main.py

#!/usr/bin/env python3
"""
FastAPI application for risk_manager (LangGraph)
Following FastAPI best practices for production deployment
"""

import os
from typing import Dict, Any, List
from contextlib import asynccontextmanager
import aiohttp

from fastapi import FastAPI, HTTPException, Request, status
from fastapi.middleware.cors import CORSMiddleware
from fastapi.middleware.trustedhost import TrustedHostMiddleware
from fastapi.responses import JSONResponse
from pydantic import BaseModel, Field
import uvicorn
from dotenv import load_dotenv


# Load environment variables
load_dotenv()

# Configure Handit

from src.agent import LangGraphAgent

# Global agent instance
agent = None

async def fetch_metrics(session_id: str) -> Dict[str, Any]:
    """
    Fetch metrics for a session from the self-improving engine
    
    Args:
        session_id: Session ID to fetch metrics for
        
    Returns:
        Dictionary containing aggregated metrics scores
    """
    if not session_id:
        return {"score": 0.0}
    
    try:
        # Get the base URL from environment or use default
        trace_url = os.getenv("TRACE_API_URL", "https://self-improving-engine-api-299768392189.us-central1.run.app/api/v1/trace")
        # Extract base URL by removing /trace or /context
        base_url = trace_url.replace("/trace", "").replace("/context", "")
        metrics_url = f"{base_url}/metrics/{session_id}"
        
        async with aiohttp.ClientSession() as session:
            async with session.get(metrics_url, timeout=aiohttp.ClientTimeout(total=300)) as response:
                if response.status == 200:
                    result = await response.json()
                    
                    # Extract and aggregate metrics
                    metrics = result.get("metrics", {})
                    total_accuracy = 0.0
                    count = 0
                    
                    # Aggregate accuracies across all runs
                    for run_id, run_metrics in metrics.items():
                        if isinstance(run_metrics, dict):
                            for metric_type, metric_data in run_metrics.items():
                                if isinstance(metric_data, dict):
                                    for subtype, subtype_data in metric_data.items():
                                        if isinstance(subtype_data, dict) and "accuracy" in subtype_data:
                                            total_accuracy += subtype_data["accuracy"]
                                            count += 1
                    
                    avg_score = total_accuracy / count if count > 0 else 0.0
                    
                    return {
                        "score": round(avg_score, 4),
                        "details": result
                    }
                else:
                    print(f"⚠️ Metrics API call failed with status {response.status}")
                    return {"score": 0.0}
    except Exception as e:
        print(f"⚠️ Failed to fetch metrics: {e}")
        return {"score": 0.0}

@asynccontextmanager
async def lifespan(app: FastAPI):
    """Application lifespan manager"""
    global agent
    
    # Startup
    print(f"🚀 Starting risk_manager (LangGraph + FastAPI)")
    
    # Initialize agent
    agent = LangGraphAgent()
    
    # Print graph information
    graph_info = agent.get_graph_info()
    print(f"✅ risk_manager initialized successfully")
    
    yield
    
    # Shutdown
    print(f"🔄 Shutting down risk_manager")

# Create FastAPI app with lifespan
app = FastAPI(
    title="risk_manager",
    description="LangGraph-powered AI agent API",
    version="1.0.0",
    lifespan=lifespan,
    docs_url="/docs",
    redoc_url="/redoc"
)

# Add security middleware - Commented out to allow all connections
# Cloud Run handles security at the infrastructure level
# app.add_middleware(
#     TrustedHostMiddleware,
#     allowed_hosts=["localhost", "127.0.0.1", "0.0.0.0"]
# )

# Add CORS middleware
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Allow all origins
    allow_credentials=True,
    allow_methods=["GET", "POST", "PUT", "DELETE", "OPTIONS"],
    allow_headers=["*"],
)

# Request/Response Models
class ProcessRequest(BaseModel):
    # Accept either a string or a dictionary for flexible input
    input_data: Any = Field(None, description="Input data to process (optional for compatibility)")
    
    # Accept array of transactions for batch processing
    transactions: List[Dict[str, Any]] = Field(None, description="Array of transactions to process")
    
    # Model types to run (vanilla, full, online)
    model_types: List[str] = Field(default=["vanilla", "full", "online"], description="Model types to execute")
    
    # Session ID for tracking multiple runs
    session_id: str = Field(None, description="Session ID to track multiple runs of the same transaction")
    
    # Run ID within session
    run_id: str = Field(None, description="Run ID within session (e.g., 'run-1', 'run-2', 'run-3')")

    # Allow direct transaction fields (fraud detection)
    transaction: Dict[str, Any] = Field(None, description="Transaction details")
    financial: Dict[str, Any] = Field(None, description="Financial information")
    card: Dict[str, Any] = Field(None, description="Card details")
    merchant: Dict[str, Any] = Field(None, description="Merchant information")
    customer: Dict[str, Any] = Field(None, description="Customer details")
    device: Dict[str, Any] = Field(None, description="Device information")
    location: Dict[str, Any] = Field(None, description="Location data")
    behavioral_profile: Dict[str, Any] = Field(None, description="Behavioral profile")
    velocity_counters: Dict[str, Any] = Field(None, description="Velocity metrics")
    risk_signals: Dict[str, Any] = Field(None, description="Risk signals")
    session: Dict[str, Any] = Field(None, description="Session data")
    channel: Dict[str, Any] = Field(None, description="Channel information")
    authentication: Dict[str, Any] = Field(None, description="Authentication data")

    # Simple transaction fields for basic testing
    user_id: str = Field(None, description="User ID")
    user_age_days: int = Field(None, description="Account age in days")
    total_transactions: int = Field(None, description="Total transaction count")
    amount: float = Field(None, description="Transaction amount")
    time: str = Field(None, description="Transaction time (HH:MM format)")
    merchant_name: str = Field(None, description="Merchant name")
    merchant_category: str = Field(None, description="Merchant category")
    currency: str = Field(None, description="Transaction currency code")

    metadata: Dict[str, Any] = Field(default_factory=dict, description="Optional metadata")
    
    class Config:
        schema_extra = {
            "example": {
                "transaction": {
                    "transaction_id": "TXN-001",
                    "transaction_type": "PURCHASE"
                },
                "financial": {
                    "amount": 150.00,
                    "currency": "USD"
                },
                "merchant": {
                    "merchant_name": "Amazon",
                    "merchant_category_code": "5732"
                },
                "customer": {
                    "customer_id": "CUST-001",
                    "age_of_account_days": 365
                }
            }
        }

class ProcessResponse(BaseModel):
    result: Any = Field(..., description="Processing result")
    success: bool = Field(..., description="Whether processing was successful")
    metadata: Dict[str, Any] = Field(..., description="Response metadata")
    
    class Config:
        schema_extra = {
            "example": {
                "result": "I can help you with various tasks...",
                "success": True,
                "metadata": {
                    "agent": "risk_manager",
                    "framework": "langgraph",
                    "processing_time_ms": 150
                }
            }
        }

class HealthResponse(BaseModel):
    status: str = Field(..., description="Health status")
    agent: str = Field(..., description="Agent name")
    framework: str = Field(..., description="Framework used")
    uptime: str = Field(..., description="Application uptime")

class GraphInfoResponse(BaseModel):
    graph_info: Dict[str, Any] = Field(..., description="Graph structure information")

# Exception handlers
@app.exception_handler(HTTPException)
async def http_exception_handler(request: Request, exc: HTTPException):
    return JSONResponse(
        status_code=exc.status_code,
        content={"detail": exc.detail, "status_code": exc.status_code}
    )

@app.exception_handler(Exception)
async def general_exception_handler(request: Request, exc: Exception):
    return JSONResponse(
        status_code=500,
        content={"detail": "Internal server error", "error": str(exc)}
    )

# API Routes
@app.get("/", tags=["Root"])
async def root():
    """Root endpoint"""
    return {
        "message": "Welcome to risk_manager API",
        "framework": "langgraph",
        "docs": "/docs",
        "health": "/health"
    }

@app.post("/process", response_model=ProcessResponse, tags=["Agent"])
async def process_endpoint(request: ProcessRequest):
    """
    Main processing endpoint - sends input through the LangGraph agent
    This is the main entry point for agent execution, so it has tracing.
    Accepts flexible transaction data in multiple formats.
    Can process single transactions or arrays of transactions.
    
    Accepts session_id and run_id as parameters for tracking multiple runs.
    """
    if not agent:
        raise HTTPException(status_code=503, detail="Agent not initialized")
    
    try:
        import time
        import uuid
        start_time = time.time()
        
        # Check if this is a batch request (array of transactions)
        if request.transactions:
            from src.utils import get_generated_bullets, clear_generated_bullets
            
            # Clear generated bullets at the start of processing
            clear_generated_bullets()
            
            # Process multiple transactions
            results = []
            for idx, tx in enumerate(request.transactions):
                try:
                    # Ensure transaction has an ID
                    if "transaction_id" not in tx:
                        tx["transaction_id"] = f"TXN-{uuid.uuid4().hex[:12].upper()}"
                    
                    # Process through the agent with specified model types
                    transaction_results = {}
                    for model_type in request.model_types:
                        try:
                            # Run once per model type
                            result = await agent.process(tx, model_type=model_type, session_id=request.session_id, run_id=request.run_id)
                            
                            # Extract only analyzer results and final decision
                            final_output = {}

                            # If result is a dict containing 'results' (from graph execution)
                            if isinstance(result, dict):
                                # Get the results section which contains analyzer outputs
                                if 'results' in result:
                                    analyzer_results = result['results']

                                    # Include all analyzer outputs
                                    for analyzer in ['pattern_detector', 'behavioral_analizer', 'velocity_checker',
                                                   'merchant_risk_analizer', 'geographic_analizer']:
                                        if analyzer in analyzer_results:
                                            final_output[analyzer] = analyzer_results[analyzer]

                                    # Include the decision aggregator (final decision)
                                    if 'decision_aggregator' in analyzer_results:
                                        final_output['decision'] = analyzer_results['decision_aggregator']

                                # Alternative: Check if analyzers are directly in result
                                else:
                                    for key in ['pattern_detector', 'behavioral_analizer', 'velocity_checker',
                                               'merchant_risk_analizer', 'geographic_analizer', 'decision_aggregator']:
                                        if key in result:
                                            if key == 'decision_aggregator':
                                                final_output['decision'] = result[key]
                                            else:
                                                final_output[key] = result[key]

                            # If no analyzers found, return the raw result
                            transaction_results[model_type] = final_output if final_output else result
                        except Exception as e:
                            transaction_results[model_type] = {"error": str(e)}
                    
                    results.append({
                        "transaction_id": tx.get("transaction_id"),
                        "results": transaction_results
                    })
                except Exception as e:
                    results.append({"error": str(e), "transaction_id": tx.get("transaction_id", "unknown")})
            
            processing_time = (time.time() - start_time) * 1000
            
            # Get generated bullets
            generated_bullets = get_generated_bullets()
            
            # Fetch metrics if session_id is provided
            metrics = {}
            if request.session_id:
                metrics = await fetch_metrics(request.session_id)
            
            return ProcessResponse(
                result=results,
                success=True,
                metadata={
                    "agent": "risk_manager",
                    "framework": "langgraph",
                    "processing_time_ms": round(processing_time, 2),
                    "transaction_count": len(request.transactions),
                    "model_types": request.model_types,
                    "session_id": request.session_id,
                    "run_id": request.run_id,
                    "metrics": metrics,
                    "generated_bullets": generated_bullets,
                    **request.metadata
                }
            )

        # Single transaction processing (existing logic)
        from src.utils import get_generated_bullets, clear_generated_bullets
        
        # Clear generated bullets at the start of processing
        clear_generated_bullets()
        
        # Build transaction data from whatever format is provided
        transaction_data = {}

        # Option 1: If input_data is provided (backwards compatibility)
        if request.input_data:
            if isinstance(request.input_data, dict):
                transaction_data = request.input_data
            elif isinstance(request.input_data, str):
                # Try to parse as JSON if it's a string
                try:
                    import json
                    transaction_data = json.loads(request.input_data)
                except:
                    # If not JSON, create a simple transaction
                    transaction_data = {"raw_input": request.input_data}

        # Option 2: Build from structured fields (complex transaction)
        elif request.transaction or request.financial or request.merchant:
            # Complex transaction format
            if request.transaction:
                transaction_data.update(request.transaction)
            if request.financial and "amount" in request.financial:
                transaction_data["amount"] = request.financial["amount"]
                transaction_data["currency"] = request.financial.get("currency", "USD")
            if request.card:
                transaction_data["card_info"] = request.card
            if request.merchant:
                if "merchant_name" in request.merchant:
                    transaction_data["merchant"] = request.merchant["merchant_name"]
                    transaction_data["merchant_category_code"] = request.merchant.get("merchant_category_code")
                transaction_data["merchant_data"] = request.merchant
            if request.customer:
                if "customer_id" in request.customer:
                    transaction_data["user_id"] = request.customer["customer_id"]
                if "age_of_account_days" in request.customer:
                    transaction_data["user_age_days"] = request.customer["age_of_account_days"]
                transaction_data["customer_data"] = request.customer
            if request.device:
                transaction_data["device_data"] = request.device
            if request.location:
                if "transaction_city" in request.location:
                    transaction_data["location"] = f"{request.location.get('transaction_city', '')}, {request.location.get('transaction_country', '')}"
                transaction_data["location_data"] = request.location
            if request.behavioral_profile:
                transaction_data["behavioral_profile"] = request.behavioral_profile
            if request.velocity_counters:
                transaction_data["velocity_counters"] = request.velocity_counters
            if request.risk_signals:
                transaction_data["risk_signals"] = request.risk_signals
            if request.session:
                transaction_data["session_data"] = request.session
            if request.channel:
                transaction_data["channel_data"] = request.channel
            if request.authentication:
                transaction_data["authentication_data"] = request.authentication

        # Option 3: Build from simple fields (basic transaction)
        elif request.user_id or request.amount:
            transaction_data = {
                "user_id": request.user_id or f"user_{uuid.uuid4().hex[:8]}",
                "user_age_days": request.user_age_days or 180,
                "total_transactions": request.total_transactions or 10,
                "amount": request.amount or 100.0,
                "time": request.time or "14:00",
                "merchant": request.merchant_name or "Unknown Merchant",
                "merchant_category": request.merchant_category or "General",
                "currency": request.currency or "USD",
                "location": "Unknown",
                "previous_location": "Unknown"
            }

        # Option 4: Use the raw request dict if nothing else works
        else:
            # Get all non-None fields from the request
            request_dict = request.dict(exclude_none=True, exclude={"metadata"})
            if request_dict:
                transaction_data = request_dict
            else:
                # Default minimal transaction
                transaction_data = {
                    "user_id": f"user_{uuid.uuid4().hex[:8]}",
                    "amount": 100.0,
                    "merchant": "Test Merchant",
                    "user_age_days": 180,
                    "total_transactions": 10
                }

        # Ensure transaction has an ID
        if "transaction_id" not in transaction_data:
            transaction_data["transaction_id"] = f"TXN-{uuid.uuid4().hex[:12].upper()}"

        # Process through the agent with specified model types
        transaction_results = {}
        for model_type in request.model_types:
            try:
                # Run once per model type
                result = await agent.process(transaction_data, model_type=model_type, session_id=request.session_id, run_id=request.run_id)
                
                # Extract only analyzer results and final decision
                final_output = {}

                # If result is a dict containing 'results' (from graph execution)
                if isinstance(result, dict):
                    # Get the results section which contains analyzer outputs
                    if 'results' in result:
                        analyzer_results = result['results']

                        # Include all analyzer outputs
                        for analyzer in ['pattern_detector', 'behavioral_analizer', 'velocity_checker',
                                       'merchant_risk_analizer', 'geographic_analizer']:
                            if analyzer in analyzer_results:
                                final_output[analyzer] = analyzer_results[analyzer]

                        # Include the decision aggregator (final decision)
                        if 'decision_aggregator' in analyzer_results:
                            final_output['decision'] = analyzer_results['decision_aggregator']

                    # Alternative: Check if analyzers are directly in result
                    else:
                        for key in ['pattern_detector', 'behavioral_analizer', 'velocity_checker',
                                   'merchant_risk_analizer', 'geographic_analizer', 'decision_aggregator']:
                            if key in result:
                                if key == 'decision_aggregator':
                                    final_output['decision'] = result[key]
                                else:
                                    final_output[key] = result[key]

                # If no analyzers found, return the raw result
                transaction_results[model_type] = final_output if final_output else result
            except Exception as e:
                transaction_results[model_type] = {"error": str(e)}

        processing_time = (time.time() - start_time) * 1000
        
        # Get generated bullets
        generated_bullets = get_generated_bullets()
        
        # Fetch metrics if session_id is provided
        metrics = {}
        if request.session_id:
            metrics = await fetch_metrics(request.session_id)
        
        return ProcessResponse(
            result={
                "transaction_id": transaction_data.get("transaction_id"),
                "results": transaction_results
            },
            success=True,
            metadata={
                "agent": "risk_manager",
                "framework": "langgraph",
                "processing_time_ms": round(processing_time, 2),
                "model_types": request.model_types,
                "session_id": request.session_id,
                "run_id": request.run_id,
                "metrics": metrics,
                "generated_bullets": generated_bullets,
                **request.metadata
            }
        )
    except Exception as e:
        raise HTTPException(
            status_code=500,
            detail=f"Processing failed: {str(e)}"
        )

@app.get("/health", response_model=HealthResponse, tags=["Health"])
async def health_check():
    """Health check endpoint"""
    import time
    uptime = time.time() - start_time if 'start_time' in globals() else 0
    
    return HealthResponse(
        status="healthy",
        agent="risk_manager",
        framework="langgraph",
        uptime=f"{uptime:.2f} seconds"
    )

@app.get("/graph/info", response_model=GraphInfoResponse, tags=["Graph"])
async def graph_info():
    """Get graph structure information"""
    if not agent:
        raise HTTPException(status_code=503, detail="Agent not initialized")
    
    return GraphInfoResponse(
        graph_info=agent.get_graph_info()
    )

# Development server
if __name__ == "__main__":
    import time
    start_time = time.time()
    
    port = 8001
    host = os.getenv("HOST", "0.0.0.0")
    
    print(f"🌐 Starting risk_manager FastAPI server")
    print(f"📍 Server will be available at: http://{host}:{port}")
    print(f"📚 API Documentation: http://{host}:{port}/docs")
    print(f"🔍 Alternative docs: http://{host}:{port}/redoc")
    
    uvicorn.run(
        "main:app",
        host=host,
        port=port,
        reload=os.getenv("ENVIRONMENT") == "development",
        log_level=os.getenv("LOG_LEVEL", "info").lower()
    )