Create monitoring_system.py

src/monitoring_system.py

@@ -0,0 +1,70 @@
+import time
+import psutil
+from datetime import datetime, timedelta
+from typing import Dict, List, Optional
+import logging
+from threading import Thread
+
+class ComprehensiveMonitor:
+    """Monitoring system for performance tracking"""
+    
+    def __init__(self):
+        self.inference_metrics = []
+        self.system_metrics = []
+        self.start_time = datetime.now()
+        self.setup_logging()
+    
+    def setup_logging(self):
+        """Setup logging"""
+        self.logger = logging.getLogger(__name__)
+    
+    def record_inference(self, metrics: Dict):
+        """Record inference metrics"""
+        metrics['timestamp'] = datetime.now()
+        self.inference_metrics.append(metrics)
+        
+        # Keep only last 1000 records
+        if len(self.inference_metrics) > 1000:
+            self.inference_metrics = self.inference_metrics[-1000:]
+    
+    def get_recent_metrics(self, minutes: int = 5) -> List[Dict]:
+        """Get metrics from recent time window"""
+        cutoff = datetime.now() - timedelta(minutes=minutes)
+        return [m for m in self.inference_metrics if m['timestamp'] > cutoff]
+    
+    def get_average_response_time(self) -> float:
+        """Get average response time"""
+        recent_metrics = self.get_recent_metrics(30)  # Last 30 minutes
+        if not recent_metrics:
+            return 0.0
+        
+        successful_metrics = [m for m in recent_metrics if m.get('success', False)]
+        if not successful_metrics:
+            return 0.0
+        
+        return sum(m['processing_time_ms'] for m in successful_metrics) / len(successful_metrics)
+    
+    def get_error_rate(self) -> float:
+        """Get error rate percentage"""
+        recent_metrics = self.get_recent_metrics(30)  # Last 30 minutes
+        if not recent_metrics:
+            return 0.0
+        
+        errors = sum(1 for m in recent_metrics if not m.get('success', True))
+        return (errors / len(recent_metrics)) * 100
+    
+    def get_uptime(self) -> float:
+        """Get system uptime in seconds"""
+        return (datetime.now() - self.start_time).total_seconds()
+    
+    def get_system_health(self) -> Dict:
+        """Get system health status"""
+        return {
+            "cpu_percent": psutil.cpu_percent(),
+            "memory_percent": psutil.virtual_memory().percent,
+            "disk_percent": psutil.disk_usage('/').percent,
+            "uptime_seconds": self.get_uptime(),
+            "total_requests": len(self.inference_metrics),
+            "error_rate": self.get_error_rate(),
+            "avg_response_time": self.get_average_response_time()
+        }