Litellm oss staging 02 14 20262

[Sameerlite]

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 16, 2026 1:36pm

[greptile-apps]

Greptile Summary

This staging PR bundles five independent changes across the codebase:

• S3 virtual-hosted-style URLs: Adds s3_use_virtual_hosted_style option to S3Logger for S3-compatible services (like MinIO) that require bucket.endpoint/key URL format instead of the default path-style endpoint/bucket/key. Includes docs and comprehensive tests.
• Anthropic effort="max": Extends effort validation to accept "max" as a valid effort level, gated to Claude Opus 4.6 only. Other models correctly reject this value.
• Shared aiohttp session fix: Introduces owns_session parameter to AiohttpTransport to prevent closing shared ClientSession instances. When a shared session is reused, owns_session=False is passed so aclose() skips closing it.
• Duplicate agent spend computation fix: Removes a redundant call to _common_add_spend_log_transaction_to_daily_transaction in add_spend_log_transaction_to_daily_agent_transaction that was computing the daily transaction twice — once before and once after the agent_id null check.
• Langfuse test isolation: Fixes flaky Langfuse tests by using reset_mock() on shared setUp mocks instead of creating entirely new mock objects that bypassed the test class's mock chain.

All changes include corresponding tests and no critical issues were found.

Confidence Score: 4/5

• This PR is safe to merge with minimal risk — all changes are well-tested and isolated.
• Each of the five changes is straightforward with clear intent. The S3 virtual-hosted-style feature adds a new opt-in configuration. The aiohttp session fix and agent spend deduplication address real bugs with targeted tests. The Anthropic effort="max" change is model-gated. Only minor style concerns exist (duplicated URL construction, missing trailing newline). Score is 4 instead of 5 because this is a bundled staging PR covering multiple unrelated changes, which increases review surface area.
• litellm/integrations/s3_v2.py has duplicated URL construction logic that would benefit from a helper method extraction.

Important Files Changed

Filename	Overview
litellm/integrations/s3_v2.py	Adds virtual-hosted-style URL support for S3-compatible services. URL construction logic is duplicated 3 times but follows existing pattern. Missing trailing newline at end of file.
litellm/llms/anthropic/chat/transformation.py	Adds effort="max" as a valid option for Claude Opus 4.6, with model-gated validation. Clean implementation that correctly reuses existing _is_claude_opus_4_6 helper.
litellm/llms/custom_httpx/aiohttp_transport.py	Adds owns_session parameter to prevent closing shared ClientSession instances. Default is True for backward compatibility. Clean implementation.
litellm/llms/custom_httpx/http_handler.py	Passes owns_session=False when using shared sessions, correctly wiring the new aiohttp_transport feature.
litellm/proxy/db/db_spend_update_writer.py	Removes duplicate _common_add_spend_log_transaction_to_daily_transaction call that was computing the daily transaction twice for agent spend. The common helper is still called once after the agent_id null check.
docs/my-website/docs/proxy/logging.md	Documents the new s3_use_virtual_hosted_style configuration option with clear description.
tests/test_litellm/integrations/test_langfuse.py	Improves test isolation by using setUp's shared mock objects with reset_mock() instead of creating entirely new mocks, preventing flaky failures from state pollution.
tests/test_litellm/integrations/test_s3_v2.py	Comprehensive mock tests for virtual-hosted-style URLs covering async upload, sync upload, download, and protocol variations. No real network calls made.
tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py	Adds tests for effort="max" validation for Opus 4.6 and rejection for non-Opus 4.6 models. Good coverage of the new validation logic.
tests/test_litellm/llms/custom_httpx/test_aiohttp_transport.py	Adds tests for owns_session behavior: shared sessions not closed, owned sessions closed, and default value is True. Uses real aiohttp sessions but no network calls.
tests/test_litellm/proxy/db/test_db_spend_update_writer.py	Adds test verifying the common helper is called exactly once for agent transactions, confirming the duplicate call fix.

Flowchart

flowchart TD
    subgraph S3["S3 URL Construction"]
        A["S3Logger upload/download"] --> B{"s3_endpoint_url set?"}
        B -->|No| C["Default AWS URL: bucket.s3.region.amazonaws.com/key"]
        B -->|Yes| D{"s3_use_virtual_hosted_style?"}
        D -->|True| E["Virtual-hosted: protocol://bucket.endpoint/key"]
        D -->|False| F["Path-style: endpoint/bucket/key"]
    end

    subgraph Aiohttp["Aiohttp Session Management"]
        G["AsyncHTTPHandler"] --> H{"shared_session provided?"}
        H -->|Yes| I["LiteLLMAiohttpTransport\nowns_session=False"]
        H -->|No| J["LiteLLMAiohttpTransport\nowns_session=True"]
        I --> K["aclose() skips session.close()"]
        J --> L["aclose() calls session.close()"]
    end

    subgraph Agent["Agent Spend Transaction"]
        M["add_spend_log_to_daily_agent_txn"] --> N{"agent_id is None?"}
        N -->|Yes| O["Skip - return early"]
        N -->|No| P["_common_add_spend_log...\n(called once)"]
        P --> Q["Queue agent spend update"]
    end

Loading

_{Last reviewed commit: 72a1bd6}

[greptile-apps]

_{11 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (2)

litellm/integrations/s3_v2.py
Missing trailing newline at end of file

The file no longer ends with a newline character. Most editors and POSIX tools expect text files to end with a newline. This can also cause noise in future diffs.

            return None

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

---

litellm/integrations/s3_v2.py
URL construction logic is duplicated three times

The virtual-hosted-style and path-style URL construction block is copy-pasted identically at lines 313-326, 474-487, and 575-588. Consider extracting this into a private helper method like _build_s3_url(self, object_key: str) -> str to reduce duplication and make future changes (e.g., trailing-slash handling) easier to maintain.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}