[litellm-agent] Staging → litellm_internal_staging (5/6/2026)

[oss-pr-review-agent-shin]

Automated staging PR created by litellm-agent.

This branch collects PRs approved by the agent on 5/6/2026.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
4 out of 9 committers have signed the CLA.

✅ KunalG67
✅ pnookala-godaddy
✅ michelligabriele
✅ Sameerlite
❌ Aneesh-Fiddler
❌ Michael Riad Zaky
❌ cursoragent
❌ oss-pr-review-agent-shin[bot]
❌ oss-agent-shin

---

Michael Riad Zaky seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This automated staging PR bundles five distinct improvements: deadlock prevention in DB spend updates (sorted lock ordering across pods), an OpenAPI schema fix so WebSocket stubs no longer overwrite HTTP operations on shared paths, OTel Responses API support (output messages, tool calls, status-based finish_reasons, system_instructions coalescing), OVHCloud field migration shims for the 2026-05-11 API transition, and compliance route access for non-admin roles.

• Deadlock prevention (db_spend_update_writer.py, proxy/utils.py): all six spend-bucket loops now iterate in sorted() order so concurrent pods acquire row-level locks in a consistent sequence. A comprehensive parametrized test suite verifies the ordering across all buckets.
• OpenAPI schema fix (proxy_server.py): _inject_websocket_stubs_into_openapi_schema now uses setdefault + a \"get\" not in path_entry guard, preventing the synthetic WebSocket GET stub from silently dropping a real POST (e.g. /v1/responses) from Swagger UI.
• OTel + OVHCloud (opentelemetry.py, ovhcloud/): Responses API output is now surfaced as OTEL GenAI 1.38 span attributes; OVHCloud seconds→duration and reasoning→reasoning_content normalisations handle falsy-zero edge cases correctly with is not None guards.

Confidence Score: 5/5

Safe to merge — all changes are additive or narrowly scoped bug fixes with targeted test coverage.

Every change is either a clear correctness fix (deadlock ordering, OpenAPI stub injection) or a non-breaking addition (OTel Responses API, OVHCloud migration shims, compliance route access). The deadlock fix is mechanical and consistent across all six spend buckets. The OpenAPI refactor adds a guard that only skips the stub when a real operation already exists, which is strictly safer than before. No auth logic is altered, no DB schema is changed, and no backwards-incompatible interfaces are introduced.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/integrations/opentelemetry.py	Adds OTel Responses API support: system_instructions coalescing (system_instructions/instructions/system kwargs), output message transformation, tool-call span attributes, and status-based finish_reasons. Logic is well-guarded and new helper methods are cleanly separated.
litellm/proxy/db/db_spend_update_writer.py	Adds sorted() around all spend-bucket iteration loops to enforce deterministic lock-acquisition order across pods and prevent PostgreSQL deadlocks. Change is consistent across all six transaction buckets (user, key, team, team_member, org, entity).
litellm/proxy/proxy_server.py	Extracts WebSocket stub injection into _inject_websocket_stubs_into_openapi_schema(); now uses setdefault + 'get' not in path_entry guard so HTTP operations on shared paths (e.g. POST /v1/responses) are never overwritten by a synthetic WebSocket GET stub.
litellm/proxy/_types.py	Adds compliance_check_routes (['/compliance/eu-ai-act', '/compliance/gdpr']) to internal_user_routes and internal_user_view_only_routes so non-admin roles can reach stateless compliance validators.
litellm/proxy/utils.py	Applies the same sorted() deadlock-prevention fix to ProxyUpdateSpend.update_end_user_spend, consistent with the db_spend_update_writer.py changes.
litellm/llms/ovhcloud/audio_transcription/transformation.py	Handles OVHCloud field migration: prefers seconds over duration (using is not None to correctly pass through 0.0), normalises to duration key for downstream consumers through the transition deadline (2026-05-11).
litellm/llms/ovhcloud/chat/transformation.py	Handles OVHCloud reasoning field migration: maps new reasoning field to reasoning_content only when reasoning_content is absent, preserving legacy behavior during transition window.

_{Reviews (2): Last reviewed commit: "Merge PR #26670 into agent staging branc..." | Re-trigger Greptile}

[greptile-apps]

The comment is now inaccurate. The module docstring added by this PR is itself an expression statement (an Expr node in Python's AST), so from __future__ import annotations is no longer the first non-comment statement. PEP 236 explicitly permits module docstrings before future imports, so behaviour is unchanged — but the inline explanation is misleading for anyone reading it in future.

Suggested change

	# from __future__ import annotations must be the first non-comment statement
	from __future__ import annotations
	# from __future__ import annotations must appear before any other imports/code
	# (module docstring above is permitted by PEP 236)
	from __future__ import annotations

[codecov]

Codecov Report

❌ Patch coverage is 97.26027% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/integrations/opentelemetry.py	96.49%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[krrish-berri-2]

@veria-labs please review

[krrish-berri-2]

@greptile please re-review