aibilly_backend_code/src/infrastructure/observability/metrics.py

from prometheus_client import Counter, Histogram, Gauge, generate_latest, REGISTRY
from prometheus_client import CollectorRegistry, multiprocess, start_http_server
from src.infrastructure.observability.logger import logger
import os

"""
Prometheus Metrics Service
Production-ready metrics collection with Prometheus format
"""
class MetricsService:
    def __init__(self):
        self.registry = REGISTRY
        
        # Default labels for all metrics
        self.default_labels = {
            'service': os.getenv('APP_NAME', 'test_project'),
            'environment': os.getenv('ENVIRONMENT', 'development')
        }
        
        # Custom metrics
        self._initialize_custom_metrics()
        
        # Start metrics server if enabled
        if os.getenv('METRICS_ENABLED', 'true').lower() == 'true':
            self._start_metrics_server()
    
    def _initialize_custom_metrics(self):
        """Initialize custom application metrics"""
        # HTTP request metrics
        self.http_request_duration = Histogram(
            'http_request_duration_seconds',
            'Duration of HTTP requests in seconds',
            ['method', 'route', 'status_code'],
            buckets=[0.1, 0.5, 1, 2, 5, 10, 30]
        )
        
        self.http_request_total = Counter(
            'http_requests_total',
            'Total number of HTTP requests',
            ['method', 'route', 'status_code']
        )
        
        # Business metrics
        self.business_events_total = Counter(
            'business_events_total',
            'Total number of business events',
            ['event_type', 'status']
        )
        
        # Database metrics
        self.database_query_duration = Histogram(
            'database_query_duration_seconds',
            'Duration of database queries in seconds',
            ['operation', 'table'],
            buckets=[0.01, 0.05, 0.1, 0.5, 1, 2, 5]
        )
        
        # Cache metrics
        self.cache_hits = Counter(
            'cache_hits_total',
            'Total number of cache hits',
            ['cache_type']
        )
        
        self.cache_misses = Counter(
            'cache_misses_total',
            'Total number of cache misses',
            ['cache_type']
        )
        
        # Kafka metrics
        self.kafka_messages_published = Counter(
            'kafka_messages_published_total',
            'Total number of Kafka messages published',
            ['topic', 'status']
        )
        
        self.kafka_messages_consumed = Counter(
            'kafka_messages_consumed_total',
            'Total number of Kafka messages consumed',
            ['topic', 'status']
        )
    
    def _start_metrics_server(self):
        """Start Prometheus metrics HTTP server"""
        try:
            port = int(os.getenv('PROMETHEUS_PORT', '9090'))
            start_http_server(port)
            logger.info(f'Metrics: Prometheus metrics server started on port {port}')
        except Exception as e:
            logger.error(f'Metrics: Failed to start metrics server: {e}')
    
    def record_http_request(self, method: str, route: str, status_code: int, duration: float):
        """
        Record HTTP request duration
        
        Args:
            method: HTTP method
            route: Route path
            status_code: HTTP status code
            duration: Duration in seconds
        """
        self.http_request_duration.labels(method=method, route=route, status_code=status_code).observe(duration)
        self.http_request_total.labels(method=method, route=route, status_code=status_code).inc()
    
    def record_business_event(self, event_type: str, status: str = 'success'):
        """
        Record business event
        
        Args:
            event_type: Event type
            status: Event status ('success' or 'error')
        """
        self.business_events_total.labels(event_type=event_type, status=status).inc()
    
    def record_database_query(self, operation: str, table: str, duration: float):
        """
        Record database query duration
        
        Args:
            operation: Operation type (select, insert, update, delete)
            table: Table name
            duration: Duration in seconds
        """
        self.database_query_duration.labels(operation=operation, table=table).observe(duration)
    
    def record_cache_hit(self, cache_type: str = 'redis'):
        """
        Record cache hit
        
        Args:
            cache_type: Cache type (e.g., 'redis', 'memory')
        """
        self.cache_hits.labels(cache_type=cache_type).inc()
    
    def record_cache_miss(self, cache_type: str = 'redis'):
        """
        Record cache miss
        
        Args:
            cache_type: Cache type
        """
        self.cache_misses.labels(cache_type=cache_type).inc()
    
    def record_kafka_published(self, topic: str, status: str = 'success'):
        """
        Record Kafka message published
        
        Args:
            topic: Topic name
            status: Status ('success' or 'error')
        """
        self.kafka_messages_published.labels(topic=topic, status=status).inc()
    
    def record_kafka_consumed(self, topic: str, status: str = 'success'):
        """
        Record Kafka message consumed
        
        Args:
            topic: Topic name
            status: Status ('success' or 'error')
        """
        self.kafka_messages_consumed.labels(topic=topic, status=status).inc()
    
    def get_metrics(self):
        """Get metrics in Prometheus format"""
        return generate_latest(self.registry)
    
    def get_registry(self):
        """Get metrics registry"""
        return self.registry

# Singleton instance
metrics_service = MetricsService()