Fix: inject routes via background thread after launch

app.py

@@ -458,27 +458,37 @@ def _benchmark_test_handler():
 
 
 # ---------------------------------------------------------------------------
-# Monkey-patch: intercept Gradio server creation to inject routes.
+# After launch, inject routes into the running server.
 # ---------------------------------------------------------------------------
 
-_orig_routes = gr.routes.App.__init__
-
-
-def _patched_init(self, *args, **kwargs):
-    _orig_routes(self, *args, **kwargs)
-    _log("Gradio server __init__ patched - adding routes")
-    try:
-        self.add_api_route("/respite/resources", _resources_handler, methods=["GET"])
-        self.add_api_route("/respite/specs", _specs_handler, methods=["GET"])
-        self.add_api_route("/respite/search", _search_handler, methods=["GET"])
-        self.add_api_route("/respite/benchmark", _benchmark_handler, methods=["GET"])
-        self.add_api_route("/respite/benchmark/test", _benchmark_test_handler, methods=["GET"])
-        _log("Routes added via __init__ patch")
-    except Exception as e:
-        _log(f"Route patching failed: {e}")
+import atexit
+
+def _inject_routes_after_launch():
+    """Wait for Gradio server to start, then add our routes."""
+    for attempt in range(30):
+        time.sleep(2)
+        server = getattr(demo, 'server', None)
+        if server is None:
+            continue
+        fa_app = getattr(server, 'app', None)
+        if fa_app is None:
+            continue
+        _log(f"Injecting routes into running server (attempt {attempt+1})...")
+        try:
+            fa_app.add_api_route("/respite/resources", _resources_handler, methods=["GET"])
+            fa_app.add_api_route("/respite/specs", _specs_handler, methods=["GET"])
+            fa_app.add_api_route("/respite/search", _search_handler, methods=["GET"])
+            fa_app.add_api_route("/respite/benchmark", _benchmark_handler, methods=["GET"])
+            fa_app.add_api_route("/respite/benchmark/test", _benchmark_test_handler, methods=["GET"])
+            _log(f"Routes injected into running server on attempt {attempt+1}")
+            return
+        except Exception as e:
+            _log(f"Route injection attempt {attempt+1} failed: {e}")
+    _log("Failed to inject routes after 30 attempts")
 
 
-gr.routes.App.__init__ = _patched_init
+_thread = threading.Thread(target=_inject_routes_after_launch, daemon=True)
+_thread.start()
 
 # Disable the Gradio Spaces file watcher thread which crashes because
 # importing `spaces` in a new thread after CUDA init fails.