Update app.py

app.py

@@ -17,6 +17,8 @@ logger = logging.getLogger(__name__)
 PREDICT_TIMEOUT = 600
 GPU_WARM_WINDOW = 900
 MAX_RETRIES = 1
+MAX_QUEUE_SIZE = 50  # Maximum number of requests in queue
+DEFAULT_CONCURRENCY_LIMIT = 3  # Number of concurrent backend calls
 
 # ─────────  Backend connection ─────────
 HF_TOKEN = os.getenv("HF_TOKEN")
@@ -76,10 +78,19 @@ def base64_to_image(b64: str) -> Image.Image | None:
         logger.error(f"Failed to decode base64 → image: {e}")
         return None
 
+# ─────────  Queue Status ─────────
+def get_queue_status():
+    """Get current queue status for display"""
+    try:
+        # This would need to be implemented based on your queue system
+        # For now, return a placeholder
+        return "📊 Queue: Ready for requests"
+    except Exception as e:
+        return f"📊 Queue status unavailable: {e}"
+
 # ─────────  UI ↔ Backend bridge ─────────
 def call_backend_with_retry(input_image: Image.Image, category: str, gender: str, *, max_retries: int = MAX_RETRIES):
-    
-    """Single‑shot call (no more than `max_retries` times)."""
+    """Single‑shot call (no more than `max_retries` times) with queue handling."""
 
     if input_image is None:
         return None, None, "❌ Please upload an image.", gr.update(interactive=True)
@@ -90,21 +101,24 @@ def call_backend_with_retry(input_image: Image.Image, category: str, gender: str
             return None, None, msg, gr.update(interactive=True)
 
     client: Client = backend_status["client"]
-    # print(Client.view_api())
     img_b64 = image_to_base64(input_image)
 
+    # Add queue position info
+    start_time = time.time()
+    logger.info(f"Request queued at {time.strftime('%H:%M:%S')}")
+
     for attempt in range(max_retries):
         try:
             logger.info(f"Backend call #{attempt+1}")
-            start = time.time()
+            
+            # Add timeout to prevent hanging
             result = client.predict(
                 img_b64,
                 category,
                 gender,
                 api_name="/predict",
-                
             )
-            dt = time.time() - start
+            dt = time.time() - start_time
 
             if not result or len(result) < 4:
                 raise ValueError("Invalid response structure from backend")
@@ -118,7 +132,9 @@ def call_backend_with_retry(input_image: Image.Image, category: str, gender: str
 
             if not status.startswith("✅"):
                 status = "✅ " + status
-            status += f" (⏱ {dt:.1f}s)"
+            status += f" (⏱ Total: {dt:.1f}s)"
+            
+            logger.info(f"Request completed successfully in {dt:.1f}s")
             return overlay_img, bg_img, status, gr.update(interactive=True)
 
         except Exception as e:
@@ -132,6 +148,10 @@ def call_backend_with_retry(input_image: Image.Image, category: str, gender: str
 def disable_button():
     return gr.update(interactive=False)
 
+def show_queue_position():
+    """Show current position in queue"""
+    return "⏳ Request added to queue..."
+
 # ─────────  CSS ─────────
 custom_css = """
 .gradio-container {
@@ -147,6 +167,24 @@ custom_css = """
     box-shadow: 0 10px 30px rgba(0, 0, 0, 0.2);
     backdrop-filter: blur(10px);
 }
+.queue-status {
+    background: linear-gradient(135deg, #4CAF50, #45a049);
+    color: white;
+    padding: 10px;
+    border-radius: 8px;
+    text-align: center;
+    margin: 10px 0;
+    font-weight: 500;
+}
+.queue-warning {
+    background: linear-gradient(135deg, #ff9800, #f57c00);
+    color: white;
+    padding: 10px;
+    border-radius: 8px;
+    text-align: center;
+    margin: 10px 0;
+    font-weight: 500;
+}
 .title-container {
     text-align: center;
     margin-bottom: 25px;
@@ -284,14 +322,24 @@ with gr.Blocks(css=custom_css, title="Jewellery Photography Preview") as demo:
 
     # Status banner
     status_html = gr.HTML()
+    
+    # Queue status display
+    queue_status = gr.HTML()
 
     def _update_status():
         ok, msg = check_backend_connection()
         cls = "status-ready" if ok else ("status-starting" if "🟡" in msg else "status-error")
         return f'<div class="status-banner {cls}">{msg}</div>'
 
+    def _update_queue_status():
+        return f'<div class="queue-status">{get_queue_status()}</div>'
+
     status_html.value = _update_status()
-    gr.Button("🔄 Check Status").click(fn=_update_status, outputs=status_html)
+    queue_status.value = _update_queue_status()
+    
+    with gr.Row():
+        gr.Button("🔄 Check Status").click(fn=_update_status, outputs=status_html)
+        gr.Button("📊 Queue Status").click(fn=_update_queue_status, outputs=queue_status)
 
     with gr.Column():
             with gr.Row():
@@ -334,6 +382,15 @@ with gr.Blocks(css=custom_css, title="Jewellery Photography Preview") as demo:
             
             
             out_status = gr.Text(label="Status", interactive=False)
+            
+            # Queue information
+            gr.HTML("""
+                <div class="tip-box">
+                    <strong>⏳ Queue Information:</strong> Requests are processed in order. 
+                    During high traffic, you may experience wait times. The button will be disabled while processing.
+                </div>
+            """)
+            
         # ──────── Footer ────────
     gr.HTML("""
         <div style="text-align:center;padding:40px 20px;background:#f8fafc;border:1px solid #e2e8f0;border-radius:16px;margin:30px 0;">
@@ -352,20 +409,32 @@ with gr.Blocks(css=custom_css, title="Jewellery Photography Preview") as demo:
         </div>
     """)
 
-    # wire button → backend
+    # Wire button → backend with queue status updates
     run_btn.click(
         fn=disable_button,
         inputs=None,
         outputs=run_btn
+    ).then(
+        fn=show_queue_position,
+        inputs=None,
+        outputs=out_status
     ).then(
         fn=call_backend_with_retry,
         inputs=[input_img, category, gender],
         outputs=[out_overlay, out_bg, out_status, run_btn],
-        concurrency_limit=1,
+        concurrency_limit=DEFAULT_CONCURRENCY_LIMIT,  # Limit concurrent backend calls
         show_progress=True,
     )
     
 
-# ─────────  Launch ─────────
+# ─────────  Launch with Queue ─────────
 if __name__ == "__main__":
-    demo.queue(max_size=20, default_concurrency_limit=1).launch(share=False)
+    demo.queue(
+        max_size=MAX_QUEUE_SIZE,  # Maximum requests in queue
+        default_concurrency_limit=DEFAULT_CONCURRENCY_LIMIT,  # Concurrent processing limit
+    ).launch(
+        share=False,
+        server_name="0.0.0.0",
+        server_port=7860,
+        show_error=True
+    )