[Question]: The agent page did not return, and the log shows “socket. send() raised exception”

[xianxh]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

My agent is shown in the picture:

After running the agent to retrieve the knowledge base, the page does not display any returns for a long time, and clicking on the page may cause lag. The backend log shows that the response took about 3 to 5 minutes, and finally an error occurred:
{
"role": "user",
"content": "Respond with a formal answer. FORGET(DO NOT mention) about complete_task. The language for the response MUST be as the same as the first user request.\n"
}
]
16:43:06 - LiteLLM:INFO: utils.py:3296 -
LiteLLM completion() model= qwen-plus; provider = dashscope
2025-11-24 16:43:06,175 INFO 18
LiteLLM completion() model= qwen-plus; provider = dashscope
[2025-11-24 16:43:09 +0800] [18] [INFO] 127.0.0.1:50812 POST /v1/canvas/completion 1.1 - - 82295830
2025-11-24 16:43:10,464 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:10,733 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:10,884 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:11,048 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:11,166 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:11,250 WARNING 18 socket.send() raised exception.
2025-11-24 16:43:11,407 WARNING 18 socket.send() raised exception.

[dosubot]

The "socket.send() raised exception" error after long delays is usually caused by backend processing taking too long—often due to slow retrieval steps, complex agent workflows, or resource bottlenecks. When the backend takes several minutes to respond, the client (browser) may disconnect, and the server logs this socket error. This is a common symptom when the retrieval pipeline is slow or blocked, especially with the qwen-plus model and dashscope provider. Similar issues have been reported by other users, and fixes have been made in recent nightly versions of RAGFlow, which resolve bugs in retrieval and agent orchestration logic [source].

To troubleshoot: profile your agent pipeline to find slow steps, check for oversized or malformed inputs, review your workflow configuration (especially output-to-input mapping between retrieval and agent nodes), and monitor backend logs for stack overflow or database errors. Reducing agent max rounds and concurrency, tuning Elasticsearch/vector DB settings, and disabling expensive retrieval features can help. Upgrading to the latest nightly version is strongly recommended, as it has resolved similar issues for others [source].

If the UI is stuck, you can fetch logs and results directly from backend API endpoints, and clean up stuck tasks in MySQL/Redis if needed. If the problem persists, try isolating it by testing with different models/providers or smaller inputs, and collect detailed logs for further diagnosis.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Magicbook1108]

@xianxh Could you try reproducing this issue using the chat interface? Alternatively, could you share the flow with us via xiaoyang.bao@infiniflow.org?

If possible, it would be even more helpful if you could also share some of the documents from the dataset.