agentscope-ai / agentscope-ai/agentscope

[Bug]: WebUI message stalls indefinitely (spinner never stops) when LLM backend crashes/freezes

Ouverte
#2,110 1 commentaire 0 réactions 0 personnes assignées Voir sur GitHub
Langage dominant
Python
Étoiles
31.5k
Forks
3.5k
Merge moyen
1 j 23 h
PR mergées (30 j)
95

Description

### Prerequisites

- [x] I have searched the existing [issues](https://github.com/agentscope-ai/agentscope/issues) and [discussions](https://github.com/agentscope-ai/agentscope/discussions), and this is not a duplicate.
- [x] This is a bug, not a usage question. (For questions, please use [Discussions](https://github.com/agentscope-ai/agentscope/discussions/new?category=general) instead.)

### Background / Description

When sending a message in the WebUI, if the underlying LLM model/service crashes, hangs, or becomes unresponsive (e.g., OOM, process killed, network partition), the frontend UI enters an unrecoverable state. The loading spinner continues to animate indefinitely—sometimes for hours—and the "Stop" button becomes non-functional. The user is forced to manually refresh the browser to recover the session.
Expected behavior: When the LLM backend fails, the WebUI should detect the connection drop or timeout, stop the spinner, and display an error message (e.g., "Model response failed" or "Connection lost"). The "Stop" button should either work to cancel the pending request or be replaced by a "Retry/Refresh" option.

### Error Messages

```shell
Backend logs typically show the LLM crash (e.g., CUDA out of memory, grpc timeout, connection reset by peer). However, the browser console shows no explicit error; the WebSocket/SSE connection often remains technically "open" on the client side even though the backend stopped streaming. No ReplyEndEventor error event is received by the frontend to trigger the UI reset.
```

### Steps to Reproduce

1. Start the AgentScope service and open the WebUI.

2. Configure a chat session with an LLM model (e.g., a local vLLM instance or a remote endpoint).
3. Simulate a model crash or freeze

4. Observe the WebUI:
The message bubble shows a loading spinner.
The spinner continues spinning indefinitely (tested up to 2+ hours).

### Environment

- AgentScope Version: 2.0.4
- Python Version: 3.12
- OS: windows 11

Guide de contribution

Ouvrir le guide de contribution

Évaluation

Cette issue n'a pas encore été évaluée.

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.