source/src/app/api/chat.py

from fastapi import APIRouter, HTTPException
from fastapi.responses import StreamingResponse
import json
import asyncio
import logging
from src.domain.transportation_graph import create_transportation_workflow
from src.state.graph_state import TransportationState
from src.app.schema.chat import ChatRequest, StreamChunk, FunctionCall
from src.config.logging_config import get_logger

logger = get_logger(__name__)
router = APIRouter()

@router.post("/chat", response_model=None)
async def chat_stream(request: ChatRequest):
    """Streaming chat endpoint sử dụng LangGraph native streaming với LangChain model"""
    try:
        async def generate_stream():
            """Generate streaming từ LangGraph với stream_mode='messages'"""
            # Tạo graph workflow
            workflow = create_transportation_workflow()
            graph = workflow.compile()
            
            # Tạo initial state với user message
            from src.state.graph_state import create_initial_state
            initial_state = create_initial_state(request.message)
            
            # Emit status starting
            chunk = StreamChunk(
                event="status",
                stage="starting",
                message="Đang xử lý câu hỏi..."
            )
            yield f"data: {chunk.model_dump_json()}\n\n"
            
            try:
                accumulated_text = ""
                final_state = None
                
                logger.info("Using LangGraph native streaming with stream_mode='messages'")
                
                # Sử dụng LangGraph streaming
                async for event in graph.astream(initial_state, stream_mode="messages"):
                    logger.debug(f"Graph event: {type(event)}")
                    
                    # Check for message content from LLM
                    if isinstance(event, tuple) and len(event) == 2:
                        content_obj, metadata = event
                        
                        # Check if this is from llm node
                        if (hasattr(content_obj, 'content') and 
                            content_obj.content and 
                            metadata.get('langgraph_node') == 'llm'):
                            
                            content = content_obj.content
                            accumulated_text += content
                            chunk_obj = StreamChunk(
                                event="delta",
                                content=content
                            )
                            yield f"data: {chunk_obj.model_dump_json()}\n\n"
                            await asyncio.sleep(0.02)
                
                logger.info(f"Graph streaming completed. Accumulated: {len(accumulated_text)} chars")
                
                # Nếu không có streaming content, fallback to invoke
                if not accumulated_text.strip():
                    logger.info("No streaming content, falling back to invoke")
                    final_state = graph.invoke(initial_state)
                    
                    if 'ai_response' in final_state and final_state['ai_response']:
                        # Simulate streaming effect
                        words = final_state['ai_response'].split()
                        for word in words:
                            accumulated_text += word + " "
                            chunk_obj = StreamChunk(
                                event="delta", 
                                content=word + " "
                            )
                            yield f"data: {chunk_obj.model_dump_json()}\n\n"
                            await asyncio.sleep(0.05)
                        accumulated_text = accumulated_text.strip()
                    else:
                        accumulated_text = "Đã xảy ra lỗi khi xử lý yêu cầu."
                else:
                    # Get final state for function calls and cache
                    try:
                        final_state = graph.invoke(initial_state)
                    except Exception as e:
                        logger.warning(f"Graph invoke for final state failed: {e}")
                
                # Include function calls nếu có (tạm thời bỏ qua để đơn giản)
                # if final_state and 'function_calls_made' in final_state and final_state['function_calls_made']:
                #     pass  # Có thể thêm function calls event sau
                
                final_response = accumulated_text if accumulated_text.strip() else "Không có phản hồi"
                        
            except Exception as e:
                logger.error(f"Graph streaming error: {e}")
                chunk = StreamChunk(
                    event="status",
                    stage="error",
                    message=f"Streaming Error: {str(e)}"
                )
                yield f"data: {chunk.model_dump_json()}\n\n"
                final_response = "Đã xảy ra lỗi trong quá trình xử lý"
            
            # Kết thúc stream
            final_chunk = StreamChunk(
                event="final",
                content=final_response
            )
            yield f"data: {final_chunk.model_dump_json()}\n\n"
            yield "data: [DONE]\n\n"
        
        return StreamingResponse(
            generate_stream(),
            media_type="text/event-stream",
            headers={
                "Cache-Control": "no-cache",
                "Connection": "keep-alive",
                "Access-Control-Allow-Origin": "*",
                "Access-Control-Allow-Headers": "Cache-Control"
            }
        )
        
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Streaming failed: {str(e)}")