fix(vps): report stable creation progress

websocket_service.py

@@ -48,6 +48,17 @@ import websockets
 
 __all__ = ['DateTimeEncoder', 'attach_to_game_console', 'ensure_node_shell_pod', 'execute_minecraft_save_all', 'handle_2fa_disable', 'handle_2fa_enable', 'handle_2fa_login', 'handle_2fa_setup', 'handle_add_credits', 'handle_authenticate', 'handle_change_server_version', 'handle_check_demo_eligibility', 'handle_convert_demo_to_paid', 'handle_copy_path', 'handle_create_backup', 'handle_create_custom_domain', 'handle_create_directory', 'handle_create_domain', 'handle_create_server', 'handle_create_ticket', 'handle_delete_backup', 'handle_delete_custom_domain', 'handle_delete_domain', 'handle_delete_path', 'handle_delete_server', 'handle_download_file', 'handle_exec_command', 'handle_expose_port', 'handle_game_console_command', 'handle_game_console_stream', 'handle_generate_sftp', 'handle_get_backups', 'handle_get_credits', 'handle_get_minecraft_settings', 'handle_get_server_creation_progress', 'handle_get_server_logs', 'handle_get_server_metrics', 'handle_get_servers', 'handle_get_ticket', 'handle_get_ticket_stats', 'handle_get_tickets', 'handle_link_domain', 'handle_list_custom_domains', 'handle_list_domains', 'handle_list_files', 'handle_login', 'handle_move_path', 'handle_node_terminal_connect', 'handle_node_terminal_disconnect', 'handle_node_terminal_input', 'handle_pack_zip', 'handle_read_file', 'handle_register', 'handle_remove_port', 'handle_restart_server', 'handle_restore_backup', 'handle_send_ticket_message', 'handle_start_chunked_upload', 'handle_start_server', 'handle_stop_server', 'handle_sync_server_status', 'handle_unlink_domain', 'handle_unpack_zip', 'handle_update_minecraft_settings', 'handle_update_ticket_status', 'handle_upload_file', 'handle_upload_file_chunk', 'handle_verify_custom_domain', 'handle_vps_terminal_connect', 'handle_vps_terminal_disconnect', 'handle_vps_terminal_input', 'handle_websocket', 'handle_write_file', 'notify_admins', 'notify_ticket_participants', 'parse_ls_output', 'send_message', 'send_verification_email', 'start_websocket_server', 'stream_game_console_output', 'stream_metrics', 'sync_user_domains_with_dns']
 
+_server_creation_progress_by_user: Dict[str, Dict] = {}
+_server_creation_progress_lock = threading.Lock()
+
+
+def _set_server_creation_progress(user_id: str, progress: Dict) -> Dict:
+    """Keep the latest creation status available after a WebSocket reconnect."""
+    snapshot = {**progress, 'timestamp': time.time()}
+    with _server_creation_progress_lock:
+        _server_creation_progress_by_user[str(user_id)] = snapshot
+    return snapshot
+
 
 
 async def handle_create_server(client_id, data):
@@ -58,6 +69,14 @@ async def handle_create_server(client_id, data):
         return
 
     user_id = client_info['user_id']
+    server_type = data.get('serverType', '')
+    _set_server_creation_progress(user_id, {
+        'phase': 'Preparing',
+        'progress': 1,
+        'message': 'Checking capacity and preparing your server.',
+        'serverType': server_type,
+        'state': 'active'
+    })
 
     try:
         # Check if this is a demo server request
@@ -67,6 +86,18 @@ async def handle_create_server(client_id, data):
         server_name = data.get('name', '').strip()
         is_valid, error_msg = validate_server_name(server_name)
         if not is_valid:
+            failed = _set_server_creation_progress(user_id, {
+                'phase': 'Failed',
+                'progress': 1,
+                'message': error_msg,
+                'serverType': server_type,
+                'state': 'failed'
+            })
+            await send_message(client_id, {'type': 'server_creation_progress', **failed})
+            await send_message(client_id, {
+                'type': 'server_creation_failed',
+                'error': error_msg
+            })
             await send_message(client_id, {
                 'type': 'error',
                 'error': error_msg
@@ -80,7 +111,15 @@ async def handle_create_server(client_id, data):
         capacity_check = validate_server_capacity(data)
 
         if not capacity_check['success']:
+            failed = _set_server_creation_progress(user_id, {
+                'phase': 'Capacity check',
+                'progress': 1,
+                'message': capacity_check['message'],
+                'serverType': server_type,
+                'state': 'failed'
+            })
             # Send detailed capacity error
+            await send_message(client_id, {'type': 'server_creation_progress', **failed})
             await send_message(client_id, {
                 'type': 'capacity_error',
                 'error': capacity_check['message'],
@@ -90,43 +129,48 @@ async def handle_create_server(client_id, data):
             })
             return
 
-        # Capacity check passed, proceed with server creation
-        # Create progress callback that sends websocket messages
+        # Capacity check passed, proceed with server creation.
         loop = asyncio.get_event_loop()
 
         def progress_callback(phase, progress, message):
-            """Thread-safe progress callback that sends websocket messages"""
+            """Store and forward progress updates from the provisioning thread."""
             try:
-                # Store progress in client info for reconnection recovery
-                if client_id in connected_clients:
-                    connected_clients[client_id]['server_creation_progress'] = {
-                        'phase': phase,
-                        'progress': progress,
-                        'message': message,
-                        'timestamp': time.time()
-                    }
+                with _server_creation_progress_lock:
+                    previous = dict(_server_creation_progress_by_user.get(str(user_id), {}))
+                state = 'failed' if phase in ('Failed', 'Error', 'Timeout') else 'active'
+                snapshot = _set_server_creation_progress(user_id, {
+                    'phase': phase,
+                    'progress': max(int(previous.get('progress', 0)), max(0, min(99, int(progress)))),
+                    'message': message,
+                    'serverType': server_type,
+                    'state': state
+                })
 
-                # Use run_coroutine_threadsafe to send async message from executor thread
                 future = asyncio.run_coroutine_threadsafe(
                     send_message(client_id, {
                         'type': 'server_creation_progress',
-                        'phase': phase,
-                        'progress': progress,
-                        'message': message
+                        **snapshot
                     }),
                     loop
                 )
-                # Don't wait for completion to avoid blocking
-                # The future will complete asynchronously
+                future.result(timeout=2)
             except Exception as e:
                 logger.error(f"Error sending progress update: {e}")
 
-        # Run blocking create_server in executor with progress callback
+        # Run blocking provisioning outside the event loop.
         server = await loop.run_in_executor(None, create_server, data, user_id, progress_callback)
 
-        # Clear server creation progress on completion
-        if client_id in connected_clients:
-            connected_clients[client_id].pop('server_creation_progress', None)
+        completed = _set_server_creation_progress(user_id, {
+            'phase': 'Provisioned',
+            'progress': 100,
+            'message': 'Server resources are ready. The VPS may take a little longer to boot.',
+            'serverType': server_type,
+            'state': 'complete'
+        })
+        await send_message(client_id, {
+            'type': 'server_creation_progress',
+            **completed
+        })
 
         # Get user credits
         user = get_user_by_id(user_id)
@@ -145,10 +189,22 @@ async def handle_create_server(client_id, data):
         })
     except Exception as e:
         logger.error(f"Error creating server: {e}", exc_info=True)
-
-        # Clear server creation progress on error
-        if client_id in connected_clients:
-            connected_clients[client_id].pop('server_creation_progress', None)
+        previous = _server_creation_progress_by_user.get(str(user_id), {})
+        failed = _set_server_creation_progress(user_id, {
+            'phase': 'Failed',
+            'progress': previous.get('progress', 0),
+            'message': str(e),
+            'serverType': data.get('serverType', ''),
+            'state': 'failed'
+        })
+        await send_message(client_id, {
+            'type': 'server_creation_progress',
+            **failed
+        })
+        await send_message(client_id, {
+            'type': 'server_creation_failed',
+            'error': str(e)
+        })
 
         # Check if it's a capacity-related error
         error_msg = str(e)
@@ -1269,25 +1325,25 @@ async def handle_get_server_creation_progress(client_id, data):
         await send_message(client_id, {'type': 'error', 'error': 'Not authenticated'})
         return
 
-    # Get stored progress if available
-    progress = client_info.get('server_creation_progress')
+    user_id = str(client_info['user_id'])
+    with _server_creation_progress_lock:
+        progress = _server_creation_progress_by_user.get(user_id)
     if progress:
-        # Check if progress is recent (within last 5 minutes)
-        if time.time() - progress.get('timestamp', 0) < 300:
+        expiry = 1800 if progress.get('state') == 'active' else 600
+        if time.time() - progress.get('timestamp', 0) < expiry:
             await send_message(client_id, {
                 'type': 'server_creation_progress',
-                'phase': progress.get('phase', ''),
-                'progress': progress.get('progress', 0),
-                'message': progress.get('message', 'Creating server...')
+                **progress
             })
         else:
-            # Progress is stale, clear it
-            client_info.pop('server_creation_progress', None)
+            with _server_creation_progress_lock:
+                _server_creation_progress_by_user.pop(user_id, None)
             await send_message(client_id, {
                 'type': 'server_creation_progress',
                 'phase': 'Unknown',
                 'progress': 0,
-                'message': 'No active server creation found'
+                'message': 'No recent server creation status is available.',
+                'state': 'unknown'
             })
     else:
         # No progress stored
@@ -1295,7 +1351,8 @@ async def handle_get_server_creation_progress(client_id, data):
             'type': 'server_creation_progress',
             'phase': 'Unknown',
             'progress': 0,
-            'message': 'No active server creation found'
+            'message': 'No recent server creation status is available.',
+            'state': 'unknown'
         })
 
 async def handle_create_domain(client_id, data):