isa-model 0.4.0__py3-none-any.whl → 0.4.3__py3-none-any.whl

isa_model/core/logging/influx_logger.py

@@ -0,0 +1,523 @@
+"""
+InfluxDB-based Inference Logging System for ISA Model
+
+This module provides comprehensive logging for model inference requests,
+optimized for time-series analysis and monitoring.
+
+Data Model:
+- measurement: 'inference_requests' (main table)
+- tags: indexed fields for fast queries (provider, model, service_type, etc.)
+- fields: numerical and text data (tokens, costs, response_time, etc.)
+- timestamp: automatic time-based partitioning
+
+Key Features:
+- Automatic data retention (30 days default)
+- Cost-effective storage with compression
+- Real-time monitoring capabilities
+- Aggregated metrics generation
+"""
+
+import os
+import json
+import hashlib
+import logging
+from datetime import datetime, timezone
+from typing import Dict, Any, Optional, List, Union
+from dataclasses import dataclass, field
+from influxdb_client import InfluxDBClient, Point, WritePrecision
+from influxdb_client.client.write_api import SYNCHRONOUS
+import uuid
+
+logger = logging.getLogger(__name__)
+
+@dataclass
+class InferenceLogEntry:
+    """
+    Structure for inference log data
+    """
+    # Required fields
+    request_id: str
+    service_type: str  # 'text', 'vision', 'audio', 'image', 'embedding'
+    task: str         # 'chat', 'analyze_image', 'generate_speech', etc.
+    provider: str     # 'openai', 'replicate', 'anthropic', etc.
+    model_name: str   # Actual model used
+    status: str       # 'completed', 'failed', 'timeout'
+    
+    # Timing data
+    start_time: datetime
+    end_time: Optional[datetime] = None
+    execution_time_ms: Optional[int] = None
+    queue_time_ms: Optional[int] = None
+    
+    # Token and usage data
+    input_tokens: Optional[int] = None
+    output_tokens: Optional[int] = None
+    total_tokens: Optional[int] = None
+    
+    # Cost data
+    estimated_cost_usd: Optional[float] = None
+    actual_cost_usd: Optional[float] = None
+    cost_breakdown: Optional[Dict[str, Any]] = None
+    
+    # Request/response data (optional, for debugging)
+    input_data_hash: Optional[str] = None
+    input_size_bytes: Optional[int] = None
+    output_size_bytes: Optional[int] = None
+    
+    # Streaming data
+    is_streaming: bool = False
+    stream_start_time: Optional[datetime] = None
+    stream_chunks_count: Optional[int] = None
+    time_to_first_token_ms: Optional[int] = None
+    
+    # Error information
+    error_message: Optional[str] = None
+    error_code: Optional[str] = None
+    
+    # Context and metadata
+    session_id: Optional[str] = None
+    user_id: Optional[str] = None
+    client_ip: Optional[str] = None
+    model_version: Optional[str] = None
+    cache_hit: bool = False
+    
+    # Quality metrics
+    quality_score: Optional[float] = None
+    user_feedback: Optional[int] = None  # 1-5 rating
+    
+    # Additional metadata
+    custom_metadata: Dict[str, Any] = field(default_factory=dict)
+
+class InfluxInferenceLogger:
+    """
+    InfluxDB-based logger for model inference activities
+    
+    Features:
+    - Time-series storage optimized for metrics
+    - Automatic data retention and compression
+    - Real-time query capabilities
+    - Cost tracking and analysis
+    - Performance monitoring
+    """
+    
+    def __init__(self):
+        """Initialize InfluxDB connection"""
+        self.enabled = os.getenv('ENABLE_INFERENCE_LOGGING', 'false').lower() == 'true'
+        
+        # Logging configuration - always set these regardless of enabled status
+        self.retention_days = int(os.getenv('LOG_RETENTION_DAYS', '30'))
+        self.log_detailed_requests = os.getenv('LOG_DETAILED_REQUESTS', 'true').lower() == 'true'
+        self.log_sensitive_data = os.getenv('LOG_SENSITIVE_DATA', 'false').lower() == 'true'
+        
+        if not self.enabled:
+            logger.info("Inference logging disabled via ENABLE_INFERENCE_LOGGING")
+            return
+        
+        # InfluxDB configuration
+        from ..config.config_manager import ConfigManager
+        config_manager = ConfigManager()
+        # Use Consul discovery for InfluxDB URL with fallback
+        self.url = os.getenv('INFLUXDB_URL', config_manager.get_influxdb_url())
+        self.token = os.getenv('INFLUXDB_TOKEN', 'dev-token-isa-model-12345')
+        self.org = os.getenv('INFLUXDB_ORG', 'isa-model')
+        self.bucket = os.getenv('INFLUXDB_BUCKET', 'isa-model-logs')
+        
+        try:
+            # Initialize InfluxDB client
+            self.client = InfluxDBClient(url=self.url, token=self.token, org=self.org)
+            self.write_api = self.client.write_api(write_options=SYNCHRONOUS)
+            self.query_api = self.client.query_api()
+            
+            # Test connection
+            self._test_connection()
+            logger.info(f"InfluxDB inference logger initialized: {self.url}")
+            
+        except Exception as e:
+            logger.error(f"Failed to initialize InfluxDB logger: {e}")
+            self.enabled = False
+    
+    def _test_connection(self):
+        """Test InfluxDB connection"""
+        try:
+            health = self.client.health()
+            if health.status == "pass":
+                logger.debug("InfluxDB connection healthy")
+            else:
+                raise Exception(f"InfluxDB health check failed: {health.message}")
+        except Exception as e:
+            raise Exception(f"InfluxDB connection test failed: {e}")
+    
+    def _create_data_hash(self, data: Any) -> str:
+        """Create SHA-256 hash of input data for deduplication"""
+        try:
+            if isinstance(data, (dict, list)):
+                data_str = json.dumps(data, sort_keys=True)
+            else:
+                data_str = str(data)
+            return hashlib.sha256(data_str.encode()).hexdigest()
+        except Exception:
+            return None
+    
+    def log_inference_start(
+        self,
+        request_id: str,
+        service_type: str,
+        task: str,
+        provider: str,
+        model_name: str,
+        input_data: Any = None,
+        session_id: Optional[str] = None,
+        user_id: Optional[str] = None,
+        client_ip: Optional[str] = None,
+        is_streaming: bool = False,
+        custom_metadata: Optional[Dict[str, Any]] = None
+    ) -> None:
+        """
+        Log the start of an inference request
+        """
+        if not self.enabled:
+            return
+        
+        try:
+            start_time = datetime.now(timezone.utc)
+            
+            # Create data hash for input
+            input_hash = None
+            input_size = None
+            if input_data and self.log_detailed_requests:
+                input_hash = self._create_data_hash(input_data)
+                try:
+                    input_size = len(str(input_data).encode('utf-8'))
+                except:
+                    input_size = None
+            
+            # Create InfluxDB point
+            point = Point("inference_requests") \
+                .tag("service_type", service_type) \
+                .tag("task", task) \
+                .tag("provider", provider) \
+                .tag("model_name", model_name) \
+                .tag("status", "started") \
+                .field("request_id", request_id) \
+                .field("is_streaming", is_streaming) \
+                .time(start_time, WritePrecision.MS)
+            
+            # Add optional tags and fields
+            if session_id:
+                point = point.tag("session_id", session_id)
+            if user_id:
+                point = point.tag("user_id", user_id)
+            if client_ip and not self.log_sensitive_data:
+                # Hash IP for privacy
+                ip_hash = hashlib.md5(client_ip.encode()).hexdigest()[:8]
+                point = point.field("client_ip_hash", ip_hash)
+            if input_hash:
+                point = point.field("input_data_hash", input_hash)
+            if input_size:
+                point = point.field("input_size_bytes", input_size)
+            if custom_metadata:
+                for key, value in custom_metadata.items():
+                    point = point.field(f"meta_{key}", str(value))
+            
+            # Write to InfluxDB
+            self.write_api.write(bucket=self.bucket, org=self.org, record=point)
+            logger.debug(f"Logged inference start: {request_id}")
+            
+        except Exception as e:
+            logger.error(f"Failed to log inference start: {e}")
+    
+    def log_inference_complete(
+        self,
+        request_id: str,
+        status: str = "completed",
+        execution_time_ms: Optional[int] = None,
+        input_tokens: Optional[int] = None,
+        output_tokens: Optional[int] = None,
+        estimated_cost_usd: Optional[float] = None,
+        output_data: Any = None,
+        stream_chunks_count: Optional[int] = None,
+        time_to_first_token_ms: Optional[int] = None,
+        error_message: Optional[str] = None,
+        error_code: Optional[str] = None,
+        cache_hit: bool = False,
+        quality_score: Optional[float] = None,
+        custom_metadata: Optional[Dict[str, Any]] = None
+    ) -> None:
+        """
+        Log the completion of an inference request
+        """
+        if not self.enabled:
+            return
+        
+        try:
+            end_time = datetime.now(timezone.utc)
+            
+            # Calculate output data size
+            output_size = None
+            if output_data and self.log_detailed_requests:
+                try:
+                    output_size = len(str(output_data).encode('utf-8'))
+                except:
+                    output_size = None
+            
+            # Create InfluxDB point
+            point = Point("inference_requests") \
+                .tag("status", status) \
+                .field("request_id", request_id) \
+                .field("cache_hit", cache_hit) \
+                .time(end_time, WritePrecision.MS)
+            
+            # Add timing data
+            if execution_time_ms is not None:
+                point = point.field("execution_time_ms", execution_time_ms)
+            
+            # Add token data
+            if input_tokens is not None:
+                point = point.field("input_tokens", input_tokens)
+            if output_tokens is not None:
+                point = point.field("output_tokens", output_tokens)
+            if input_tokens and output_tokens:
+                point = point.field("total_tokens", input_tokens + output_tokens)
+            
+            # Add cost data
+            if estimated_cost_usd is not None:
+                point = point.field("estimated_cost_usd", float(estimated_cost_usd))
+            
+            # Add output data size
+            if output_size:
+                point = point.field("output_size_bytes", output_size)
+            
+            # Add streaming metrics
+            if stream_chunks_count is not None:
+                point = point.field("stream_chunks_count", stream_chunks_count)
+            if time_to_first_token_ms is not None:
+                point = point.field("time_to_first_token_ms", time_to_first_token_ms)
+            
+            # Add error information
+            if error_message:
+                point = point.field("error_message", error_message[:500])  # Limit length
+            if error_code:
+                point = point.field("error_code", error_code)
+            
+            # Add quality metrics
+            if quality_score is not None:
+                point = point.field("quality_score", float(quality_score))
+            
+            # Add custom metadata
+            if custom_metadata:
+                for key, value in custom_metadata.items():
+                    point = point.field(f"meta_{key}", str(value))
+            
+            # Write to InfluxDB
+            self.write_api.write(bucket=self.bucket, org=self.org, record=point)
+            logger.debug(f"Logged inference completion: {request_id} ({status})")
+            
+        except Exception as e:
+            logger.error(f"Failed to log inference completion: {e}")
+    
+    def log_token_usage(
+        self,
+        request_id: str,
+        provider: str,
+        model_name: str,
+        prompt_tokens: int,
+        completion_tokens: int,
+        prompt_cost_usd: Optional[float] = None,
+        completion_cost_usd: Optional[float] = None
+    ) -> None:
+        """
+        Log detailed token usage data
+        """
+        if not self.enabled:
+            return
+        
+        try:
+            timestamp = datetime.now(timezone.utc)
+            total_tokens = prompt_tokens + completion_tokens
+            total_cost = (prompt_cost_usd or 0) + (completion_cost_usd or 0)
+            
+            point = Point("token_usage") \
+                .tag("provider", provider) \
+                .tag("model_name", model_name) \
+                .field("request_id", request_id) \
+                .field("prompt_tokens", prompt_tokens) \
+                .field("completion_tokens", completion_tokens) \
+                .field("total_tokens", total_tokens) \
+                .time(timestamp, WritePrecision.MS)
+            
+            if prompt_cost_usd is not None:
+                point = point.field("prompt_cost_usd", float(prompt_cost_usd))
+            if completion_cost_usd is not None:
+                point = point.field("completion_cost_usd", float(completion_cost_usd))
+            if total_cost > 0:
+                point = point.field("total_cost_usd", float(total_cost))
+                point = point.field("cost_per_token_usd", float(total_cost / total_tokens))
+            
+            self.write_api.write(bucket=self.bucket, org=self.org, record=point)
+            logger.debug(f"Logged token usage: {request_id}")
+            
+        except Exception as e:
+            logger.error(f"Failed to log token usage: {e}")
+    
+    def log_error(
+        self,
+        request_id: str,
+        error_type: str,
+        error_message: str,
+        error_code: Optional[str] = None,
+        provider: Optional[str] = None,
+        model_name: Optional[str] = None,
+        retry_count: int = 0
+    ) -> None:
+        """
+        Log error events
+        """
+        if not self.enabled:
+            return
+        
+        try:
+            timestamp = datetime.now(timezone.utc)
+            
+            point = Point("inference_errors") \
+                .tag("error_type", error_type) \
+                .field("request_id", request_id) \
+                .field("error_message", error_message[:500]) \
+                .field("retry_count", retry_count) \
+                .time(timestamp, WritePrecision.MS)
+            
+            if error_code:
+                point = point.field("error_code", error_code)
+            if provider:
+                point = point.tag("provider", provider)
+            if model_name:
+                point = point.tag("model_name", model_name)
+            
+            self.write_api.write(bucket=self.bucket, org=self.org, record=point)
+            logger.debug(f"Logged error: {request_id} - {error_type}")
+            
+        except Exception as e:
+            logger.error(f"Failed to log error: {e}")
+    
+    def get_recent_requests(
+        self,
+        limit: int = 100,
+        hours: int = 24,
+        service_type: Optional[str] = None,
+        provider: Optional[str] = None,
+        status: Optional[str] = None
+    ) -> List[Dict[str, Any]]:
+        """
+        Query recent inference requests
+        """
+        if not self.enabled:
+            return []
+        
+        try:
+            # Build query with simpler filtering
+            filters = []
+            if service_type:
+                filters.append(f'r.service_type == "{service_type}"')
+            if provider:
+                filters.append(f'r.provider == "{provider}"')
+            if status:
+                filters.append(f'r.status == "{status}"')
+            
+            # Build filter clause
+            if filters:
+                filter_clause = " and " + " and ".join(filters)
+            else:
+                filter_clause = ""
+            
+            query = f'''
+            from(bucket: "{self.bucket}")
+                |> range(start: -{hours}h)
+                |> filter(fn: (r) => r._measurement == "inference_requests"{filter_clause})
+                |> filter(fn: (r) => r._field == "request_id")
+                |> sort(columns: ["_time"], desc: true)
+                |> limit(n: {limit})
+            '''
+            
+            result = self.query_api.query(org=self.org, query=query)
+            
+            # Process results - get unique request IDs first
+            request_ids = []
+            for table in result:
+                for record in table.records:
+                    request_id = record.get_value()
+                    if request_id not in [r.get('request_id') for r in request_ids]:
+                        request_ids.append({
+                            'request_id': request_id,
+                            'time': record.get_time(),
+                            'service_type': record.values.get('service_type'),
+                            'provider': record.values.get('provider'),
+                            'model_name': record.values.get('model_name'),
+                            'status': record.values.get('status'),
+                            'task': record.values.get('task')
+                        })
+            
+            return request_ids
+            
+        except Exception as e:
+            logger.error(f"Failed to query recent requests: {e}")
+            return []
+    
+    def get_usage_statistics(
+        self,
+        hours: int = 24,
+        group_by: str = "provider"  # "provider", "model_name", "service_type"
+    ) -> Dict[str, Any]:
+        """
+        Get usage statistics and metrics
+        """
+        if not self.enabled:
+            return {}
+        
+        try:
+            # Simplified query to count unique request IDs by group
+            query = f'''
+            from(bucket: "{self.bucket}")
+                |> range(start: -{hours}h)
+                |> filter(fn: (r) => r._measurement == "inference_requests")
+                |> filter(fn: (r) => r._field == "request_id")
+                |> group(columns: ["{group_by}"])
+                |> count()
+                |> yield(name: "request_counts")
+            '''
+            
+            result = self.query_api.query(org=self.org, query=query)
+            
+            # Process results into statistics
+            stats = {}
+            for table in result:
+                for record in table.records:
+                    key = record.values.get(group_by, 'unknown')
+                    stats[key] = {
+                        'total_requests': record.get_value() or 0,
+                        'group_by': group_by,
+                        'time_range_hours': hours
+                    }
+            
+            return stats
+            
+        except Exception as e:
+            logger.error(f"Failed to get usage statistics: {e}")
+            return {}
+    
+    def close(self):
+        """Close InfluxDB connection"""
+        if self.enabled and hasattr(self, 'client'):
+            self.client.close()
+
+# Global logger instance
+_inference_logger: Optional[InfluxInferenceLogger] = None
+
+def get_inference_logger() -> InfluxInferenceLogger:
+    """Get or create global inference logger instance"""
+    global _inference_logger
+    if _inference_logger is None:
+        _inference_logger = InfluxInferenceLogger()
+    return _inference_logger
+
+def generate_request_id() -> str:
+    """Generate unique request ID"""
+    return f"req_{uuid.uuid4().hex[:12]}"

isa_model/core/logging/loki_logger.py

@@ -0,0 +1,160 @@
+"""
+Centralized Logging Configuration with Loki Integration for ISA Model
+
+This module provides centralized application logging with Loki support,
+complementing the existing InfluxDB inference logging system.
+
+Architecture:
+- Loki: General application logs (INFO, WARNING, ERROR, DEBUG)
+- InfluxDB: Inference metrics and performance data (tokens, costs, timing)
+
+Usage:
+    from isa_model.core.logging import app_logger, api_logger
+
+    app_logger.info("Service starting...")
+    api_logger.error(f"Request failed: {error}", exc_info=True)
+"""
+
+import logging
+import sys
+import os
+from typing import Optional
+
+
+def setup_logger(
+    name: str,
+    level: Optional[str] = None,
+    format_str: Optional[str] = None
+) -> logging.Logger:
+    """
+    Setup logger with centralized Loki integration
+
+    Args:
+        name: Logger name (e.g., "ISAModel.API")
+        level: Log level (DEBUG, INFO, WARNING, ERROR, CRITICAL)
+        format_str: Log format string (optional)
+
+    Returns:
+        Configured logger instance
+
+    Example:
+        >>> from isa_model.core.logging import setup_logger
+        >>> my_logger = setup_logger("ISAModel.MyModule")
+        >>> my_logger.info("Processing started")
+    """
+    logger = logging.getLogger(name)
+
+    # Avoid duplicate handlers
+    if logger.handlers:
+        return logger
+
+    # Get configuration from environment
+    log_level_env = os.getenv("LOG_LEVEL", "INFO").upper()
+    log_format_env = os.getenv(
+        "LOG_FORMAT",
+        "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
+    )
+    from ..config.config_manager import ConfigManager
+    config_manager = ConfigManager()
+    # Use Consul discovery for Loki URL with fallback
+    loki_url = os.getenv("LOKI_URL", config_manager.get_loki_url())
+    loki_enabled = os.getenv("LOKI_ENABLED", "true").lower() == "true"
+
+    # Set log level
+    final_level = (level or log_level_env).upper()
+    logger.setLevel(getattr(logging, final_level, logging.INFO))
+
+    # Disable propagation to prevent duplicate logs
+    logger.propagate = False
+
+    # Log format
+    formatter = logging.Formatter(format_str or log_format_env)
+
+    # 1. Console Handler (for local development and debugging)
+    console_handler = logging.StreamHandler(sys.stdout)
+    console_handler.setFormatter(formatter)
+    logger.addHandler(console_handler)
+
+    # 2. Loki Handler (for centralized logging)
+    if loki_enabled:
+        try:
+            from logging_loki import LokiHandler
+
+            # Extract service name and logger component
+            # e.g., "ISAModel.API" -> service="isa_model", logger="API"
+            service_name = "isa_model"
+            logger_component = name.replace("ISAModel.", "").replace("ISAModel", "main")
+
+            # Labels for Loki (used for filtering and searching)
+            # Use service_name to match other services (mcp, agent, etc.)
+            loki_labels = {
+                "service_name": "model",  # Use "model" to match service naming convention
+                "logger": logger_component,
+                "environment": os.getenv("ENVIRONMENT", "development"),
+                "job": "isa_model_service"
+            }
+
+            # Create Loki handler
+            loki_handler = LokiHandler(
+                url=f"{loki_url}/loki/api/v1/push",
+                tags=loki_labels,
+                version="1",
+            )
+
+            # Only send INFO and above to Loki (reduce network traffic)
+            loki_handler.setLevel(logging.INFO)
+
+            logger.addHandler(loki_handler)
+
+        except ImportError:
+            # Silently fall back to console-only logging during initialization
+            pass
+        except Exception as e:
+            # Loki unavailable - silently fall back to console
+            pass
+
+    return logger
+
+
+# Create application loggers
+# Main application logger
+app_logger = setup_logger("ISAModel")
+
+# API/Server logger
+api_logger = setup_logger("ISAModel.API")
+
+# Client logger
+client_logger = setup_logger("ISAModel.Client")
+
+# Inference logger (application-level, not metrics)
+inference_logger = setup_logger("ISAModel.Inference")
+
+# Training logger
+training_logger = setup_logger("ISAModel.Training")
+
+# Evaluation logger
+eval_logger = setup_logger("ISAModel.Evaluation")
+
+# Database logger
+db_logger = setup_logger("ISAModel.Database")
+
+# Deployment logger
+deployment_logger = setup_logger("ISAModel.Deployment")
+
+# Model manager logger
+model_logger = setup_logger("ISAModel.Models")
+
+
+# Export all loggers
+__all__ = [
+    'setup_logger',
+    'app_logger',
+    'api_logger',
+    'client_logger',
+    'inference_logger',
+    'training_logger',
+    'eval_logger',
+    'db_logger',
+    'deployment_logger',
+    'model_logger',
+]