Update app.py

app.py

@@ -33,11 +33,12 @@ os.makedirs("results", exist_ok=True)
 
 templates = Jinja2Templates(directory="templates")
 
-# --- الگوریتم برش (بدون تغییر) ---
+# --- الگوریتم برش ---
 def find_split_points(audio_path, sr=24000):
     y, _ = librosa.load(audio_path, sr=sr)
     total_samples = len(y)
     
+    # تنظیمات برش برای پایداری بیشتر
     MIN_PREFERRED = 8.0
     MAX_PREFERRED = 12.0
     MIN_HARD = 6.0
@@ -91,28 +92,46 @@ def find_split_points(audio_path, sr=24000):
         
     return split_points, y
 
-async def process_chunk_on_worker(worker_url, chunk_path, ref_path, index, job_id):
-    async with aiohttp.ClientSession() as session:
+async def process_chunk_with_retry(worker_url, chunk_path, ref_path, index, job_id, max_retries=3):
+    """ تابع ارسال به کارگر با قابلیت تلاش مجدد """
+    
+    # افزایش تایم‌اوت به ۶۰۰ ثانیه (۱۰ دقیقه) برای هر تکه
+    timeout = aiohttp.ClientTimeout(total=600, connect=60, sock_read=600)
+    
+    for attempt in range(max_retries):
         try:
-            with open(chunk_path, 'rb') as f_chunk, open(ref_path, 'rb') as f_ref:
-                data = aiohttp.FormData()
-                data.add_field('source_file', f_chunk, filename='chunk.wav', content_type='audio/wav')
-                data.add_field('ref_file', f_ref, filename='ref.wav', content_type='audio/wav')
-                
-                async with session.post(worker_url, data=data, timeout=300) as response:
-                    if response.status == 200:
-                        result_data = await response.read()
-                        out_path = f"temp/{job_id}_part_{index}.wav"
-                        with open(out_path, 'wb') as f_out:
-                            f_out.write(result_data)
-                        return index, out_path
-                    else:
-                        print(f"Worker {worker_url} failed with {response.status}")
-                        return index, None
+            async with aiohttp.ClientSession(timeout=timeout) as session:
+                with open(chunk_path, 'rb') as f_chunk, open(ref_path, 'rb') as f_ref:
+                    data = aiohttp.FormData()
+                    # هدرهای صحیح برای جلوگیری از رد شدن درخواست
+                    data.add_field('source_file', f_chunk, filename='chunk.wav', content_type='audio/wav')
+                    data.add_field('ref_file', f_ref, filename='ref.wav', content_type='audio/wav')
+                    
+                    async with session.post(worker_url, data=data) as response:
+                        if response.status == 200:
+                            result_data = await response.read()
+                            if len(result_data) < 1000: # چک کردن اینکه فایل خیلی کوچک (احتمالا ارور متنی) نباشد
+                                print(f"Warning: Chunk {index} returned suspicious small data from {worker_url}")
+                                
+                            out_path = f"temp/{job_id}_part_{index}.wav"
+                            with open(out_path, 'wb') as f_out:
+                                f_out.write(result_data)
+                            return index, out_path
+                        else:
+                            # اگر سرور ارور داد (مثلا 500 یا 503)
+                            print(f"Attempt {attempt+1}/{max_retries} failed for chunk {index} on {worker_url}: Status {response.status}")
+                            
         except Exception as e:
-            print(f"Error calling worker {worker_url}: {e}")
-            return index, None
+            print(f"Attempt {attempt+1}/{max_retries} connection error for chunk {index} on {worker_url}: {e}")
+        
+        # اگر خطا خورد، کمی صبر کن و دوباره تلاش کن (Backoff)
+        if attempt < max_retries - 1:
+            await asyncio.sleep(2 * (attempt + 1))
+            
+    print(f"All {max_retries} attempts failed for chunk {index}.")
+    return index, None
 
+# حافظه موقت جاب‌ها
 JOBS = {}
 
 async def job_manager_task(job_id, source_path, ref_path):
@@ -120,12 +139,12 @@ async def job_manager_task(job_id, source_path, ref_path):
         JOBS[job_id]["status"] = "processing"
         sr = 24000
         
-        # تبدیل و تمیزکاری رفرنس
+        # آماده‌سازی رفرنس
         ref_y, _ = librosa.load(ref_path, sr=sr)
         clean_ref_path = f"temp/{job_id}_ref_clean.wav"
         sf.write(clean_ref_path, ref_y, sr)
         
-        # برش
+        # برش فایل اصلی
         split_points, y = find_split_points(source_path, sr)
         total_chunks = len(split_points) - 1
         JOBS[job_id]["total_chunks"] = total_chunks
@@ -146,9 +165,10 @@ async def job_manager_task(job_id, source_path, ref_path):
             chunk_files.append((i, chunk_path, read_start, read_end, start, end))
             
             worker_url = WORKER_URLS[i % len(WORKER_URLS)]
-            tasks.append(process_chunk_on_worker(worker_url, chunk_path, clean_ref_path, i, job_id))
+            # استفاده از تابع جدید با قابلیت Retry
+            tasks.append(process_chunk_with_retry(worker_url, chunk_path, clean_ref_path, i, job_id))
         
-        # اجرا
+        # اجرای تسک‌ها
         results = []
         for f in asyncio.as_completed(tasks):
             idx, path = await f
@@ -156,9 +176,9 @@ async def job_manager_task(job_id, source_path, ref_path):
                 JOBS[job_id]["completed_chunks"] += 1
                 results.append((idx, path))
             else:
-                print(f"Chunk {idx} failed completely.")
+                print(f"Chunk {idx} failed completely after retries.")
         
-        # چسباندن
+        # چسباندن (Stitching)
         results.sort(key=lambda x: x[0])
         processed_map = {idx: path for idx, path in results}
         final_audio = []
@@ -187,10 +207,12 @@ async def job_manager_task(job_id, source_path, ref_path):
                             final_audio[-1][-fade_len:] = mixed
                             valid_part = valid_part[fade_len:]
                     final_audio.append(valid_part)
-                except:
+                except Exception as e:
+                    print(f"Stitch error {i}: {e}")
                     real_len = split_points[i+1] - split_points[i]
                     final_audio.append(np.zeros(real_len))
             else:
+                # اگر تکه‌ای کلاً خراب شد، سکوت می‌گذاریم که تایم کلی به هم نریزد
                 real_len = split_points[i+1] - split_points[i]
                 final_audio.append(np.zeros(real_len))
 
@@ -206,7 +228,7 @@ async def job_manager_task(job_id, source_path, ref_path):
         JOBS[job_id]["status"] = "completed"
         JOBS[job_id]["filename"] = out_filename
         
-        # Cleanup
+        # پاکسازی
         if os.path.exists(clean_ref_path): os.remove(clean_ref_path)
         for _, c_path, _, _, _, _ in chunk_files:
             if os.path.exists(c_path): os.remove(c_path)
@@ -214,7 +236,7 @@ async def job_manager_task(job_id, source_path, ref_path):
             if os.path.exists(path): os.remove(path)
             
     except Exception as e:
-        print(f"Job failed: {e}")
+        print(f"Job failed fatal: {e}")
         JOBS[job_id]["status"] = "failed"
     finally:
         if os.path.exists(source_path): os.remove(source_path)
@@ -233,6 +255,7 @@ async def create_job(
     job_id = str(uuid.uuid4())
     source_path = f"temp/{job_id}_src.wav"
     ref_path = f"temp/{job_id}_ref.wav"
+    
     with open(source_path, "wb") as b: shutil.copyfileobj(source_audio.file, b)
     with open(ref_path, "wb") as b: shutil.copyfileobj(ref_audio.file, b)
     
@@ -242,7 +265,10 @@ async def create_job(
         "completed_chunks": 0,
         "filename": None
     }
+    
+    # اجرای تسک در پس‌زمینه (حتی اگر کاربر قطع شود ادامه می‌یابد)
     background_tasks.add_task(job_manager_task, job_id, source_path, ref_path)
+    
     return {"job_id": job_id}
 
 @app.get("/status/{job_id}")
@@ -255,10 +281,10 @@ def get_status(job_id: str):
 def download_file(filename: str):
     path = f"results/{filename}"
     if os.path.exists(path):
-        # اضافه کردن هدر برای اطمینان از پخش صحیح در مرورگر
         return FileResponse(path, filename=filename, media_type="audio/wav")
     return {"error": "File not found"}
 
 if __name__ == "__main__":
     import uvicorn
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    # افزایش تایم‌اوت خود uvicorn برای اطمینان
+    uvicorn.run(app, host="0.0.0.0", port=7860, timeout_keep_alive=300)