[thermal control] Backport changes from master to 201911 branch (#90)

Author: Junchao-Mellanox

.gitignore

@@ -1,3 +1,5 @@
-build/
+*.pyc
 */__pycache__/
+build/
 sonic_platform_common.egg-info/
+.cache

setup.py

@@ -17,6 +17,7 @@
         'sonic_platform_base.sonic_eeprom',
         'sonic_platform_base.sonic_sfp',
         'sonic_platform_base.sonic_ssd',
+        'sonic_platform_base.sonic_thermal_control',
         'sonic_psu',
         'sonic_sfp',
     ],
@@ -33,5 +34,5 @@
         'Programming Language :: Python :: 3.6',
         'Topic :: Utilities',
     ],
-    keywords='sonic SONiC platform hardware interface api API',
+    keywords='sonic SONiC platform hardware interface api API'
 )

sonic_platform_base/chassis_base.py

@@ -333,6 +333,14 @@ def get_thermal(self, index):
 
         return thermal
 
+    def get_thermal_manager(self):
+        """
+        Retrieves thermal manager class on this chassis
+        :return: A class derived from ThermalManagerBase representing the
+        specified thermal manager. ThermalManagerBase is returned as default
+        """
+        raise NotImplementedError
+
     ##############################################
     # SFP methods
     ##############################################

sonic_platform_base/sonic_thermal_control/__init__.py

@@ -0,0 +1,17 @@
+from .thermal_json_object import ThermalJsonObject
+
+
+class ThermalPolicyActionBase(ThermalJsonObject):
+    """
+    Base class for thermal action. Once all thermal conditions in a thermal policy are matched,
+    all predefined thermal action will be executed.
+    """
+    def execute(self, thermal_info_dict):
+        """
+        Take action when thermal condition matches. For example, adjust speed of fan or shut
+        down the switch.
+        :param thermal_info_dict: A dictionary stores all thermal information.
+        :return:
+        """
+        raise NotImplementedError
+

sonic_platform_base/sonic_thermal_control/thermal_action_base.py

@@ -0,0 +1,14 @@
+from .thermal_json_object import ThermalJsonObject
+
+
+class ThermalPolicyConditionBase(ThermalJsonObject):
+    """
+    Base class for thermal condition
+    """
+    def is_match(self, thermal_info_dict):
+        """
+        Indicate if this condition is matched.
+        :param thermal_info_dict: A dictionary stores all thermal information.
+        :return: True if condition matched else False.
+        """
+        raise NotImplementedError

sonic_platform_base/sonic_thermal_control/thermal_condition_base.py

@@ -0,0 +1,15 @@
+from .thermal_json_object import ThermalJsonObject
+
+
+class ThermalPolicyInfoBase(object):
+    """
+    Base class for thermal information
+    """
+    def collect(self, chassis):
+        """
+        Collect thermal information for thermal policy.
+        :param chassis: The chassis object.
+        :return:
+        """
+        raise NotImplementedError
+

sonic_platform_base/sonic_thermal_control/thermal_info_base.py

@@ -0,0 +1,72 @@
+class ThermalJsonObject(object):
+    """
+    Base class for thermal json object. 
+    """
+    # JSON field definition
+    JSON_FIELD_TYPE = 'type'
+
+    # Dictionary of ThermalJsonObject-derived class representing all thermal json types
+    _object_type_dict = {}
+
+    def load_from_json(self, json_obj):
+        """
+        Initialize this object by a json object. The json object is read from policy json file.
+        Derived class can define any field in policy json file and interpret them in this function.
+        :param json_obj: A json object representing an object.
+        :return:
+        """
+        pass
+
+    def __eq__(self, other):
+        """
+        Compare input object with this object, return True if equal. Subclass should override this
+        if necessary.
+        :param other: Object to compare with.
+        :return: True if equal else False
+        """
+        return self.__class__ == other.__class__
+
+    @classmethod
+    def register_concrete_type(cls, type_name, object_type):
+        """
+        Register a concrete class by type name. The concrete class must derive from
+        ThermalJsonObject.
+        :param type_name: Name of the class.
+        :param object_type: A concrete class.
+        :return:
+        """
+        if type_name not in cls._object_type_dict:
+            cls._object_type_dict[type_name] = object_type
+        else:
+            raise Exception('ThermalJsonObject type {} already exists'.format(type_name))
+
+    @classmethod
+    def get_type(cls, json_obj):
+        """
+        Get a concrete class by json object. The json object represents an object and must
+        have a 'type' field. This function returns a pre-registered concrete class if the specific
+        'type' is found.
+        :param json_obj: A json object representing an action.
+        :return: A concrete class if requested type exists; Otherwise None.
+        """
+        if ThermalJsonObject.JSON_FIELD_TYPE in json_obj:
+            type_str = json_obj[ThermalJsonObject.JSON_FIELD_TYPE]
+            if type_str in cls._object_type_dict:
+                return cls._object_type_dict[type_str]
+            else:
+                raise Exception('ThermalJsonObject type {} not found'.format(type_str) )
+
+        raise Exception('Invalid policy file, {} field must be presented'.format(ThermalJsonObject.JSON_FIELD_TYPE))
+
+
+def thermal_json_object(type_name):
+    """
+    Decorator to auto register a ThermalJsonObject-derived class
+    :param type_name: Type name of the concrete class which corresponding to the 'type' field of
+    a condition, action or info.
+    :return: Wrapper function
+    """
+    def wrapper(object_type):
+        ThermalJsonObject.register_concrete_type(type_name, object_type)
+        return object_type
+    return wrapper

sonic_platform_base/sonic_thermal_control/thermal_json_object.py

@@ -0,0 +1,203 @@
+import json
+from .thermal_policy import ThermalPolicy
+from .thermal_json_object import ThermalJsonObject
+
+
+class ThermalManagerBase(object):
+    """
+    Base class of ThermalManager representing a manager to control all thermal policies.
+    """
+    # JSON field definition.
+    JSON_FIELD_POLICIES = 'policies'
+    JSON_FIELD_INFO_TYPES = 'info_types'
+    JSON_FIELD_POLICY_NAME = 'name'
+    JSON_FIELD_THERMAL_ALGORITHM = "thermal_control_algorithm"
+    JSON_FIELD_FAN_SPEED_WHEN_SUSPEND = "fan_speed_when_suspend"
+    JSON_FIELD_RUN_AT_BOOT_UP = "run_at_boot_up"
+
+    # Dictionary of ThermalPolicy objects.
+    _policy_dict = {}
+
+    # Dictionary of thermal information objects. A thermal information object is used by Thermal Policy
+    _thermal_info_dict = {}
+
+    _fan_speed_when_suspend = None
+
+    _run_thermal_algorithm_at_boot_up = None
+
+    @classmethod
+    def initialize(cls):
+        """
+        Initialize thermal manager, including register thermal condition types and thermal action types
+        and any other vendor specific initialization.
+        :return:
+        """
+        pass
+
+    @classmethod
+    def deinitialize(cls):
+        """
+        Destroy thermal manager, including any vendor specific cleanup. The default behavior of this function
+        is a no-op.
+        :return:
+        """
+        pass
+
+    @classmethod
+    def start_thermal_control_algorithm(cls):
+        """
+        Start vendor specific thermal control algorithm. The default behavior of this function is a no-op.
+        :return:
+        """
+        pass
+
+    @classmethod
+    def stop_thermal_control_algorithm(cls):
+        """
+        Stop vendor specific thermal control algorithm. The default behavior of this function is a no-op.
+        :return:
+        """
+        pass
+
+    @classmethod
+    def load(cls, policy_file_name):
+        """
+        Load all thermal policies from JSON policy file. An example looks like:
+        {
+          "thermal_control_algorithm": {
+            "run_at_boot_up": "false",
+            "fan_speed_when_suspend": "60"
+          },
+          "info_types": [
+            {
+              "type": "fan_info" # collect fan information for each iteration
+            },
+            {
+              "type": "psu_info" # collect psu information for each iteration
+            }
+          ],
+          "policies": [
+            {
+              "name": "any fan absence", # if any fan absence, set all fan speed to 100% and disable thermal control algorithm
+              "conditions": [
+                {
+                  "type": "fan.any.absence" # see sonic-platform-daemons.sonic-thermalctld.thermal_policy.thermal_conditions
+                }
+              ],
+              "actions": [
+                {
+                  "type": "fan.all.set_speed", # see sonic-platform-daemons.sonic-thermalctld.thermal_policy.thermal_actions
+                  "speed": "100"
+                },
+                {
+                  "type": "thermal_control.control",
+                  "status": "false"
+                }
+              ]
+            },
+            {
+              "name": "all fan absence", # if all fan absence, shutdown the switch
+              "conditions": [
+                {
+                  "type": "fan.all.absence"
+                }
+              ],
+              "actions": [
+                {
+                  "type": "switch.shutdown"
+                }
+              ]
+            }
+          ]
+        }
+        :param policy_file_name: Path of JSON policy file.
+        :return:
+        """
+        with open(policy_file_name, 'r') as policy_file:
+            json_obj = json.load(policy_file)
+            if cls.JSON_FIELD_POLICIES in json_obj:
+                json_policies = json_obj[cls.JSON_FIELD_POLICIES]
+                for json_policy in json_policies:
+                    cls._load_policy(json_policy)
+
+            if cls.JSON_FIELD_INFO_TYPES in json_obj:
+                for json_info in json_obj[cls.JSON_FIELD_INFO_TYPES]:
+                    info_type = ThermalJsonObject.get_type(json_info)
+                    info_obj = info_type()
+                    cls._thermal_info_dict[json_info[ThermalJsonObject.JSON_FIELD_TYPE]] = info_obj
+
+            if cls.JSON_FIELD_THERMAL_ALGORITHM in json_obj:
+                json_thermal_algorithm_config = json_obj[cls.JSON_FIELD_THERMAL_ALGORITHM]
+                if cls.JSON_FIELD_RUN_AT_BOOT_UP in json_thermal_algorithm_config:
+                    cls._run_thermal_algorithm_at_boot_up = \
+                        True if json_thermal_algorithm_config[cls.JSON_FIELD_RUN_AT_BOOT_UP].lower() == 'true' else False
+
+                if cls.JSON_FIELD_FAN_SPEED_WHEN_SUSPEND in json_thermal_algorithm_config:
+                    # if the string is not a valid int, let it raise
+                    cls._fan_speed_when_suspend = \
+                        int(json_thermal_algorithm_config[cls.JSON_FIELD_FAN_SPEED_WHEN_SUSPEND])
+
+    @classmethod
+    def _load_policy(cls, json_policy):
+        """
+        Load a policy object from a JSON object.
+        :param json_policy: A JSON object representing a thermal policy.
+        :return:
+        """
+        if cls.JSON_FIELD_POLICY_NAME in json_policy:
+            name = json_policy[cls.JSON_FIELD_POLICY_NAME]
+            if name in cls._policy_dict:
+                raise Exception('Policy {} already exists'.format(name))
+
+            policy = ThermalPolicy()
+            policy.load_from_json(json_policy)
+            policy.validate_duplicate_policy(cls._policy_dict.values())
+            cls._policy_dict[name] = policy
+        else:
+            raise Exception('{} not found in policy'.format(cls.JSON_FIELD_POLICY_NAME))
+
+    @classmethod
+    def run_policy(cls, chassis):
+        """
+        Collect thermal information, run each policy, if one policy matches, execute the policy's action.
+        :param chassis: The chassis object.
+        :return:
+        """
+        if not cls._policy_dict:
+            return
+
+        cls._collect_thermal_information(chassis)
+
+        for policy in cls._policy_dict.values():
+            if policy.is_match(cls._thermal_info_dict):
+                policy.do_action(cls._thermal_info_dict)
+
+    @classmethod
+    def _collect_thermal_information(cls, chassis):
+        """
+        Collect thermal information. This function will be called before run_policy.
+        :param chassis: The chassis object.
+        :return:
+        """
+        for thermal_info in cls._thermal_info_dict.values():
+            thermal_info.collect(chassis)
+
+    @classmethod
+    def init_thermal_algorithm(cls, chassis):
+        """
+        Initialize thermal algorithm according to policy file.
+        :param chassis: The chassis object.
+        :return:
+        """
+        if cls._run_thermal_algorithm_at_boot_up is not None:
+            if cls._run_thermal_algorithm_at_boot_up:
+                cls.start_thermal_control_algorithm()
+            else:
+                cls.stop_thermal_control_algorithm()
+                if cls._fan_speed_when_suspend is not None:
+                    for fan in chassis.get_all_fans():
+                        fan.set_speed(cls._fan_speed_when_suspend)
+
+                    for psu in chassis.get_all_psus():
+                        for fan in psu.get_all_fans():
+                            fan.set_speed(cls._fan_speed_when_suspend)