Managed batches - Address PR bot comments from #22464

[ephrimstanley]

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/test_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	Error		Mar 3, 2026 4:02pm

[ephrimstanley]

@greptile-apps

[greptile-apps]

Greptile Summary

This PR addresses three review bot comments from #22464 for the managed batches / Vertex AI files feature:

• file_retrieve Literal type fix: Adds "gemini" and "vertex_ai" to the custom_llm_provider Literal in the sync file_retrieve function, aligning it with the already-updated afile_retrieve. Without this, passing these providers would fail type-checking and fall through to a BadRequestError.
• GCS delete URI reconstruction fix: Fixes transform_delete_file_response to include the bucket name in the reconstructed gs:// URI by extracting the bucket from the /b/ segment of the GCS URL, not just the object path from /o/.
• Logging improvement: Removes an unnecessary hasattr guard on httpx.HTTPStatusError.response and switches from f-string to lazy %-formatting in the error logger for the async batch create path.

All changes are well-scoped to the Vertex AI provider directory (per custom rule on provider-specific code) and accompanied by mock-only regression tests.

Confidence Score: 5/5

• This PR is safe to merge — it fixes three small, well-tested bugs with no behavioral regressions.
• All three fixes are straightforward, well-scoped, and backed by mock-only regression tests. The Literal type fix aligns with the already-correct async counterpart, the GCS URI fix correctly reconstructs the full gs:// path, and the logging change is a clean improvement. No new dependencies, no architectural changes, no risk of regressions.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/files/main.py	Added "gemini" and "vertex_ai" to the Literal type for custom_llm_provider in file_retrieve, aligning it with afile_retrieve. Correct fix.
litellm/llms/vertex_ai/batches/handler.py	Removed unnecessary hasattr guard on httpx.HTTPStatusError (always has .response) and switched to lazy %-formatting for logger. Clean improvement.
litellm/llms/vertex_ai/files/transformation.py	Fixed transform_delete_file_response to include the GCS bucket name in the reconstructed gs:// URI by also splitting on /b/. Bug fix with good guard clause.
tests/test_litellm/llms/vertex_ai/files/test_file_retrieve_provider_routing.py	New mock-only test file for file_retrieve routing with vertex_ai and gemini providers. Tests are well-structured and use proper mocking.
tests/test_litellm/llms/vertex_ai/files/test_vertex_ai_files_transformation.py	Updated existing test assertion and added two regression tests for the bucket-in-delete-id fix. All mock-only, no network calls.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["afile_retrieve(custom_llm_provider='vertex_ai')"] --> B["file_retrieve(custom_llm_provider='vertex_ai')"]
    B --> C{"Provider in OpenAI-compatible list?"}
    C -->|No| D{"Provider == 'azure'?"}
    D -->|No| E["ProviderConfigManager.get_provider_files_config()"]
    E --> F["base_llm_http_handler.retrieve_file()"]
    F --> G["GCS API call"]
    G --> H["Return OpenAIFileObject"]

    I["transform_delete_file_response()"] --> J{"URL contains /b/ and /o/?"}
    J -->|Yes| K["Extract bucket from /b/ segment"]
    K --> L["Extract object from /o/ segment"]
    L --> M["Reconstruct gs://bucket/object"]
    J -->|No| N["Fallback: id = 'deleted'"]

Loading

_{Last reviewed commit: b83373d}