Litellm tuesday cicd release final

[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

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 3, 2026 9:56am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR reverts three previous features and fixes test failures in the MCP semantic tool filter tests.

Reverted Changes:

• Access Group Filtering (fix: models loadbalancing billing issue by filter #18891/fix: models loadbalancing billing issue by filter (#18891) #19220): Removed the deployment filtering logic that prevented cross-team load balancing based on access groups. This included removing filter_deployments_by_access_groups() from litellm/router_utils/common_utils.py, get_access_groups_from_models() from litellm/proxy/auth/model_checks.py, and the access group extraction logic from litellm/proxy/litellm_pre_call_utils.py.
• VerificationToken Indexes (add missing indexes on VerificationToken table #20040): Removed database indexes on the LiteLLM_VerificationToken table (user_id + team_id, team_id, budget_reset_at + expires) from all three schema.prisma files and deleted the migration file.
• Fallback Exhaustion Check (fix: #19985 prevent error when max_fallbacks exceeds available models  #20071): Removed the additional check that prevented errors when max_fallbacks exceeds available models. The code now only checks if fallback_depth >= max_fallbacks without checking the fallback list length.

Test Fixes:

• Fixed MCP semantic tool filter tests by calling filter_instance._build_router(tools) before filtering and initializing the metadata dict for the hook to store filter stats.
• Added semantic-router import check with @pytest.mark.skipif to skip tests when the package is not installed.

Documentation:

• Added four new MCP semantic filter configuration environment variables to the documentation.

Confidence Score: 4/5

• This PR is safe to merge as it reverts previously committed features and fixes test failures
• Score reflects that this is a revert PR for CICD stabilization. The reverts are clean and comprehensive, removing all related code. Test fixes are straightforward. Score is 4 instead of 5 because reverting features suggests the original implementations had issues, and the PR description lacks information about why these reverts were necessary or what problems they caused.
• No files require special attention - all changes are clean reverts

Important Files Changed

Filename	Overview
litellm/proxy/auth/model_checks.py	Removed get_access_groups_from_models function and minor formatting changes
litellm/proxy/litellm_pre_call_utils.py	Removed access group extraction logic for router filtering (reverted)
litellm/router.py	Removed access group filtering from router deployment selection (reverted)
litellm/router_utils/common_utils.py	Removed filter_deployments_by_access_groups function and minor formatting
litellm/router_utils/fallback_event_handlers.py	Removed fallback list exhaustion check from fallback handler (reverted)
schema.prisma	Removed VerificationToken table indexes from schema (reverted)

Sequence Diagram

sequenceDiagram
    participant Client
    participant ProxyServer as Proxy Server
    participant PreCallUtils as litellm_pre_call_utils
    participant Router
    participant Fallback as Fallback Handler
    participant LLM

    Note over ProxyServer,Router: REVERTED: Access Group Filtering
    Client->>ProxyServer: Request with API Key
    ProxyServer->>PreCallUtils: add_litellm_data_to_request()
    Note over PreCallUtils: ❌ REMOVED: Extract access groups<br/>from key/team models
    PreCallUtils->>Router: Request without access group metadata
    
    Note over Router: ❌ REMOVED: filter_deployments_by_access_groups()
    Router->>Router: Select deployment (no access group filtering)
    Router->>LLM: Forward request
    
    alt Request Fails
        LLM-->>Router: Error
        Router->>Fallback: run_async_fallback()
        Note over Fallback: ❌ REMOVED: Check if fallback list exhausted<br/>(fallback_depth >= fallback_group_length)
        Note over Fallback: ✅ KEPT: Only check max_fallbacks limit
        Fallback->>Router: Try next fallback model
        Router->>LLM: Retry with fallback
        LLM-->>Client: Response
    end
    
    Note over ProxyServer,Router: REVERTED: Database Indexes
    Note over ProxyServer: ❌ REMOVED: Prisma indexes on<br/>VerificationToken table<br/>(user_id, team_id, budget_reset_at, expires)

Loading

[greptile-apps]

_{6 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}