feat: support per-request enable_json_schema_validation for thread safety

[Chesars]

Relevant issues

Fixes thread-safety issue with litellm.enable_json_schema_validation global flag in concurrent environments (FastAPI, multi-threaded services).

Pre-Submission checklist

• 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

Type

🆕 New Feature

Changes

Allow passing enable_json_schema_validation as a per-request parameter to completion() and acompletion(), instead of only relying on the global litellm.enable_json_schema_validation flag.

The global flag isn't thread-safe. In a FastAPI service with concurrent users, one request can change the flag while another is mid-execution:

# Thread A sets it to True
litellm.enable_json_schema_validation = True
# Thread B sets it to False  
litellm.enable_json_schema_validation = False
# Thread A calls completion() — gets False instead of True

Solution: Pass the flag dynamically per-request:

# Thread-safe: each request controls its own validation
completion(..., enable_json_schema_validation=True)
completion(..., enable_json_schema_validation=False)

Priority: per-request value > global value. If not passed, falls back to the global (fully backward compatible).

Files changed:

• litellm/main.py — added enable_json_schema_validation param to completion() and acompletion()
• litellm/types/utils.py — registered in all_litellm_params so it doesn't leak to provider APIs
• litellm/utils.py — post_call_processing() checks per-request value first, falls back to global
• tests/litellm/litellm_core_utils/test_json_schema_validation.py — 6 tests covering the full matrix

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Mar 3, 2026 5:55pm

[greptile-apps]

Greptile Summary

This PR adds per-request enable_json_schema_validation support to completion() and acompletion(), solving a thread-safety issue where concurrent requests sharing the global litellm.enable_json_schema_validation flag could interfere with each other.

• Adds enable_json_schema_validation: Optional[bool] = None parameter to completion() and acompletion() in litellm/main.py
• Updates post_call_processing() in litellm/utils.py to check the per-request value first, falling back to the global flag (fully backward-compatible)
• Registers the param in all_litellm_params to prevent it from leaking to provider APIs
• Includes 6 mock-only tests covering the full matrix of global/per-request flag combinations

Confidence Score: 5/5

• This PR is safe to merge — it's a clean, backward-compatible addition with no risk of breaking existing behavior.
• The changes are minimal and well-scoped: a new optional parameter with None default that falls back to the existing global flag. The param is properly registered in all_litellm_params to prevent leaking. Tests cover all flag combinations without making real network calls. No backwards-incompatible changes.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/main.py	Adds enable_json_schema_validation as an optional parameter to both completion() and acompletion(), and forwards it through completion_kwargs and the MCP handler call. Clean, minimal changes.
litellm/types/utils.py	Registers enable_json_schema_validation in all_litellm_params to prevent the parameter from leaking to provider APIs. Single-line addition.
litellm/utils.py	Updates post_call_processing to check per-request enable_json_schema_validation from optional_params before falling back to the global flag. Logic is correct and backward-compatible.
tests/litellm/litellm_core_utils/test_json_schema_validation.py	Comprehensive test suite covering all 4 combinations of global/per-request flags, valid response pass-through, and param registration. No real network calls—uses mock data directly.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["completion() / acompletion()"] -->|"enable_json_schema_validation param"| B["@client decorator wrapper"]
    B --> C["post_call_processing()"]
    C --> D{"Per-request flag\nin optional_params?"}
    D -->|Yes| E["Use per-request value"]
    D -->|No / None| F["Use global\nlitellm.enable_json_schema_validation"]
    E --> G{"Flag is True?"}
    F --> G
    G -->|Yes| H["Validate response\nagainst JSON schema"]
    G -->|No| I["Skip validation"]
    H -->|Invalid| J["Raise JSONSchemaValidationError"]
    H -->|Valid| K["Continue normally"]

Loading

_{Last reviewed commit: 909e3ce}

[greptile-apps]

_{4 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

litellm/main.py
Missing enable_json_schema_validation in MCP delegation

When completion() delegates to acompletion_with_mcp for MCP tool handling, the new enable_json_schema_validation parameter is not forwarded. Since it's a named parameter in the completion() signature, it is NOT included in **kwargs at line 1173 — it's a separate local variable that gets silently dropped.

This means per-request JSON schema validation will not work when MCP tools are present. Add enable_json_schema_validation=enable_json_schema_validation, before **kwargs:

                shared_session=shared_session,
                enable_json_schema_validation=enable_json_schema_validation,
                **kwargs,