From d1d41106bd3970835b6b90aa6ae56c2caa410892 Mon Sep 17 00:00:00 2001
From: Jared Vititoe <jjvititoe1@gmail.com>
Date: Thu, 5 Dec 2024 21:17:09 -0500
Subject: [PATCH] Changed detection for smart

---
 hwmonDaemon.py | 84 +++++++++++++++++---------------------------------
 1 file changed, 29 insertions(+), 55 deletions(-)

diff --git a/hwmonDaemon.py b/hwmonDaemon.py
index 4d53a97..23a4318 100644
--- a/hwmonDaemon.py
+++ b/hwmonDaemon.py
@@ -248,7 +248,7 @@ class SystemHealthMonitor:
             'temp': None,
             'attributes': {}
         }
-        
+    
         try:
             result = subprocess.run(
                 ['smartctl', '-A', '-H', device],
@@ -256,74 +256,48 @@ class SystemHealthMonitor:
                 stderr=subprocess.PIPE,
                 text=True
             )
-            
+        
             output = result.stdout
-            
+        
             # Check overall SMART status first
-            if 'FAILED' in output and not 'PASSED' in output:
+            if 'FAILED' in output and 'PASSED' not in output:
                 smart_health['status'] = 'UNHEALTHY'
                 smart_health['issues'].append("SMART overall health check failed")
-                
-            # Define critical attributes and their thresholds
-            critical_attributes = {
-                'Reallocated_Sector_Ct': {'threshold': 0, 'critical': True},
-                'Current_Pending_Sector': {'threshold': 0, 'critical': True},
-                'Offline_Uncorrectable': {'threshold': 0, 'critical': True},
-                'Reported_Uncorrect': {'threshold': 0, 'critical': True},
-                'Command_Timeout': {'threshold': 5, 'critical': False},
-                'Temperature_Celsius': {'threshold': 65, 'critical': False},
-                'Wear_Leveling_Count': {'threshold': 10, 'critical': True},
-                'Media_Wearout_Indicator': {'threshold': 20, 'critical': True}
-            }
             
+            # Parse SMART attributes
             for line in output.split('\n'):
-                # Skip header lines
-                if 'ATTRIBUTE_NAME' in line or '===' in line:
+                if 'ATTRIBUTE_NAME' in line:
                     continue
-                    
-                for attr_name, limits in critical_attributes.items():
-                    if attr_name in line:
+                
+                # Check for current failures only
+                if 'WHEN_FAILED' in line and 'In_the_past' not in line and '-' not in line:
+                    smart_health['status'] = 'UNHEALTHY'
+                    smart_health['issues'].append(f"Current failure detected: {line}")
+                
+                # Monitor critical attributes
+                for attr in ['Reallocated_Sector_Ct', 'Current_Pending_Sector', 
+                            'Offline_Uncorrectable', 'Reported_Uncorrect']:
+                    if attr in line:
                         parts = line.split()
                         if len(parts) >= 10:
-                            value = int(parts[9])  # Raw value
-                            normalized = int(parts[3])  # Normalized value
+                            raw_value = int(parts[9])
+                            if raw_value > 0:
+                                smart_health['status'] = 'UNHEALTHY'
+                                smart_health['issues'].append(f"{attr} has value {raw_value}")
                             
-                            smart_health['attributes'][attr_name] = {
-                                'raw': value,
-                                'normalized': normalized
-                            }
-                            
-                            # Check thresholds
-                            if attr_name == 'Temperature_Celsius':
-                                smart_health['temp'] = value
-                                if value > limits['threshold']:
-                                    smart_health['issues'].append(
-                                        f"Drive temperature critical: {value}°C"
-                                    )
-                            elif value > limits['threshold']:
-                                if limits['critical']:
-                                    smart_health['status'] = 'UNHEALTHY'
-                                smart_health['issues'].append(
-                                    f"{attr_name} above threshold: {value}"
-                                )
-                                
-                            # Check for very low normalized values
-                            if normalized <= 10 and attr_name != 'Temperature_Celsius':
-                                smart_health['issues'].append(
-                                    f"{attr_name} normalized value critical: {normalized}"
-                                )
-                                
-            # Check if WHEN_FAILED is present and not in the past
-            if 'WHEN_FAILED' in output:
-                for line in output.split('\n'):
-                    if 'WHEN_FAILED' in line and 'In_the_past' not in line and '-' not in line:
-                        smart_health['status'] = 'UNHEALTHY'
-                        smart_health['issues'].append(f"Current failure detected: {line}")
+                # Check temperature
+                if 'Temperature_Celsius' in line or 'Airflow_Temperature_Cel' in line:
+                    parts = line.split()
+                    if len(parts) >= 10:
+                        temp = int(parts[9])
+                        smart_health['temp'] = temp
+                        if temp > 65:
+                            smart_health['issues'].append(f"High drive temperature: {temp}°C")
                         
         except Exception as e:
             smart_health['status'] = 'ERROR'
             smart_health['issues'].append(f"Error checking SMART: {str(e)}")
-        
+    
         return smart_health
     def _check_drives_health(self) -> Dict[str, Any]:
         """