[SmartSwitch] Extend implementation of the DPU chassis daemon. (sonic-net#563)

Author: oleksandrivantsiv

sonic-chassisd/scripts/chassisd

@@ -791,21 +791,74 @@ class ChassisdDaemon(daemon_base.DaemonBase):
         self.log_info("Shutting down...")
 
 
+class DpuStateManagerTask(ProcessTaskBase):
+
+    def __init__(self, log_identifier, dpu_state_updater):
+        super(DpuStateManagerTask, self).__init__()
+
+        self.logger = logger.Logger(log_identifier)
+        self.dpu_state_updater = dpu_state_updater
+        self.state_db = daemon_base.db_connect('STATE_DB')
+        self.app_db = daemon_base.db_connect('APPL_DB')
+
+    def task_worker(self):
+        sel = swsscommon.Select()
+        selectable = [
+            swsscommon.SubscriberStateTable(self.app_db, 'PORT_TABLE'),
+            swsscommon.SubscriberStateTable(self.state_db, 'SYSTEM_READY')
+        ]
+
+        for s in selectable:
+            sel.addSelectable(s)
+
+        try:
+            while True:
+                (state, c) = sel.select(SELECT_TIMEOUT)
+
+                if state == swsscommon.Select.TIMEOUT:
+                    continue
+
+                if state != swsscommon.Select.OBJECT:
+                    continue
+
+                for s in selectable:
+                    s.pops()
+
+                self.dpu_state_updater.update_state()
+
+        except KeyboardInterrupt:
+            pass
+
+
 class DpuChassisdDaemon(ChassisdDaemon):
 
     def run(self):
         self.log_info("Starting up...")
 
+        poll_dpu_state = True
+        if not try_get(self.platform_chassis.get_dataplane_state, default=None) and not \
+                try_get(self.platform_chassis.get_controlplane_state, default=None):
+            poll_dpu_state = False
+
         dpu_updater = DpuStateUpdater(SYSLOG_IDENTIFIER, self.platform_chassis)
+        dpu_state_mng = None
+
+        if not poll_dpu_state:
+            dpu_state_mng = DpuStateManagerTask(SYSLOG_IDENTIFIER, dpu_updater)
+            dpu_state_mng.task_run()
 
         # Start main loop
         self.log_info("Start daemon main loop")
 
         while not self.stop.wait(CHASSIS_INFO_UPDATE_PERIOD_SECS):
-            dpu_updater.update_state()
+            if poll_dpu_state:
+                dpu_updater.update_state()
 
         self.log_info("Stop daemon main loop")
 
+        if dpu_state_mng:
+            dpu_state_mng.task_stop()
+
         dpu_updater.deinit()
 
         self.log_info("Shutting down...")

sonic-chassisd/tests/mock_swsscommon.py

@@ -54,6 +54,7 @@ def __init__(self, fvs):
         pass
 
 class Select:
+    OBJECT = 0
     TIMEOUT = 1
 
     def addSelectable(self, selectable):
@@ -66,7 +67,12 @@ def select(self, timeout=-1, interrupt_on_signal=False):
         return self.TIMEOUT, None
 
 class SubscriberStateTable(Table):
-    pass
+
+    def pop(self):
+        return None
+
+    def pops(self):
+        return None
 
 class RedisPipeline:
     def __init__(self, db):

sonic-chassisd/tests/test_dpu_chassisd.py

@@ -116,6 +116,52 @@ def hset(key, field, value):
                  'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
 
 
+@pytest.mark.parametrize('dpu_id, dp_state, cp_state, expected_state', [
+    (0, False, False, {'DPU0':
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
+         'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+    (0, False, True, {'DPU0':
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+    (0, True, True, {'DPU0':
+        {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': '2000-01-01 00:00:00',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+])
+def test_dpu_state_manager(dpu_id, dp_state, cp_state, expected_state):
+    chassis = MockDpuChassis()
+
+    chassis.get_dpu_id = MagicMock(return_value=dpu_id)
+    chassis.get_dataplane_state = MagicMock(return_value=dp_state)
+    chassis.get_controlplane_state = MagicMock(return_value=cp_state)
+
+    chassis_state_db = {}
+
+    def hset(key, field, value):
+        print(key, field, value)
+        if key not in chassis_state_db:
+            chassis_state_db[key] = {}
+
+        chassis_state_db[key][field] = value
+
+    with mock.patch.object(swsscommon.Table, 'hset', side_effect=hset):
+        with mock.patch.object(swsscommon.Select, 'select', side_effect=((swsscommon.Select.OBJECT, None), (swsscommon.Select.OBJECT, None), KeyboardInterrupt)):
+            dpu_updater = DpuStateUpdater(SYSLOG_IDENTIFIER, chassis)
+            dpu_updater._time_now = MagicMock(return_value='2000-01-01 00:00:00')
+
+            dpu_state_mng = DpuStateManagerTask(SYSLOG_IDENTIFIER, dpu_updater)
+
+            dpu_state_mng.task_worker()
+
+            assert chassis_state_db == expected_state
+
+            dpu_updater.deinit()
+
+            # After the deinit we assume that the DPU state is down.
+            assert chassis_state_db == {'DPU0':
+                {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
+                 'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
+
+
 def test_dpu_chassis_daemon():
     # Test the chassisd run
     chassis = MockDpuChassis()