Implementation:BerriAI Litellm Logging Setup

Attribute	Value
Sources	litellm/_logging.py
Domains	Logging, Observability, Configuration
last_updated	2026-02-15 16:00 GMT

Overview

The Logging Setup module initializes and configures the Python logging infrastructure for LiteLLM, providing three named loggers, a JSON formatter, and utility functions for controlling log verbosity.

Description

This module is executed at import time and sets up the foundational logging subsystem for the entire LiteLLM library. It creates three named loggers:

verbose_logger -- "LiteLLM" (general library logging)
verbose_router_logger -- "LiteLLM Router" (router/load-balancing logging)
verbose_proxy_logger -- "LiteLLM Proxy" (proxy server logging)

The module reads environment variables to configure behavior:

LITELLM_LOG -- sets the log level (default: DEBUG)
JSON_LOGS -- when truthy, applies the JsonFormatter for structured JSON output

The JsonFormatter class produces JSON log lines with ISO 8601 timestamps, log level, message, and optional stacktrace. It also attempts to parse embedded JSON or Python dict representations in log messages, promoting their fields to top-level JSON keys.

The module suppresses noisy third-party loggers (httpx, APScheduler) by setting them to WARNING level. Helper functions provide runtime control: _turn_on_debug(), _disable_debugging(), _enable_debugging(), _turn_on_json(), and _is_debugging_on().

Usage

Import the loggers directly for use throughout the LiteLLM codebase:

from litellm._logging import verbose_logger

Code Reference

Source Location

/litellm/_logging.py (352 lines)

Key Symbols

Symbol	Type	Description
`verbose_logger`	`logging.Logger`	Main LiteLLM logger
`verbose_router_logger`	`logging.Logger`	Router subsystem logger
`verbose_proxy_logger`	`logging.Logger`	Proxy server logger
`JsonFormatter`	`class(Formatter)`	Formats log records as JSON
`_turn_on_debug`	`def _turn_on_debug() -> None`	Sets all loggers to DEBUG level
`_turn_on_json`	`def _turn_on_json() -> None`	Switches all loggers to JSON formatting
`_is_debugging_on`	`def _is_debugging_on() -> bool`	Returns whether DEBUG logging is active
`_get_uvicorn_json_log_config`	`def _get_uvicorn_json_log_config() -> dict`	Generates uvicorn log_config dict for JSON mode
`print_verbose`	`def print_verbose(print_statement) -> None`	Legacy verbose printing (deprecated)

Import

from litellm._logging import verbose_logger, verbose_proxy_logger, verbose_router_logger

I/O Contract

Inputs (Environment Variables)

Variable	Type	Description
`LITELLM_LOG`	`str`	Log level string (e.g., `"DEBUG"`, `"INFO"`)
`JSON_LOGS`	`str`	When truthy, enables JSON log formatting

Outputs

Export	Type	Description
`verbose_logger`	`logging.Logger`	Pre-configured logger for general LiteLLM use
`verbose_router_logger`	`logging.Logger`	Pre-configured logger for router subsystem
`verbose_proxy_logger`	`logging.Logger`	Pre-configured logger for proxy server
`ALL_LOGGERS`	`list[logging.Logger]`	List of all loggers including root

Usage Examples

from litellm._logging import verbose_logger, _turn_on_debug, _is_debugging_on

# Standard logging
verbose_logger.info("Processing request for model: %s", model_name)

# Enable debug logging at runtime
_turn_on_debug()

# Check if debugging is active
if _is_debugging_on():
    verbose_logger.debug("Detailed parameter dump: %s", params)

# JSON log output example (when JSON_LOGS=True)
# {"message": "Processing request for model: gpt-4", "level": "INFO", "timestamp": "2026-02-15T16:00:00"}

Related Pages

BerriAI_Litellm_Service_Logger - uses these loggers for service health monitoring
BerriAI_Litellm_Logging_Worker - background logging worker that uses verbose_logger
BerriAI_Litellm_Logging_Callback_Manager - manages callback registration and invocation

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment