OpenAI Responses API Instrumentation

[TimPansino]

Overview

• Instrument the newer openai.responses APIs.
• Add a new set of tests that mirrors the original chat_completions tests.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Max errors	Warnings	Elapsed time
✅ ACTION	actionlint	9		0		0	1.55s
✅ MARKDOWN	markdownlint	7	0	0		0	1.32s
✅ PYTHON	ruff	1108	0	0		0	1.49s
✅ PYTHON	ruff-format	1108	0	0		0	0.34s
✅ YAML	prettier	23	0	0		0	1.55s
✅ YAML	v8r	23		0		0	7.37s
✅ YAML	yamllint	23		0		0	1.34s

See detailed reports in MegaLinter artifacts

Show us your support by starring ⭐ the repository

[codecov-commenter]

Codecov Report

❌ Patch coverage is 87.93103% with 14 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.09%. Comparing base (44c3db2) to head (fa0c000).

Files with missing lines	Patch %	Lines
newrelic/hooks/mlmodel_openai.py	87.82%	2 Missing and 12 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1813      +/-   ##
==========================================
- Coverage   82.13%   82.09%   -0.04%     
==========================================
  Files         220      220              
  Lines       27344    27447     +103     
  Branches     4330     4353      +23     
==========================================
+ Hits        22458    22532      +74     
- Misses       3381     3391      +10     
- Partials     1505     1524      +19     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[umaannamalai]

For chat completions, we deliberately suppress content-less tool call messages (on line 595 by checking "tool_call" in finish_reason). Since we don't want to report every tool call as an LlmChatCompletionMessage with content: "", do we want to mirror that suppression behavior here if possible?

[umaannamalai]

Or is that why "role" is being checked for explicitly?

[umaannamalai]

Are we capturing errors surfaced via response.failed somewhere? From my scan of the codebase, it seems like APIErrors are raised when the raw payload has a top-level error key. response.failed events nest errors inside response.error and the stream just ends normally afterwards. I just want to make sure we are capturing errors in this case also.