[system health daemon] Support PSU power threshold checking (#11864)

sonic-net · Nov 21, 2022 · 7b4032e · 7b4032e
1 parent a618728
commit 7b4032e
Show file tree

Hide file tree

Showing 2 changed files with 44 additions and 0 deletions.
diff --git a/src/system-health/health_checker/hardware_checker.py b/src/system-health/health_checker/hardware_checker.py
@@ -239,6 +239,18 @@ def _check_psu_status(self, config):
                                                                                                                voltage_min_th,
                                                                                                                voltage_max_th))
                         continue
+
+            if not self._ignore_check(config.ignore_devices, 'psu', name, 'power_threshold'):
+                power_overload = data_dict.get('power_overload', None)
+                if power_overload == 'True':
+                    try:
+                        power = data_dict['power']
+                        power_critical_threshold = data_dict['power_critical_threshold']
+                        self.set_object_not_ok('PSU', name, 'power of {} ({}w) exceeds threshold ({}w)'.format(name, power, power_critical_threshold))
+                    except KeyError:
+                        self.set_object_not_ok('PSU', name, 'power of {} exceeds threshold but power or power_critical_threshold is invalid'.format(name))
+                    continue
+
             self.set_object_ok('PSU', name)
 
     def reset(self):

diff --git a/src/system-health/tests/test_system_health.py b/src/system-health/tests/test_system_health.py
@@ -362,6 +362,30 @@ def test_hardware_checker():
             'voltage': '10',
             'voltage_min_threshold': '12',
             'voltage_max_threshold': '15',
+        },
+        'PSU_INFO|PSU 6': {
+            'presence': 'True',
+            'status': 'True',
+            'temp': '55',
+            'temp_threshold': '100',
+            'voltage': '12',
+            'voltage_min_threshold': '12',
+            'voltage_max_threshold': '15',
+            'power_overload': 'True',
+            'power': '101.0',
+            'power_critical_threshold': '100.0',
+            'power_warning_suppress_threshold': '90.0'
+        },
+        'PSU_INFO|PSU 7': {
+            'presence': 'True',
+            'status': 'True',
+            'temp': '55',
+            'temp_threshold': '100',
+            'voltage': '12',
+            'voltage_min_threshold': '12',
+            'voltage_max_threshold': '15',
+            'power_overload': 'True',
+            'power': '101.0'
         }
     })
 
@@ -400,6 +424,14 @@ def test_hardware_checker():
     assert 'PSU 5' in checker._info
     assert checker._info['PSU 5'][HealthChecker.INFO_FIELD_OBJECT_STATUS] == HealthChecker.STATUS_NOT_OK
 
+    assert 'PSU 6' in checker._info
+    assert checker._info['PSU 6'][HealthChecker.INFO_FIELD_OBJECT_MSG] == 'power of PSU 6 (101.0w) exceeds threshold (100.0w)'
+    assert checker._info['PSU 6'][HealthChecker.INFO_FIELD_OBJECT_STATUS] == HealthChecker.STATUS_NOT_OK
+
+    assert 'PSU 7' in checker._info
+    assert checker._info['PSU 7'][HealthChecker.INFO_FIELD_OBJECT_STATUS] == HealthChecker.STATUS_NOT_OK
+    assert checker._info['PSU 7'][HealthChecker.INFO_FIELD_OBJECT_MSG] == 'power of PSU 7 exceeds threshold but power or power_critical_threshold is invalid'
+
 
 def test_config():
     config = Config()