How do you use Sentry?
Sentry Saas (sentry.io)
Version
2.43.0
Steps to Reproduce
I'm using openai lib with gemini. For couple of my tasks gemini's response was like this:
{
"id": "vOALaYOpG4CJ7M8PlpOM0QY",
"choices": [
{
"finish_reason": "content_filter: PROHIBITED_CONTENT",
"index": 0,
"logprobs": null,
"message": null
}
],
"created": 1762386108,
"model": "gemini-2.5-flash-lite",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": null,
"usage": {
"completion_tokens": 0,
"prompt_tokens": 5431,
"total_tokens": 5431,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}This made openai integration choke like following:
File /usr/local/lib/python3.12/site-packages/sentry_sdk/integrations/openai.py:246, in _set_output_data(span, response, kwargs, integration, finish_span)
244 if hasattr(response, "choices"):
245 if should_send_default_pii() and integration.include_prompts:
--> 246 response_text = [choice.message.model_dump() for choice in response.choices]
247 if len(response_text) > 0:
248 set_data_normalized(span, SPANDATA.GEN_AI_RESPONSE_TEXT, response_text)
AttributeError: 'NoneType' object has no attribute 'model_dump'
Which is understandable, as integration clearly expects message to be present.
I'm using latest sdk and openai client (openai==2.7.1)
Expected Result
It should ignore such messages or report them separately
Actual Result
It dies with a traceback even before I can read the message from openai lib.
How do you use Sentry?
Sentry Saas (sentry.io)
Version
2.43.0
Steps to Reproduce
I'm using openai lib with gemini. For couple of my tasks gemini's response was like this:
{ "id": "vOALaYOpG4CJ7M8PlpOM0QY", "choices": [ { "finish_reason": "content_filter: PROHIBITED_CONTENT", "index": 0, "logprobs": null, "message": null } ], "created": 1762386108, "model": "gemini-2.5-flash-lite", "object": "chat.completion", "service_tier": null, "system_fingerprint": null, "usage": { "completion_tokens": 0, "prompt_tokens": 5431, "total_tokens": 5431, "completion_tokens_details": null, "prompt_tokens_details": null } }This made openai integration choke like following:
Which is understandable, as integration clearly expects message to be present.
I'm using latest sdk and openai client (openai==2.7.1)
Expected Result
It should ignore such messages or report them separately
Actual Result
It dies with a traceback even before I can read the message from openai lib.