fix(openai): Attach response model with streamed Completions API#5557
fix(openai): Attach response model with streamed Completions API#5557alexander-alderman-webb wants to merge 2 commits intowebb/openai/add-response-modelfrom
Conversation
Semver Impact of This PR🟢 Patch (bug fixes) 📋 Changelog PreviewThis is how your changes will appear in the changelog. Bug Fixes 🐛Openai
Other
Documentation 📚
Internal Changes 🔧Agents
Openai
Openai Agents
Other
🤖 This preview updates automatically when you update the PR. |
Codecov Results 📊✅ 6 passed | Total: 6 | Pass Rate: 100% | Execution Time: 1.65s 📊 Comparison with Base Branch
All tests are passing successfully. ❌ Patch coverage is 0.00%. Project has 15378 uncovered lines. Files with missing lines (181)
Coverage diff@@ Coverage Diff @@
## main #PR +/-##
==========================================
+ Coverage 22.48% 22.66% +0.18%
==========================================
Files 189 189 —
Lines 19830 19884 +54
Branches 6428 6462 +34
==========================================
+ Hits 4457 4506 +49
- Misses 15373 15378 +5
- Partials 374 378 +4Generated by Codecov Action |
…sponse-model-completions
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| nonlocal ttft | ||
| count_tokens_manually = True | ||
| async for x in old_iterator: | ||
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) |
There was a problem hiding this comment.
Accessing x.model without defensive check could raise AttributeError and break streaming
The new code accesses x.model directly without checking if the attribute exists, and this call is placed outside the capture_internal_exceptions() block. If a streaming chunk doesn't have a model attribute (which could happen with certain API responses or edge cases), an AttributeError will propagate to the user's application and break their streaming logic. The rest of the code in this function uses defensive checks like hasattr(x, "choices") before accessing attributes.
Verification
Read the full function context at lines 658-686. Verified that the synchronous version at line 616 has the same pattern (also potentially problematic). Confirmed that capture_internal_exceptions() is used elsewhere in the file to protect against attribute access errors. Checked that set_data is a simple dictionary assignment that won't throw, so the issue is specifically with x.model access.
Suggested fix: Wrap the model attribute access inside the existing capture_internal_exceptions block or add a hasattr check
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) | |
| if hasattr(x, "model"): | |
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) | |
Also found at 1 additional location
sentry_sdk/integrations/openai.py:616-616
Identified by Warden code-review · CWR-SCT
There was a problem hiding this comment.
Fix attempt detected (commit f12cf5d)
The commit added unprotected x.model access outside the capture_internal_exceptions() block at both lines 616 and 662, directly introducing the exact issue reported: an AttributeError could propagate if a streaming chunk lacks the model attribute.
The original issue appears unresolved. Please review and try again.
Evaluated by Warden
There was a problem hiding this comment.
Spec says it's required: https://developers.openai.com/api/reference/resources/chat/subresources/completions/streaming-events
| nonlocal ttft | ||
| count_tokens_manually = True | ||
| for x in old_iterator: | ||
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) |
There was a problem hiding this comment.
Unprotected attribute access on streaming chunk may cause runtime error
The x.model access on line 616 is placed outside the capture_internal_exceptions() block, unlike other model accesses in this file (e.g., lines 761, 810 which are inside the block, or line 472 which uses hasattr guard). If a streaming chunk lacks the model attribute or it's malformed, an AttributeError will propagate to the user's application instead of being silently logged by Sentry's internal error handler.
Verification
Read openai.py lines 590-670 and 740-815. Verified pattern: line 472 uses if hasattr(response, 'model') guard, lines 761 and 810 place model access inside capture_internal_exceptions(). The new code at line 616 has neither protection and could raise AttributeError to user code.
Suggested fix: Move the set_data call inside the existing capture_internal_exceptions() block
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) | |
| if hasattr(x, "model"): | |
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) | |
Identified by Warden code-review · 26J-GUA
There was a problem hiding this comment.
Spec says it's required: developers.openai.com/api/reference/resources/chat/subresources/completions/streaming-events
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| nonlocal ttft | ||
| count_tokens_manually = True | ||
| for x in old_iterator: | ||
| span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, x.model) |
There was a problem hiding this comment.
Ungarded streamed model access can raise
Medium Severity
x.model is read directly in the streaming iterators before capture_internal_exceptions(). If a streamed chunk does not expose model (or uses a dict-like/event variant), this raises and interrupts iteration, making instrumentation break the caller’s stream instead of failing safely.
Additional Locations (1)
There was a problem hiding this comment.
Spec says it's required: developers.openai.com/api/reference/resources/chat/subresources/completions/streaming-events
2 participants
Description
Issues
Reminders
tox -e linters.feat:,fix:,ref:,meta:)