src/monitoring/interaction_monitor.py

import logging
from typing import Dict, Any
from prometheus_client import Counter, Histogram, start_http_server
import time
from datetime import datetime

class InteractionMonitor:
    """Monitor and track user interactions with the platform"""
    
    def __init__(self, enable_metrics_server: bool = True, metrics_port: int = 8000):
        """Initialize the monitoring system
        
        Args:
            enable_metrics_server: Whether to start the Prometheus metrics server
            metrics_port: Port for the metrics server
        """
        self.logger = logging.getLogger(__name__)
        
        # Initialize Prometheus metrics
        self.interaction_counter = Counter(
            'mental_wellness_interactions_total',
            'Total number of interactions',
            ['agent_type', 'interaction_type']
        )
        
        self.response_time = Histogram(
            'mental_wellness_response_time_seconds',
            'Time taken to generate response',
            ['agent_type']
        )
        
        self.error_counter = Counter(
            'mental_wellness_errors_total',
            'Total number of errors',
            ['agent_type', 'error_type']
        )
        
        # Start metrics server if enabled
        if enable_metrics_server:
            try:
                start_http_server(metrics_port)
                self.logger.info(f"Metrics server started on port {metrics_port}")
            except Exception as e:
                self.logger.error(f"Failed to start metrics server: {str(e)}")
    
    def track_interaction(self, data: Dict[str, Any]):
        """Track a user interaction
        
        Args:
            data: Dictionary containing interaction data
                - agent_type: Type of agent (e.g., therapeutic, crisis)
                - interaction_type: Type of interaction (e.g., chat, voice)
                - response_time: Time taken to generate response
                - error: Optional error information
        """
        try:
            # Log the interaction
            self.logger.info(
                f"Interaction tracked - Agent: {data.get('agent_type')}, "
                f"Type: {data.get('interaction_type')}"
            )
            
            # Update Prometheus metrics
            self.interaction_counter.labels(
                agent_type=data.get('agent_type', 'unknown'),
                interaction_type=data.get('interaction_type', 'unknown')
            ).inc()
            
            # Track response time if available
            if 'response_time' in data:
                self.response_time.labels(
                    agent_type=data.get('agent_type', 'unknown')
                ).observe(data['response_time'])
            
            # Track errors if any
            if 'error' in data:
                self.error_counter.labels(
                    agent_type=data.get('agent_type', 'unknown'),
                    error_type=type(data['error']).__name__
                ).inc()
                
        except Exception as e:
            self.logger.error(f"Error tracking interaction: {str(e)}")
    
    def get_metrics(self) -> Dict[str, Any]:
        """Get current metrics
        
        Returns:
            Dictionary containing current metrics
        """
        return {
            "total_interactions": self.interaction_counter._value.sum(),
            "total_errors": self.error_counter._value.sum(),
            "average_response_time": self.response_time._sum.sum() / max(self.response_time._count.sum(), 1)
        }