Summary

• Root cause: Test's _FakeCompletionGeneratorArgs lacks reasoning_parser and the fake server lacks _input_proc_executor, both of which the real openai_completion path reads (openai_server.py:1923, 1955); the resulting AttributeError is caught and returned as a JSONResponse (no body_iterator).
• Fix: Refresh the fakes only — add reasoning_parser = None to _FakeCompletionGeneratorArgs (safe: add_thinking_budget_logits_processor short-circuits when thinking_token_budget is None), and set server._input_proc_executor = None (safe: loop.run_in_executor(None, ...) uses the default executor).
• Automated fix generated by repair-bot

Test plan

• Verify fix on the same GPU type as the original failure
• Check for regressions in related tests

Links

• Bug: https://nvbugs/6507081

Dev Engineer Review

• Updated tests/unittest/llmapi/test_llm.py OpenAI prompt-streaming reuse test scaffolding to better match the real server interface by adding missing fake attributes:
  • Set _FakeCompletionGeneratorArgs.reasoning_parser = None.
  • Set server._input_proc_executor = None.
• These changes are minimal and intended to prevent AttributeError failures from test doubles that didn’t expose fields expected by the openai_completion path.
• Updated tests/integration/test_lists/waives.txt to remove the waiver entry for unittest/llmapi/test_llm.py::test_openai_completion_list_prompt_stream_reuses_stream_metadata, re-enabling the test in integration runs.

QA Engineer Review

• Modified test scaffolding in tests/unittest/llmapi/test_llm.py:
  • test_openai_completion_list_prompt_stream_reuses_stream_metadata: adjusted mocked server/fake objects to include reasoning_parser and _input_proc_executor attributes required by the code under test.
  • No test functions were added or removed; behavior changes are limited to test double compatibility.
• Integration test-list coverage:
  • The affected test is now un-avoided/unwaived via tests/integration/test_lists/waives.txt.
• Modified test-list file:
  • tests/integration/test_lists/waives.txt: removed waiver for unittest/llmapi/test_llm.py::test_openai_completion_list_prompt_stream_reuses_stream_metadata.
• Verdict: sufficient.