remote-gpu-server/app.py

"""
Remote GPU Execution Server v2.2 (Streaming Support)
=====================================================
Questo script usa FastAPI + Gradio per ricevere codice Python ed eseguirlo sulla GPU remota.
Supporta sia esecuzione sincrona che streaming in tempo reale.
"""

from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
import gradio as gr
import subprocess
import tempfile
import base64
import glob
import shutil
import os
import sys

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)


def execute_code(python_code: str) -> dict:
    """
    Esegue codice e restituisce output + file generati (png, jpg, csv).
    """
    if not python_code or not python_code.strip():
        return {"output": "❌ Errore: Nessun codice ricevuto.", "files": []}
    
    # Crea directory temporanea unica per questa esecuzione
    work_dir = tempfile.mkdtemp()
    script_path = os.path.join(work_dir, "script.py")
    
    # Scrive il codice
    with open(script_path, 'w', encoding='utf-8') as f:
        f.write(python_code)
    
    output_parts = []
    generated_files = []
    
    try:
        output_parts.append("🚀 Avvio esecuzione...")
        output_parts.append(f"📂 Directory: {work_dir}")
        output_parts.append("=" * 50)
        
        # Esegue
        result = subprocess.run(
            [sys.executable, script_path],
            capture_output=True,
            text=True,
            cwd=work_dir, # Esegue nella dir temporanea
            timeout=3600,  # 1 hour timeout for long training runs
            env={**os.environ, "PYTHONUNBUFFERED": "1"}
        )
        
        # Output
        if result.stdout: output_parts.append(result.stdout)
        if result.stderr: 
            output_parts.append("\n--- STDERR ---")
            output_parts.append(result.stderr)
            
        output_parts.append("=" * 50)
        status = "✅ Successo" if result.returncode == 0 else f"❌ Errore ({result.returncode})"
        output_parts.append(f"{status}")

        # Cerca file generati (immagini, csv, json)
        for ext in ['*.png', '*.jpg', '*.svg', '*.csv', '*.json', '*.txt', '*.pth', '*.pt']:
            if ext == '*.txt': continue # Salta txt per evitare di rileggere log
            for filepath in glob.glob(os.path.join(work_dir, ext)):
                filename = os.path.basename(filepath)
                try:
                    with open(filepath, "rb") as f:
                        if os.path.getsize(filepath) > 10 * 1024 * 1024:
                            output_parts.append(f"⚠️ File '{filename}' troppo grande (>10MB), saltato.")
                            continue
                        encoded = base64.b64encode(f.read()).decode('utf-8')
                        generated_files.append({
                            "name": filename,
                            "data": encoded,
                            "mime": "application/octet-stream"
                        })
                    output_parts.append(f"📦 File generato: {filename}")
                except Exception as e:
                    output_parts.append(f"❌ Errore lettura {filename}: {e}")
        
    except Exception as e:
        import traceback
        output_parts.append(f"❌ Eccezione: {str(e)}\n{traceback.format_exc()}")
    
    finally:
        shutil.rmtree(work_dir, ignore_errors=True)
    
    return {
        "output": "\n".join(output_parts),
        "files": generated_files
    }


def get_system_info():
    info = ["System Info:"]
    info.append(f"Python: {sys.version}")
    try:
        import torch
        info.append(f"PyTorch: {torch.__version__}")
        info.append(f"CUDA Available: {torch.cuda.is_available()}")
        if torch.cuda.is_available():
            info.append(f"GPU: {torch.cuda.get_device_name(0)}")
    except ImportError:
        info.append("PyTorch not installed")
    return "\n".join(info)


@app.get("/")
def read_root():
    return {"status": "running", "message": "Remote GPU Execution Server"}


@app.post("/execute")
async def execute(data: dict):
    """Endpoint POST per eseguire codice Python"""
    try:
        code = data.get("code", "")
        return execute_code(code)
    except Exception as e:
        import traceback
        return {"output": f"❌ Server Error: {str(e)}\n\n{traceback.format_exc()}", "files": []}


def stream_generator(python_code: str):
    """Generator che esegue codice e yielda output in tempo reale"""
    if not python_code or not python_code.strip():
        yield "❌ Errore: Nessun codice ricevuto.\n"
        return
    
    work_dir = tempfile.mkdtemp()
    script_path = os.path.join(work_dir, "script.py")
    
    with open(script_path, 'w', encoding='utf-8') as f:
        f.write(python_code)
    
    yield f"🚀 Avvio esecuzione...\n"
    yield f"📂 Directory: {work_dir}\n"
    yield "=" * 50 + "\n"
    
    try:
        process = subprocess.Popen(
            [sys.executable, "-u", script_path],  # -u for unbuffered
            stdout=subprocess.PIPE,
            stderr=subprocess.STDOUT,
            text=True,
            cwd=work_dir,
            env={**os.environ, "PYTHONUNBUFFERED": "1"}
        )
        
        # Stream output line by line
        for line in process.stdout:
            yield line
        
        process.wait()
        
        yield "\n" + "=" * 50 + "\n"
        status = "✅ Successo" if process.returncode == 0 else f"❌ Errore ({process.returncode})"
        yield f"{status}\n"
        
    except Exception as e:
        import traceback
        yield f"❌ Eccezione: {str(e)}\n{traceback.format_exc()}\n"
    finally:
        shutil.rmtree(work_dir, ignore_errors=True)


@app.post("/execute_stream")
async def execute_stream(data: dict):
    """Endpoint POST per eseguire codice Python con streaming in tempo reale"""
    code = data.get("code", "")
    return StreamingResponse(stream_generator(code), media_type="text/plain")


# Interfaccia Gradio (Wrapper semplificato)
def gradio_wrapper(code):
    res = execute_code(code)
    return res["output"]

demo = gr.Interface(
    fn=gradio_wrapper,
    inputs=gr.Textbox(
        label="📝 Codice Python",
        placeholder="Incolla qui il tuo codice Python...",
        lines=15
    ),
    outputs=gr.Textbox(label="📤 Output Esecuzione (i file non sono visualizzati qui, usa l'API)"),
    title="🚀 Remote GPU Execution Server",
    description=f"Server attivo. {get_system_info()}"
)

# Monta Gradio su FastAPI
app = gr.mount_gradio_app(app, demo, path="/gradio")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)