import os
import json
import time
import asyncio
import aiohttp
from typing import Dict, List, Set, Optional
from urllib.parse import quote, urljoin
from datetime import datetime
from pathlib import Path
from datasets import Dataset, DatasetDict
import huggingface_hub

from fastapi import FastAPI, BackgroundTasks, HTTPException, status
from fastapi.responses import JSONResponse
from pydantic import BaseModel, Field
import uvicorn
import aiohttp

# Path for storing caption data
CAPTIONS_DIR = Path("captions_data")
CAPTIONS_DIR.mkdir(exist_ok=True)

# Hugging Face configuration
HF_TOKEN = os.getenv("HF_TOKEN")
HF_DATASET_ID = os.getenv("HF_DATASET_ID", "fred808/helium")

if not HF_TOKEN:
    raise ValueError("HF_TOKEN environment variable is required")

def get_caption_file_path(course: str) -> Path:
    """Get the path to the JSON file for storing course captions"""
    safe_name = quote(course, safe='')
    return CAPTIONS_DIR / f"{safe_name}_captions.json"

def save_captions_to_file(course: str, captions: List[Dict]) -> None:
    """Save captions to a JSON file"""
    try:
        file_path = get_caption_file_path(course)
        with open(file_path, 'w', encoding='utf-8') as f:
            json.dump(captions, f, indent=2, ensure_ascii=False)
        print(f"✓ Saved {len(captions)} captions for {course}")
    except Exception as e:
        print(f"Error saving captions for {course}: {e}")

def load_captions_from_file(course: str) -> List[Dict]:
    """Load existing captions from JSON file"""
    try:
        file_path = get_caption_file_path(course)
        if file_path.exists():
            with open(file_path, 'r', encoding='utf-8') as f:
                captions = json.load(f)
                print(f"✓ Loaded {len(captions)} existing captions for {course}")
                return captions
    except Exception as e:
        print(f"Error loading captions for {course}: {e}")
    return []

# Configuration
SOURCE_SERVER = "https://samelias1-vs2.hf.space"
CAPTION_SERVERS = [
    "https://fred808-pil-4-1.hf.space/analyze",
    "https://fred808-pil-4-2.hf.space/analyze",
    "https://fred808-pil-4-3.hf.space/analyze",
    "https://fred1012-fred1012-gw0j2h.hf.space/analyze",
    "https://fred1012-fred1012-wqs6c2.hf.space/analyze",
    "https://fred1012-fred1012-oncray.hf.space/analyze",
    "https://fred1012-fred1012-4goge7.hf.space/analyze",
    "https://fred1012-fred1012-z0eh7m.hf.space/analyze",
    "https://fred1012-fred1012-u95rte.hf.space/analyze",
    "https://fred1012-fred1012-igje22.hf.space/analyze",
    "https://fred1012-fred1012-ibkuf8.hf.space/analyze",
    "https://fred1012-fred1012-nwqthy.hf.space/analyze",
    "https://fred1012-fred1012-4ldqj4.hf.space/analyze",
    "https://fred1012-fred1012-pivlzg.hf.space/analyze",
    "https://fred1012-fred1012-ptlc5u.hf.space/analyze",
    "https://fred1012-fred1012-u7lh57.hf.space/analyze",
    "https://fred1012-fred1012-q8djv1.hf.space/analyze",
    "https://fredalone-fredalone-ozugrp.hf.space/analyze",
    "https://fredalone-fredalone-9brxj2.hf.space/analyze",
    "https://fredalone-fredalone-p8vq9a.hf.space/analyze",
    "https://fredalone-fredalone-vbli2y.hf.space/analyze",
    "https://fredalone-fredalone-uggger.hf.space/analyze",
    "https://fredalone-fredalone-nmi7e8.hf.space/analyze",
    "https://fredalone-fredalone-d1f26d.hf.space/analyze",
    "https://fredalone-fredalone-461jp2.hf.space/analyze",
    "https://fredalone-fredalone-3enfg4.hf.space/analyze",
    "https://fredalone-fredalone-dqdbpv.hf.space/analyze",
    "https://fredalone-fredalone-ivtjua.hf.space/analyze",
    "https://fredalone-fredalone-6bezt2.hf.space/analyze",
    "https://fredalone-fredalone-e0wfnk.hf.space/analyze",
    "https://fredalone-fredalone-zu2t7j.hf.space/analyze",
    "https://fredalone-fredalone-dqtv1o.hf.space/analyze",
    "https://fredalone-fredalone-wclyog.hf.space/analyze",
    "https://fredalone-fredalone-t27vig.hf.space/analyze",
    "https://fredalone-fredalone-gahbxh.hf.space/analyze",
    "https://fredalone-fredalone-kw2po4.hf.space/analyze",
    "https://fredalone-fredalone-8h285h.hf.space/analyze"
]
MODEL_TYPE = "Florence-2-large"  # Explicitly request large model

# FastAPI Models
class CourseInfo(BaseModel):
    course_folder: str

class ImageInfo(BaseModel):
    filename: str

class CaptionRequest(BaseModel):
    image_url: str
    model_choice: str = MODEL_TYPE

class CaptionResponse(BaseModel):
    success: bool
    caption: Optional[str] = None
    error: Optional[str] = None

class ServerStatus(BaseModel):
    url: str
    model: str
    busy: bool
    total_processed: int
    total_time: float
    fps: float

class ProcessingStatus(BaseModel):
    course: str
    total_images: int
    processed_images: int
    progress_percent: float
    status: str

class StartProcessingRequest(BaseModel):
    courses: Optional[List[str]] = None  # If None, process all courses
    continuous: bool = True  # Default to continuous like original

# FastAPI App
app = FastAPI(
    title="Caption Coordinator API",
    description="Distributed caption processing coordinator",
    version="1.0.0"
)

# Global state
processed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
course_captions: Dict[str, List[Dict]] = {}  # {course: [{image, caption, metadata}]}
failed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
servers = []
is_processing = False
current_processing_task = None
auto_start_processing = True  # Set to False if you don't want auto-start

# Map of course -> vs2 callback URL
pending_vs2_callbacks: Dict[str, str] = {}

class CaptionServer:
    def __init__(self, url):
        self.url = url
        self.busy = False
        self.model = "unknown"
        self.total_processed = 0
        self.total_time = 0

    @property
    def fps(self):
        return self.total_processed / self.total_time if self.total_time > 0 else 0

# Initialize servers
def initialize_servers():
    global servers
    servers = [CaptionServer(url) for url in CAPTION_SERVERS]

# API Routes
@app.get("/")
async def root():
    return {
        "message": "Caption Coordinator API", 
        "status": "running",
        "auto_processing": auto_start_processing,
        "is_processing": is_processing
    }

@app.get("/health")
async def health():
    return {
        "status": "healthy",
        "servers_available": len([s for s in servers if not s.busy]),
        "total_servers": len(servers),
        "is_processing": is_processing,
        "auto_processing": auto_start_processing
    }

@app.get("/courses")
async def get_courses():
    """Fetch available courses from source server"""
    try:
        async with aiohttp.ClientSession() as session:
            async with session.get(f"{SOURCE_SERVER}/courses") as resp:
                data = await resp.json()
                if isinstance(data, dict) and 'courses' in data:
                    return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
                return []
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error fetching courses: {e}")


@app.post("/vs2/register")
async def vs2_register(payload: Dict):
    """Register a VS2 callback and optionally start processing for the given course.
    Expected payload: {"course": "course_name", "callback_url": "http://vs2-host/flow/done", "start": true}
    """
    try:
        course = payload.get("course")
        callback = payload.get("callback_url")
        start = payload.get("start", True)

        if not callback:
            raise HTTPException(status_code=400, detail="callback_url is required")

        # Store callback for later notification
        if course:
            pending_vs2_callbacks[course] = callback
        else:
            # store under wildcard key if course not provided
            pending_vs2_callbacks["*"] = callback

        # If caller asks to start processing this course immediately, and we're not currently processing,
        # kick off a one-shot processing loop for that course.
        if start:
            global is_processing, current_processing_task
            if not is_processing:
                is_processing = True
                current_processing_task = asyncio.create_task(processing_loop([course] if course else None, False))

        return {"registered": True, "course": course}
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/courses/{course}/images")
async def get_course_images(course: str):
    """Fetch images list for a course"""
    try:
        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
        url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
        async with aiohttp.ClientSession() as session:
            async with session.get(url) as resp:
                data = await resp.json()
                if isinstance(data, dict) and 'images' in data:
                    return data['images']
                return []
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error fetching images: {e}")

@app.get("/servers/status")
async def get_servers_status():
    """Get status of all caption servers"""
    server_statuses = []
    for server in servers:
        server_statuses.append(ServerStatus(
            url=server.url,
            model=server.model,
            busy=server.busy,
            total_processed=server.total_processed,
            total_time=server.total_time,
            fps=server.fps
        ))
    return server_statuses

@app.get("/processing/status")
async def get_processing_status():
    """Get current processing status"""
    status_info = {}
    for course in processed_images:
        total = len(processed_images[course])
        processed = len(course_captions.get(course, []))
        failed = len(failed_images.get(course, set()))
        status_info[course] = {
            "course": course,
            "total_images": total,
            "processed_images": processed,
            "failed_images": failed,
            "progress_percent": (processed / total * 100) if total > 0 else 0,
            "status": "completed" if processed + failed >= total else "processing"
        }
    return status_info

@app.post("/processing/start")
async def start_processing(request: StartProcessingRequest = StartProcessingRequest()):
    """Start caption processing"""
    global is_processing, current_processing_task
    
    if is_processing:
        raise HTTPException(status_code=400, detail="Processing is already running")
    
    is_processing = True
    current_processing_task = asyncio.create_task(
        processing_loop(request.courses, request.continuous)
    )
    
    return {
        "message": "Processing started", 
        "continuous": request.continuous,
        "specific_courses": request.courses
    }

@app.post("/processing/stop")
async def stop_processing():
    """Stop caption processing"""
    global is_processing, current_processing_task
    
    if not is_processing:
        raise HTTPException(status_code=400, detail="Processing is not running")
    
    is_processing = False
    if current_processing_task:
        current_processing_task.cancel()
        try:
            await current_processing_task
        except asyncio.CancelledError:
            pass
        current_processing_task = None
    
    return {"message": "Processing stopped"}

@app.get("/captions/{course}")
async def get_captions(course: str):
    """Get captions for a specific course"""
    captions = load_captions_from_file(course)
    return {
        "course": course,
        "total_captions": len(captions),
        "captions": captions
    }

@app.delete("/captions/{course}")
async def delete_captions(course: str):
    """Delete captions for a specific course"""
    try:
        file_path = get_caption_file_path(course)
        if file_path.exists():
            file_path.unlink()
            if course in processed_images:
                del processed_images[course]
            if course in course_captions:
                del course_captions[course]
            if course in failed_images:
                del failed_images[course]
            return {"message": f"Captions for {course} deleted"}
        else:
            raise HTTPException(status_code=404, detail=f"No captions found for {course}")
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error deleting captions: {e}")

# Core processing functions
async def fetch_courses() -> List[str]:
    """Fetch available courses from source server"""
    async with aiohttp.ClientSession() as session:
        async with session.get(f"{SOURCE_SERVER}/courses") as resp:
            data = await resp.json()
            if isinstance(data, dict) and 'courses' in data:
                return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
            return []

async def fetch_course_images(course: str) -> List[Dict]:
    """Fetch images list for a course"""
    course_frames = f"{course}_frames" if not course.endswith("_frames") else course
    url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
    async with aiohttp.ClientSession() as session:
        async with session.get(url) as resp:
            data = await resp.json()
            if isinstance(data, dict) and 'images' in data:
                return data['images']
            return []

async def get_caption(server: str, image_url: str) -> Dict:
    """Get caption from a specific server"""
    params = {
        'image_url': image_url,
        'model_choice': MODEL_TYPE
    }
    try:
        async with aiohttp.ClientSession() as session:
            async with session.get(server, params=params, timeout=30) as resp:
                return await resp.json()
    except Exception as e:
        print(f"Error from {server}: {e}")
        return None

async def get_model_info():
    """Get model information from caption servers"""
    model_info = []
    async with aiohttp.ClientSession() as session:
        for server in CAPTION_SERVERS:
            try:
                health_url = server.rsplit('/analyze', 1)[0] + '/health'
                async with session.get(health_url) as resp:
                    info = await resp.json()
                    model_info.append({
                        'url': server,
                        'model': info.get('model_choice', 'unknown')
                    })
            except Exception as e:
                print(f"Couldn't get model info from {server}: {e}")
    return model_info


async def wait_for_vs2_ready(course: str, timeout: Optional[int] = None, interval: int = 5):
    """Poll the SOURCE_SERVER /vs2/state endpoint until VS2 reports 'ready' for the given course.
    If timeout is None, this will poll indefinitely until VS2 is ready or idle.
    """
    url = f"{SOURCE_SERVER}/vs2/state"
    elapsed = 0
    async with aiohttp.ClientSession() as session:
        while True:
            try:
                async with session.get(url, timeout=10) as resp:
                    if resp.status == 200:
                        data = await resp.json()
                        # data may be either {'state': ..., 'current_course': ...} or {'states': {...}}
                        state = data.get('state') or None
                        current = data.get('current_course') or data.get('current_file')
                        if state is None and 'states' in data:
                            # per-course states dict was returned
                            states = data['states']
                            state = states.get(course)
                            current = course

                        print(f"VS2 state: {state}, current: {current}")
                        # If VS2 explicitly ready for this course, proceed
                        if state == 'ready':
                            return True
                        # If VS2 idle for this course (or unknown), proceed
                        if state in (None, 'idle'):
                            return True
                    else:
                        print(f"VS2 state endpoint returned {resp.status}")
            except Exception as e:
                print(f"Could not query VS2 state: {e}")

            # if timeout set and exceeded, raise; otherwise continue indefinitely
            if timeout is not None:
                elapsed += interval
                if elapsed >= timeout:
                    raise Exception(f"Timeout waiting for VS2 to be ready for course {course}")

            await asyncio.sleep(interval)

async def process_image(server: CaptionServer, course: str, image: Dict) -> Dict:
    """Process single image through one caption server with better error handling"""
    if server.busy:
        return None
        
    server.busy = True
    start_time = time.time()
    
    try:
        # Structure URL correctly: /images/COURSE_NAME_frames/IMAGE.png
        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
        image_url = urljoin(SOURCE_SERVER, f"/images/{quote(course_frames)}/{quote(image['filename'])}")
        result = await get_caption(server.url, image_url)
        
        processing_time = time.time() - start_time
        server.total_time += processing_time
        
        if result and result.get('success') and result.get('caption'):
            server.total_processed += 1
            metadata = {
                "image": image['filename'],
                "caption": result['caption'],
                "server": server.url,
                "processing_time": processing_time,
                "timestamp": datetime.now().isoformat()
            }
            print(f"Server {server.url} processed {image['filename']} in {processing_time:.2f}s ({server.fps:.2f} fps)")
            return metadata
        else:
            # Server responded but no caption (might be error or empty response)
            error_msg = result.get('error', 'Unknown error') if result else 'No response'
            print(f"Server {server.url} failed for {image['filename']}: {error_msg}")
            return None
            
    except asyncio.TimeoutError:
        print(f"Server {server.url} timeout for {image['filename']}")
        return None
    except Exception as e:
        print(f"Error processing {image['filename']} on {server.url}: {e}")
        return None
        
    finally:
        server.busy = False

async def upload_to_huggingface(course: str, metadata_list: List[Dict]):
    """Upload course captions to Hugging Face dataset"""
    try:
        print(f"📤 Uploading {len(metadata_list)} captions for {course} to Hugging Face...")
        
        # Prepare data for Hugging Face dataset
        dataset_data = {
            "course": [],
            "image_filename": [],
            "caption": [],
            "processing_server": [],
            "processing_time": [],
            "timestamp": []
        }
        
        for metadata in metadata_list:
            dataset_data["course"].append(course)
            dataset_data["image_filename"].append(metadata["image"])
            dataset_data["caption"].append(metadata["caption"])
            dataset_data["processing_server"].append(metadata["server"])
            dataset_data["processing_time"].append(metadata["processing_time"])
            dataset_data["timestamp"].append(metadata["timestamp"])
        
        # Create dataset
        dataset = Dataset.from_dict(dataset_data)
        
        # Login to Hugging Face
        huggingface_hub.login(token=HF_TOKEN)
        
        # Push to hub
        dataset.push_to_hub(
            HF_DATASET_ID,
            config_name=course.replace("/", "_").replace(" ", "_"),
            split="train",  # You can change this to "train", "validation", "test" as needed
            commit_message=f"Add captions for course {course} - {len(metadata_list)} images"
        )
        
        print(f"✅ Successfully uploaded {len(metadata_list)} captions for {course} to {HF_DATASET_ID}")
        # Notify VS2 (if VS2 provided a callback for this course)
        try:
            await notify_vs2_flow_done(course, success=True)
        except Exception as e:
            print(f"Warning: failed to notify VS2 about completion for {course}: {e}")
        return True
        
    except Exception as e:
        print(f"❌ Error uploading to Hugging Face: {e}")
        return False


async def notify_vs2_flow_done(course: str, success: bool):
    """If VS2 provided a callback URL for this course, POST a completion signal."""
    callback = pending_vs2_callbacks.get(course)
    if not callback:
        # try fallback: look for any callback registered under partial names
        for key, cb in pending_vs2_callbacks.items():
            if key in course:
                callback = cb
                break
    if not callback:
        # nothing to do
        return

    payload = {
        "course": course,
        "status": "done" if success else "failed",
        "timestamp": datetime.now().isoformat()
    }

    print(f"Notifying VS2 at {callback} about course {course} -> {payload['status']}")
    try:
        async with aiohttp.ClientSession() as session:
            async with session.post(callback, json=payload, timeout=30) as resp:
                if resp.status >= 400:
                    text = await resp.text()
                    print(f"VS2 callback returned {resp.status}: {text}")
    except Exception as e:
        print(f"Error notifying VS2 callback {callback}: {e}")

async def process_course(course: str, servers: List[CaptionServer]):
    """Process all images in a course using available servers with proper retry logic"""
    # Initialize course tracking
    if course not in processed_images:
        processed_images[course] = set()
    if course not in course_captions:
        course_captions[course] = load_captions_from_file(course)
        # Update processed images set from loaded captions
        for cap in course_captions[course]:
            processed_images[course].add(cap['image'])
    if course not in failed_images:
        failed_images[course] = set()
    
    # Get list of images
    images = await fetch_course_images(course)
    if not images:
        print(f"No images found for course {course}")
        return
    
    print(f"\nProcessing {len(images)} images for course {course}")
    
    # Track images that need processing with retry count (5 retries)
    pending_images = {}
    for img in images:
        filename = img['filename']
        if filename not in processed_images[course] and filename not in failed_images[course]:
            pending_images[filename] = {'image': img, 'retries': 0, 'max_retries': 5}
    
    if not pending_images:
        print(f"All images already processed or failed for course {course}")
        print(f"- Processed: {len(processed_images[course])}, Failed: {len(failed_images[course])}")
        
        # If course is completed, upload to Hugging Face
        if len(processed_images[course]) + len(failed_images[course]) >= len(images):
            if course_captions[course]:
                print(f"📤 Course {course} completed, uploading to Hugging Face...")
                await upload_to_huggingface(course, course_captions[course])
        return
    
    print(f"Images to process: {len(pending_images)} (already processed: {len(processed_images[course])}, failed: {len(failed_images[course])})")
    
    batch_size = len([s for s in servers if not s.busy])
    processed_in_this_run = 0
    
    while pending_images and is_processing:
        # Create tasks for each available server
        tasks = []
        assigned_images = []
        
        for server in servers:
            if not server.busy and pending_images:
                # Get the next pending image
                filename, img_data = next(iter(pending_images.items()))
                img = img_data['image']
                
                # Assign this image to the server
                tasks.append(process_image(server, course, img))
                assigned_images.append((filename, img, img_data['retries']))
                # Remove from pending temporarily while it's being processed
                del pending_images[filename]
                
        if not tasks:
            # If no servers available, wait a bit
            await asyncio.sleep(0.1)
            continue
            
        # Process images in parallel across servers
        results = await asyncio.gather(*tasks)
        
        # Handle results and retry logic
        has_new_results = False
        for (filename, img, current_retries), result in zip(assigned_images, results):
            if result:
                # Success - image was processed
                processed_images[course].add(filename)
                course_captions[course].append(result)
                has_new_results = True
                processed_in_this_run += 1
                print(f"✓ Successfully processed {filename}")
            else:
                # Failure - check if we should retry
                if current_retries < 5:  # max_retries
                    # Put back in pending for retry with incremented retry count
                    pending_images[filename] = {
                        'image': img, 
                        'retries': current_retries + 1, 
                        'max_retries': 5
                    }
                    print(f"↻ Retry {current_retries + 1}/5 for {filename}")
                else:
                    # Max retries exceeded, mark as failed
                    failed_images[course].add(filename)
                    print(f"✗ Failed to process {filename} after 5 retries")
        
        # Save progress after each batch with new results
        if has_new_results:
            save_captions_to_file(course, course_captions[course])
        
        # Show progress
        total = len(images)
        done = len(processed_images[course])
        failed_count = len(failed_images[course])
        pending_count = len(pending_images)
        progress_percent = (done / total * 100) if total > 0 else 0
        
        print(f"\rProgress: {done}/{total} ({progress_percent:.1f}%) - {pending_count} pending, {failed_count} failed, {processed_in_this_run} new", end="", flush=True)
        
        # Small delay to prevent overwhelming the servers
        await asyncio.sleep(0.5)
    
    # Final status for this course
    total = len(images)
    done = len(processed_images[course])
    failed_count = len(failed_images[course])
    
    if done + failed_count >= total:
        if failed_count > 0:
            print(f"\n✓ Course {course} completed with {failed_count} failed images")
        else:
            print(f"\n✓ Course {course} fully completed")
        
        # Upload to Hugging Face when course is completed
        if course_captions[course]:
            print(f"📤 Uploading {len(course_captions[course])} captions to Hugging Face...")
            success = await upload_to_huggingface(course, course_captions[course])
            if success:
                print(f"✅ Successfully uploaded {course} to Hugging Face")
            else:
                print(f"❌ Failed to upload {course} to Hugging Face")
    else:
        print(f"\n→ Course {course} partially completed: {done}/{total} processed, {failed_count} failed")

async def processing_loop(specific_courses: Optional[List[str]] = None, continuous: bool = True):
    """Main processing loop with proper error handling"""
    global is_processing
    
    # Get model information and verify Florence-2-large availability
    model_info = await get_model_info()
    print("\nCaption Servers:")
    available_servers = []
    for info, server in zip(model_info, servers):
        server.model = info['model']
        if MODEL_TYPE in info.get('model', ''):
            available_servers.append(server)
            print(f"✓ {server.url} confirmed {MODEL_TYPE}")
        else:
            print(f"✗ {server.url} using {server.model} - skipping (requires {MODEL_TYPE})")
    
    if not available_servers:
        print(f"\nError: No servers with {MODEL_TYPE} available!")
        is_processing = False
        return
    
    # Update servers list to only use those with large model
    processing_servers = available_servers
    print(f"\nUsing {len(processing_servers)} servers with {MODEL_TYPE}")
    
    # Check for existing caption files and report
    existing_captions = list(CAPTIONS_DIR.glob("*_captions.json"))
    if existing_captions:
        print("\nFound existing caption files:")
        for cap_file in existing_captions:
            course = cap_file.stem.replace("_captions", "")
            try:
                with open(cap_file, 'r', encoding='utf-8') as f:
                    captions = json.load(f)
                    print(f"- {course}: {len(captions)} captions")
            except Exception as e:
                print(f"- Error reading {cap_file.name}: {e}")
        print()
    
    start_time = time.time()
    iteration = 0
    
    while is_processing:
        try:
            iteration += 1
            print(f"\n{'='*50}")
            print(f"Processing Iteration {iteration}")
            print(f"{'='*50}")
            
            # Get available courses
            if specific_courses:
                courses = specific_courses
                print(f"Processing specific courses: {courses}")
            else:
                courses = await fetch_courses()
                print(f"Found {len(courses)} courses")
                
            if not courses:
                print("No courses found, waiting...")
                if not continuous:
                    break
                await asyncio.sleep(10)
                continue
            
            # Process each course with all available servers
            for course in courses:
                if not is_processing:
                    break

                print(f"\n--- Processing course: {course} ---")
                # Before processing, ensure VS2 has finished extracting frames for this course
                try:
                    await wait_for_vs2_ready(course)
                except Exception as e:
                    print(f"Warning: error while checking VS2 readiness for {course}: {e}")

                await process_course(course, processing_servers)
                
            # Show server stats
            print("\nServer Stats:")
            total_processed = sum(s.total_processed for s in processing_servers)
            elapsed = time.time() - start_time
            if elapsed > 0:
                print(f"Total images processed: {total_processed}")
                print(f"Overall speed: {total_processed/elapsed:.2f} fps")
                for s in processing_servers:
                    print(f"- {s.url}: {s.total_processed} images, {s.fps:.2f} fps")
            print()
            
            if not continuous:
                print("One-time processing completed")
                break
                
            # Wait before next check
            print("Waiting for new courses...")
            await asyncio.sleep(5)
            
        except asyncio.CancelledError:
            print("Processing cancelled")
            break
        except Exception as e:
            print(f"Error in processing loop: {str(e)}")
            import traceback
            traceback.print_exc()
            await asyncio.sleep(10)
    
    is_processing = False
    print("Processing loop stopped")

# Startup event
@app.on_event("startup")
async def startup_event():
    """Initialize servers and start processing on startup"""
    initialize_servers()
    print("Caption Coordinator API started")
    print(f"Source server: {SOURCE_SERVER}")
    print(f"Caption servers: {len(CAPTION_SERVERS)}")
    print(f"Hugging Face dataset: {HF_DATASET_ID}")
    print(f"HF Token: {'✅ Set' if HF_TOKEN else '❌ Missing'}")
    
    # Start processing automatically (like original main())
    if auto_start_processing:
        print("Auto-starting processing loop...")
        global is_processing, current_processing_task
        is_processing = True
        current_processing_task = asyncio.create_task(processing_loop())


@app.post("/vs2/ready")
async def vs2_ready(course: str, callback_url: str = None):
    """Called by VS2 when it has finished extracting frames for a course.
    VS2 should POST course (string) and its callback_url (where Flow will POST when captioning is done).
    """
    if not course:
        raise HTTPException(status_code=400, detail="course is required")

    if callback_url:
        pending_vs2_callbacks[course] = callback_url
        print(f"Registered VS2 callback for {course} -> {callback_url}")

    # Acknowledge. The processing loop will discover the new course via SOURCE_SERVER /courses.
    return {"status": "accepted", "course": course, "callback_url": callback_url}


@app.get("/vs2/callbacks")
async def list_vs2_callbacks():
    """List pending VS2 callbacks (debug)"""
    return pending_vs2_callbacks


if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000, reload=True)