dosubot[bot]
commented
2 months ago Hey there, @Nitewriter! I'm here to help you with any bugs, questions, or contributions you might have. Let's tackle this issue together!
To resolve the error "The message must be an iterator or an async iterator" when using the ConditionalRouter component in your flow via the API, you need to ensure that the INPUT_FIELD_NAME parameter is set to an iterator or an async iterator. This is necessary for the flow to handle the message correctly.
Here's how you can address this issue:
-
Ensure
INPUT_FIELD_NAMEis an Iterator or Async Iterator: Make sure that theINPUT_FIELD_NAMEparameter in your flow is set to an iterator or an async iterator. Here is an example of how to set this parameter correctly:from typing import AsyncIterator async def example_async_iterator() -> AsyncIterator[str]: for i in range(5): yield f"Message {i}" # Assuming `vertex` is an instance of the class containing the `stream` method vertex.params[INPUT_FIELD_NAME] = example_async_iterator() await vertex.stream() -
Check the Implementation of the
streamMethod: Ensure that thestreammethod in your vertex class is correctly implemented to handle the iterator or async iterator. Here is a relevant code snippet that checks this condition:async def stream(self): iterator = self.params.get(INPUT_FIELD_NAME, None) if not isinstance(iterator, (AsyncIterator, Iterator)): raise ValueError("The message must be an iterator or an async iterator.") ... -
Example of Handling Streaming in Langflow: In the context of Langflow, the
build_vertex_streamfunction insrc/backend/base/langflow/api/v1/chat.pydemonstrates how to handle streaming for a vertex. Here is an excerpt:@router.get("/build/{flow_id}/{vertex_id}/stream", response_class=StreamingResponse) async def build_vertex_stream( flow_id: uuid.UUID, vertex_id: str, session_id: Optional[str] = None, chat_service: "ChatService" = Depends(get_chat_service), session_service: "SessionService" = Depends(get_session_service), ): try: flow_id_str = str(flow_id) async def stream_vertex(): try: cache = await chat_service.get_cache(flow_id_str) if not cache: raise ValueError(f"No cache found for {flow_id_str}.") else: graph = cache.get("result") vertex: "InterfaceVertex" = graph.get_vertex(vertex_id) if not hasattr(vertex, "stream"): raise ValueError(f"Vertex {vertex_id} does not support streaming") if isinstance(vertex._built_result, str) and vertex._built_result: stream_data = StreamData( event="message", data={"message": f"Streaming vertex {vertex_id}"}, ) yield str(stream_data) stream_data = StreamData( event="message", data={"chunk": vertex._built_result}, ) yield str(stream_data) elif not vertex.frozen or not vertex._built: logger.debug(f"Streaming vertex {vertex_id}") stream_data = StreamData( event="message", data={"message": f"Streaming vertex {vertex_id}"}, ) yield str(stream_data) async for chunk in vertex.stream(): stream_data = StreamData( event="message", data={"chunk": chunk}, ) yield str(stream_data) elif vertex.result is not None: stream_data = StreamData( event="message", data={"chunk": vertex._built_result}, ) yield str(stream_data) else: raise ValueError(f"No result found for vertex {vertex_id}") except Exception as exc: logger.exception(f"Error building Component: {exc}") exc_message = parse_exception(exc) if exc_message == "The message must be an iterator or an async iterator.": exc_message = "This stream has already been closed." yield str(StreamData(event="error", data={"error": exc_message})) finally: logger.debug("Closing stream") await chat_service.set_cache(flow_id_str, graph) yield str(StreamData(event="close", data={"message": "Stream closed"})) return StreamingResponse(stream_vertex(), media_type="text/event-stream") except Exception as exc: raise HTTPException(status_code=500, detail="Error building Component") from exc
By ensuring that the INPUT_FIELD_NAME parameter is correctly set to an iterator or async iterator and that the stream method is properly implemented, you should be able to resolve the error and successfully run the flow via the API [1][2][3][4].
To continue talking to Dosu, mention @dosu.
nicoloboschi
Nitewriter
carlosrcoelho
I'm attempting to implement a
supervisorpattern where a language model classifies the current message given the conversation so far. I can get this to work perfectly within the UI by using a conditional router component, however, when I attempt to use the flow from the API it throws an error.I suspect this is due to the need for two separate chat outputs in the flow to facilitate the branch in conversation. Here is an example flow that demonstrates the issue. I'm currently using the latest changes from
mainto test.dont-talk-about-bruno.json