""" Flare – Main Application (Refactored with Event-Driven Architecture) ==================================================================== """ # FastAPI imports from fastapi import FastAPI, WebSocket, Request, status from fastapi.staticfiles import StaticFiles from fastapi.responses import FileResponse, JSONResponse from fastapi.middleware.cors import CORSMiddleware from fastapi.encoders import jsonable_encoder # Standard library import uvicorn import os from pathlib import Path import mimetypes import uuid import traceback from datetime import datetime import asyncio import time from pydantic import ValidationError from dotenv import load_dotenv # Event-driven architecture imports from chat_session.event_bus import event_bus from chat_session.state_orchestrator import StateOrchestrator from chat_session.websocket_manager import WebSocketManager from chat_session.resource_manager import ResourceManager from stt.stt_lifecycle_manager import STTLifecycleManager from tts.tts_lifecycle_manager import TTSLifecycleManager from llm.llm_lifecycle_manager import LLMLifecycleManager from stt.audio_buffer_manager import AudioBufferManager # Project imports from routes.admin_routes import router as admin_router, start_cleanup_task from llm.llm_startup import run_in_thread from chat_session.session import session_store, start_session_cleanup from config.config_provider import ConfigProvider # Logger imports from utils.logger import log_error, log_info, log_warning # Exception imports from utils.exceptions import ( DuplicateResourceError, RaceConditionError, ValidationError as FlareValidationError, ResourceNotFoundError, AuthenticationError, AuthorizationError, ConfigurationError, get_http_status_code, FlareException ) # Load .env file if exists load_dotenv() ALLOWED_ORIGINS = os.getenv("ALLOWED_ORIGINS", "http://localhost:4200").split(",") # ===================== Environment Setup ===================== def setup_environment(): """Setup environment based on deployment mode""" cfg = ConfigProvider.get() log_info("=" * 60) log_info("🚀 Flare Starting", version="2.0.0") log_info(f"🔌 LLM Provider: {cfg.global_config.llm_provider.name}") log_info(f"🎤 TTS Provider: {cfg.global_config.tts_provider.name}") log_info(f"🎧 STT Provider: {cfg.global_config.stt_provider.name}") log_info("=" * 60) if cfg.global_config.is_cloud_mode(): log_info("☁️ Cloud Mode: Using HuggingFace Secrets") log_info("📌 Required secrets: JWT_SECRET, FLARE_TOKEN_KEY") # Check for provider-specific tokens llm_config = cfg.global_config.get_provider_config("llm", cfg.global_config.llm_provider.name) if llm_config and llm_config.requires_repo_info: log_info("📌 LLM requires SPARK_TOKEN for repository operations") else: log_info("🏢 On-Premise Mode: Using .env file") if not Path(".env").exists(): log_warning("⚠️ WARNING: .env file not found!") log_info("📌 Copy .env.example to .env and configure it") # Run setup setup_environment() # Fix MIME types for JavaScript files mimetypes.add_type("application/javascript", ".js") mimetypes.add_type("text/css", ".css") app = FastAPI( title="Flare Orchestration Service", version="2.0.0", description="LLM-driven intent & API flow engine with multi-provider support", ) # CORS for development if os.getenv("ENVIRONMENT", "development") == "development": app.add_middleware( CORSMiddleware, allow_origins=ALLOWED_ORIGINS, allow_credentials=True, allow_methods=["GET", "POST", "PUT", "DELETE", "OPTIONS"], allow_headers=["*"], max_age=3600, expose_headers=["X-Request-ID"] ) log_info(f"🔧 CORS enabled for origins: {ALLOWED_ORIGINS}") # Request ID middleware @app.middleware("http") async def add_request_id(request: Request, call_next): """Add request ID for tracking""" request_id = str(uuid.uuid4()) request.state.request_id = request_id # Log request start log_info( "Request started", request_id=request_id, method=request.method, path=request.url.path, client=request.client.host if request.client else "unknown" ) try: response = await call_next(request) # Add request ID to response headers response.headers["X-Request-ID"] = request_id # Log request completion log_info( "Request completed", request_id=request_id, status_code=response.status_code, method=request.method, path=request.url.path ) return response except Exception as e: log_error( "Request failed", request_id=request_id, error=str(e), traceback=traceback.format_exc() ) raise # ===================== Event-Driven Architecture Initialization ===================== @app.on_event("startup") async def startup_event(): """Initialize event-driven components on startup""" try: # Initialize event bus await event_bus.start() log_info("✅ Event bus started") # Initialize resource manager resource_manager = ResourceManager(event_bus) await resource_manager.start() log_info("✅ Resource manager started") # Initialize managers state_orchestrator = StateOrchestrator(event_bus) websocket_manager = WebSocketManager(event_bus) audio_buffer_manager = AudioBufferManager(event_bus) stt_manager = STTLifecycleManager(event_bus, resource_manager) tts_manager = TTSLifecycleManager(event_bus, resource_manager) llm_manager = LLMLifecycleManager(event_bus, resource_manager) # Store in app state for access in routes app.state.event_bus = event_bus app.state.resource_manager = resource_manager app.state.state_orchestrator = state_orchestrator app.state.websocket_manager = websocket_manager app.state.audio_buffer_manager = audio_buffer_manager app.state.stt_manager = stt_manager app.state.tts_manager = tts_manager app.state.llm_manager = llm_manager log_info("✅ All managers initialized") # Start existing background tasks run_in_thread() # Start LLM startup notifier if needed start_cleanup_task() # Activity log cleanup start_session_cleanup() # Session cleanup log_info("✅ Background tasks started") except Exception as e: log_error("❌ Failed to start event-driven components", error=str(e), traceback=traceback.format_exc()) raise @app.on_event("shutdown") async def shutdown_event(): """Cleanup event-driven components on shutdown""" try: # Stop event bus await event_bus.stop() log_info("✅ Event bus stopped") # Stop resource manager if hasattr(app.state, 'resource_manager'): await app.state.resource_manager.stop() log_info("✅ Resource manager stopped") # Close all WebSocket connections if hasattr(app.state, 'websocket_manager'): await app.state.websocket_manager.close_all_connections() log_info("✅ All WebSocket connections closed") except Exception as e: log_error("❌ Error during shutdown", error=str(e)) # ---------------- Core chat/session routes -------------------------- from routes.chat_handler import router as chat_router app.include_router(chat_router, prefix="/api") # ---------------- Audio (TTS/STT) routes ------------------------------ from routes.audio_routes import router as audio_router app.include_router(audio_router, prefix="/api") # ---------------- Admin API routes ---------------------------------- app.include_router(admin_router, prefix="/api/admin") # ---------------- WebSocket route for real-time chat ------------------ @app.websocket("/ws/conversation/{session_id}") async def websocket_route(websocket: WebSocket, session_id: str): """Handle WebSocket connections using the new WebSocketManager""" if hasattr(app.state, 'websocket_manager'): await app.state.websocket_manager.handle_connection(websocket, session_id) else: log_error("WebSocketManager not initialized") await websocket.close(code=1011, reason="Server not ready") # ---------------- Test endpoint for event-driven flow ------------------ @app.post("/api/test/realtime") async def test_realtime(): """Test endpoint for event-driven realtime flow""" from chat_session.event_bus import Event, EventType try: # Create a test session session = session_store.create_session( project_name="kronos_jet", version_no=1, is_realtime=True ) # Get version config cfg = ConfigProvider.get() project = next((p for p in cfg.projects if p.name == "kronos_jet"), None) if project: version = next((v for v in project.versions if v.no == 1), None) if version: session.set_version_config(version) # Publish session started event await app.state.event_bus.publish(Event( type=EventType.SESSION_STARTED, session_id=session.session_id, data={ "session": session, "has_welcome": bool(version and version.welcome_prompt), "welcome_text": version.welcome_prompt if version and version.welcome_prompt else "Hoş geldiniz!" } )) return { "session_id": session.session_id, "message": "Test session created. Connect via WebSocket to continue." } except Exception as e: log_error("Test endpoint error", error=str(e)) raise HTTPException(500, f"Test failed: {str(e)}") # ---------------- Exception Handlers ---------------------------------- @app.exception_handler(Exception) async def global_exception_handler(request: Request, exc: Exception): """Handle all unhandled exceptions""" request_id = getattr(request.state, 'request_id', 'unknown') # Log the full exception log_error( "Unhandled exception", request_id=request_id, endpoint=str(request.url), method=request.method, error=str(exc), error_type=type(exc).__name__, traceback=traceback.format_exc() ) # Special handling for FlareExceptions if isinstance(exc, FlareException): status_code = get_http_status_code(exc) response_body = { "error": type(exc).__name__, "message": str(exc), "request_id": request_id, "timestamp": datetime.utcnow().isoformat(), "details": getattr(exc, 'details', {}) } # Special message for race conditions if isinstance(exc, RaceConditionError): response_body["user_action"] = "Please reload the data and try again" return JSONResponse( status_code=status_code, content=jsonable_encoder(response_body) ) # Generic error response return JSONResponse( status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, content=jsonable_encoder({ "error": "InternalServerError", "message": "An unexpected error occurred. Please try again later.", "request_id": request_id, "timestamp": datetime.utcnow().isoformat() }) ) # Add custom exception handlers @app.exception_handler(DuplicateResourceError) async def duplicate_resource_handler(request: Request, exc: DuplicateResourceError): """Handle duplicate resource errors""" return JSONResponse( status_code=409, content={ "detail": str(exc), "error_type": "duplicate_resource", "resource_type": exc.details.get("resource_type"), "identifier": exc.details.get("identifier") } ) @app.exception_handler(RaceConditionError) async def race_condition_handler(request: Request, exc: RaceConditionError): """Handle race condition errors""" return JSONResponse( status_code=409, content=exc.to_http_detail() ) @app.exception_handler(FlareValidationError) async def validation_error_handler(request: Request, exc: FlareValidationError): """Handle validation errors""" return JSONResponse( status_code=422, content={ "detail": str(exc), "error_type": "validation_error", "details": exc.details } ) @app.exception_handler(ResourceNotFoundError) async def resource_not_found_handler(request: Request, exc: ResourceNotFoundError): """Handle resource not found errors""" return JSONResponse( status_code=404, content={ "detail": str(exc), "error_type": "resource_not_found", "resource_type": exc.details.get("resource_type"), "identifier": exc.details.get("identifier") } ) @app.exception_handler(AuthenticationError) async def authentication_error_handler(request: Request, exc: AuthenticationError): """Handle authentication errors""" return JSONResponse( status_code=401, content={ "detail": str(exc), "error_type": "authentication_error" } ) @app.exception_handler(AuthorizationError) async def authorization_error_handler(request: Request, exc: AuthorizationError): """Handle authorization errors""" return JSONResponse( status_code=403, content={ "detail": str(exc), "error_type": "authorization_error" } ) @app.exception_handler(ConfigurationError) async def configuration_error_handler(request: Request, exc: ConfigurationError): """Handle configuration errors""" return JSONResponse( status_code=500, content={ "detail": str(exc), "error_type": "configuration_error", "config_key": exc.details.get("config_key") } ) # ---------------- Metrics endpoint ----------------- @app.get("/metrics") async def get_metrics(): """Get system metrics including event-driven components""" import psutil import gc # Memory info process = psutil.Process() memory_info = process.memory_info() # Session stats session_stats = session_store.get_session_stats() # Event-driven component stats event_stats = {} if hasattr(app.state, 'stt_manager'): event_stats['stt'] = app.state.stt_manager.get_stats() if hasattr(app.state, 'tts_manager'): event_stats['tts'] = app.state.tts_manager.get_stats() if hasattr(app.state, 'llm_manager'): event_stats['llm'] = app.state.llm_manager.get_stats() if hasattr(app.state, 'websocket_manager'): event_stats['websocket'] = { 'active_connections': app.state.websocket_manager.get_connection_count() } if hasattr(app.state, 'resource_manager'): event_stats['resources'] = app.state.resource_manager.get_stats() if hasattr(app.state, 'audio_buffer_manager'): event_stats['audio_buffers'] = app.state.audio_buffer_manager.get_all_stats() metrics = { "memory": { "rss_mb": memory_info.rss / 1024 / 1024, "vms_mb": memory_info.vms / 1024 / 1024, "percent": process.memory_percent() }, "cpu": { "percent": process.cpu_percent(interval=0.1), "num_threads": process.num_threads() }, "sessions": session_stats, "event_driven_components": event_stats, "gc": { "collections": gc.get_count(), "objects": len(gc.get_objects()) }, "uptime_seconds": time.time() - process.create_time() } return metrics # ---------------- Health probe (HF Spaces watchdog) ----------------- @app.get("/api/health") def health_check(): """Health check endpoint - moved to /api/health""" # Check if event-driven components are healthy event_bus_healthy = hasattr(app.state, 'event_bus') and app.state.event_bus._running return { "status": "ok" if event_bus_healthy else "degraded", "version": "2.0.0", "timestamp": datetime.utcnow().isoformat(), "environment": os.getenv("ENVIRONMENT", "development"), "event_driven": { "event_bus": "running" if event_bus_healthy else "not_running", "managers": { "state_orchestrator": "initialized" if hasattr(app.state, 'state_orchestrator') else "not_initialized", "websocket_manager": "initialized" if hasattr(app.state, 'websocket_manager') else "not_initialized", "stt_manager": "initialized" if hasattr(app.state, 'stt_manager') else "not_initialized", "tts_manager": "initialized" if hasattr(app.state, 'tts_manager') else "not_initialized", "llm_manager": "initialized" if hasattr(app.state, 'llm_manager') else "not_initialized" } } } # ---------------- Serve static files ------------------------------------ # UI static files (production build) static_path = Path(__file__).parent / "static" if static_path.exists(): app.mount("/static", StaticFiles(directory=str(static_path)), name="static") # Serve index.html for all non-API routes (SPA support) @app.get("/", response_class=FileResponse) async def serve_index(): """Serve Angular app""" index_path = static_path / "index.html" if index_path.exists(): return FileResponse(str(index_path)) else: return JSONResponse( status_code=404, content={"error": "UI not found. Please build the Angular app first."} ) # Catch-all route for SPA @app.get("/{full_path:path}") async def serve_spa(full_path: str): """Serve Angular app for all routes""" # Skip API routes if full_path.startswith("api/"): return JSONResponse(status_code=404, content={"error": "Not found"}) # Serve static files file_path = static_path / full_path if file_path.exists() and file_path.is_file(): return FileResponse(str(file_path)) # Fallback to index.html for SPA routing index_path = static_path / "index.html" if index_path.exists(): return FileResponse(str(index_path)) return JSONResponse(status_code=404, content={"error": "Not found"}) else: log_warning(f"⚠️ Static files directory not found at {static_path}") log_warning(" Run 'npm run build' in flare-ui directory to build the UI") @app.get("/") async def no_ui(): """No UI available""" return JSONResponse( status_code=503, content={ "error": "UI not available", "message": "Please build the Angular UI first. Run: cd flare-ui && npm run build", "api_docs": "/docs" } ) if __name__ == "__main__": log_info("🌐 Starting Flare backend on port 7860...") uvicorn.run(app, host="0.0.0.0", port=7860)