BerriAI/litellm · error · ValueError

Max depth of {DEFAULT_MAX_RECURSE_DEPTH} exceeded while proc

Error message

Max depth of {DEFAULT_MAX_RECURSE_DEPTH} exceeded while processing schema.

What it means

Recursive schema sanitizer guard: _fix_enum_empty_strings (or a sibling processor) exceeded DEFAULT_MAX_RECURSE_DEPTH while walking the tool/response_schema, protecting against pathological or cyclic nested schemas.

Source

Thrown at litellm/llms/vertex_ai/common_utils.py:512

def _check_text_in_content(parts: list[PartType]) -> bool:
    """
    check that user_content has 'text' parameter.
        - Known Vertex Error: Unable to submit request because it must have a text parameter.
        - 'text' param needs to be present (empty strings are valid)
        - Relevant Issue: https://github.com/BerriAI/litellm/issues/5515
    """
    has_text_param = False
    for part in parts:
        if "text" in part and part.get("text") is not None:
            has_text_param = True

    return has_text_param


def _fix_enum_empty_strings(schema, depth=0):
    """Fix empty strings in enum values by replacing them with None. Gemini doesn't accept empty strings in enums."""
    if depth > DEFAULT_MAX_RECURSE_DEPTH:
        raise ValueError(f"Max depth of {DEFAULT_MAX_RECURSE_DEPTH} exceeded while processing schema.")

    if "enum" in schema and isinstance(schema["enum"], list):
        schema["enum"] = [None if value == "" else value for value in schema["enum"]]

    # Reuse existing recursion pattern from convert_anyof_null_to_nullable
    properties: Final = schema.get("properties", None)
    if properties is not None:
        for _, value in properties.items():
            _fix_enum_empty_strings(value, depth=depth + 1)

    items: Final = schema.get("items", None)
    if items is not None:
        _fix_enum_empty_strings(items, depth=depth + 1)


def _fix_enum_types(schema, depth=0):
    """Remove `enum` fields when the schema type is not string.

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Reduce nesting depth of the schema.
2. Flatten or simplify the JSON schema passed in the request.

Example fix

# remove deeply nested $ref/anyOf chains from the response schema.

When it happens

Trigger: Triggered when schema processing exceeds DEFAULT_MAX_RECURSE_DEPTH due to excessive nesting.

Common situations: See trigger scenarios.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/3829b752bbde83cf. Report an issue: GitHub.