[Bug]: gemini-2.5-flash ERROR: argument of type 'NoneType' is not iterable

[taoari]

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.

RAGFlow workspace code commit ID

v0.21.1

RAGFlow image version

v0.21.1

Other environment information

Actual behavior

Gemini model does not work with ERROR: argument of type 'NoneType' is not iterable

ragflow-server | Traceback (most recent call last):
ragflow-server | File "/ragflow/api/apps/conversation_app.py", line 223, in stream
ragflow-server | for ans in chat(dia, msg, True, **req):
ragflow-server | File "/ragflow/api/db/services/dialog_service.py", line 598, in chat
ragflow-server | for ans in chat_mdl.chat_streamly(prompt + prompt4citation, msg[1:], gen_conf):
ragflow-server | File "/ragflow/api/db/services/llm_service.py", line 255, in chat_streamly
ragflow-server | for txt in chat_partial(**use_kwargs):
ragflow-server | File "/ragflow/rag/llm/cv_model.py", line 673, in chat_streamly
ragflow-server | yield total_token_count_from_response(response)
ragflow-server | File "/ragflow/rag/utils/init.py", line 61, in total_token_count_from_response
ragflow-server | if 'usage' in resp and 'total_tokens' in resp['usage']:

Expected behavior

No response

Steps to reproduce

Add Gemini models with API key, and try in chat

Additional information

No response

[dosubot]

The error happens because total_token_count_from_response expects the response object to have a 'usage' field, but in some Gemini model code paths, the response can be None or even a string, not a dict or object with 'usage'. This leads to the 'NoneType' is not iterable error when the function tries to check for keys in a None value. The function also doesn't check if the response is None before using it, which is the immediate cause of the crash.

In GeminiCV.chat_streamly(), the response variable can be None or a generator that's already been exhausted, and in GeminiCV.chat(), a string is sometimes passed instead of the expected response object. This inconsistent handling is the root cause of the issue. The code for GoogleChat.chat_streamly() uses a more robust approach by accumulating token counts during streaming, rather than relying on extracting token usage from the final response object, which may not be available or valid after streaming.

A defensive fix would be to add a check at the start of total_token_count_from_response to return 0 if the response is None or not a dict/object, and to ensure that the Gemini integration always passes the correct type to this function. For reference, see the implementation details and discussion here:
total_token_count_from_response implementation
GeminiCV usage and error context

To reply, just mention @dosu.

---

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

[zhangzhefang-github]

I'd like to take this issue and try to fix/test it.
A defensive check in total_token_count_from_response should prevent the NoneType error.
Will submit a PR shortly.