpu_id = 1
+    chassis.set_dpu_id(dpu_id)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    assert temperature_updater.chassis_table
+    # Table name in chassis state db = TEMPERATURE_INFO_0 for dpu_id 0
+    assert temperature_updater.chassis_table.table_name == f"{TEMPER_INFO_TABLE_NAME}_{dpu_id}"
+    temperature_updater.table = Table("STATE_DB", "xtable")
+    temperature_updater.table._del = mock.MagicMock()
+
+
+def test_dpu_chassis_state_deinit():
+    # Confirm that the chassis_table entries for DPU Chassis are removed on deletion
+    chassis = MockChassis()
+    chassis.set_smartswitch(True)
+    chassis.set_modular_chassis(False)
+    chassis.set_dpu(True)
+    chassis.set_dpu_id(1)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    assert temperature_updater.chassis_table
+    temperature_updater.table = Table("STATE_DB", "xtable")
+    temperature_updater.phy_entity_table = None
+    temperature_updater.table.getKeys = mock.MagicMock(return_value=['key1', 'key2'])
+    temperature_updater.table._del = mock.MagicMock()
+    temperature_updater.chassis_table = Table("CHASSIS_STATE_DB", "ctable")
+    temperature_updater.chassis_table._del = mock.MagicMock()
+    temperature_updater.__del__()
+    assert temperature_updater.chassis_table._del.call_count == 2
+    expected_calls = [mock.call('key1'), mock.call('key2')]
+    temperature_updater.chassis_table._del.assert_has_calls(expected_calls, any_order=True)
+
+
+def test_updater_dpu_thermal_check_chassis_table():
+    chassis = MockChassis()
+
+    thermal1 = MockThermal()
+    chassis.get_all_thermals().append(thermal1)
+
+    chassis.set_dpu(True)
+    chassis.set_smartswitch(True)
+    chassis.set_dpu_id(1)
+    temperature_updater = thermalctld.TemperatureUpdater(chassis, multiprocessing.Event())
+    temperature_updater.update()
+    assert temperature_updater.chassis_table.get_size() == chassis.get_num_thermals()
+
+    thermal2 = MockThermal()
+    chassis.get_all_thermals().append(thermal2)
+    temperature_updater.update()
+    assert temperature_updater.chassis_table.get_size() == chassis.get_num_thermals()
+
+
 # Modular chassis-related tests