feat: --experimental-performance flag for faster performance

solara/server/flask.py

@@ -57,6 +57,7 @@ class WebsocketWrapper(websocket.WebsocketWrapper):
     def __init__(self, ws: simple_websocket.Server) -> None:
         self.ws = ws
         self.lock = threading.Lock()
+        super().__init__()
 
     def close(self):
         with self.lock:
@@ -78,6 +79,9 @@ async def receive(self):
         except simple_websocket.ws.ConnectionClosed:
             raise websocket.WebSocketDisconnect()
 
+    def flush(self):
+        pass  # we do not implement queueing messages in flask (yet)
+
 
 class ServerFlask(ServerBase):
     server: Any

solara/server/server.py

@@ -152,11 +152,14 @@ async def app_loop(ws: websocket.WebsocketWrapper, session_id: str, kernel_id: s
                 else:
                     msg = deserialize_binary_message(message)
                 t1 = time.time()
-                if not process_kernel_messages(kernel, msg):
-                    # if we shut down the kernel, we do not keep the page session alive
-                    context.close()
-                    return
+                hold_messages = ws.hold_messages() if settings.main.experimental_performance else solara.util.nullcontext()
+                with hold_messages:
+                    if not process_kernel_messages(kernel, msg):
+                        # if we shut down the kernel, we do not keep the page session alive
+                        context.close()
+                        return
                 t2 = time.time()
+
                 if settings.main.timing:
                     widgets_ids_after = set(patch.widgets)
                     created_widgets_count = len(widgets_ids_after - widgets_ids)

solara/server/settings.py

@@ -151,6 +151,7 @@ class MainSettings(BaseSettings):
     base_url: str = ""  # e.g. https://myapp.solara.run/myapp/
     platform: str = sys.platform
     host: str = HOST_DEFAULT
+    experimental_performance: bool = True
 
     class Config:
         env_prefix = "solara_"

solara/server/starlette.py

@@ -72,17 +72,37 @@ class WebsocketWrapper(websocket.WebsocketWrapper):
     ws: starlette.websockets.WebSocket
 
     def __init__(self, ws: starlette.websockets.WebSocket, portal: anyio.from_thread.BlockingPortal) -> None:
+        super().__init__()
         self.ws = ws
         self.portal = portal
+        self.to_send: List[Union[str, bytes]] = []
+
+    def flush(self):
+        async def _flush():
+            to_send, self.to_send = self.to_send, []
+            for data in to_send:
+                if isinstance(data, bytes):
+                    await self.ws.send_bytes(data)
+                else:
+                    await self.ws.send_text(data)
+
+        if settings.main.experimental_performance:
+            self.portal.call(_flush)
 
     def close(self):
         self.portal.call(self.ws.close)
 
     def send_text(self, data: str) -> None:
-        self.portal.call(self.ws.send_text, data)
+        if self._queuing_messages:
+            self.to_send.append(data)
+        else:
+            self.portal.call(self.ws.send_bytes, data)
 
     def send_bytes(self, data: bytes) -> None:
-        self.portal.call(self.ws.send_bytes, data)
+        if self._queuing_messages:
+            self.to_send.append(data)
+        else:
+            self.portal.call(self.ws.send_bytes, data)
 
     async def receive(self):
         if hasattr(self.portal, "start_task_soon"):

solara/server/websocket.py

@@ -3,6 +3,7 @@
 Async implementation have to come up with a way how to do this sync (see e.g. the starlette implementation)
 """
 import abc
+import contextlib
 import json
 from typing import Union
 
@@ -12,6 +13,9 @@ class WebSocketDisconnect(Exception):
 
 
 class WebsocketWrapper(abc.ABC):
+    def __init__(self):
+        self._queuing_messages = False
+
     @abc.abstractmethod
     def send_text(self, data: str) -> None:
         pass
@@ -42,3 +46,17 @@ async def receive(self) -> Union[str, bytes]:
     async def receive_json(self):
         text = await self.receive()
         return json.loads(text)
+
+    @abc.abstractmethod
+    def flush(self):
+        pass
+
+    @contextlib.contextmanager
+    def hold_messages(self):
+        # we're assuming this only get used from a single thread (only at server.py app loop)
+        self._queuing_messages = True
+        try:
+            yield
+        finally:
+            self._queuing_messages = False
+            self.flush()