Implementation:Bentoml BentoML DefaultMonitor

Knowledge Sources	Bentoml_BentoML
Domains	Monitoring, Logging
Last Updated	2026-02-13 15:00 GMT

Overview

Provides a file-based monitoring implementation that logs inference data and schema to rotating log files in JSON format.

Description

The DefaultMonitor class extends MonitorBase to provide the default monitoring backend for BentoML services. It uses Python's standard logging infrastructure with TimedRotatingFileHandler for data logs and RotatingFileHandler for schema logs. Log data is formatted as JSON using pythonjsonlogger. The monitor automatically creates per-worker log directories and files, segregating schema definitions from actual monitoring data. It includes preserved columns for timestamp, request ID, and trace ID that are automatically appended to each logged record. A YAML-based logging configuration (either default or user-provided) controls the handler and formatter setup.

Usage

Use this class when you need to monitor BentoML service predictions and want data written to local rotating log files. It is the default monitor selected by BentoML when no custom monitoring backend is configured.

Code Reference

Source Location

• Repository: Bentoml_BentoML
• File: src/bentoml/_internal/monitoring/default.py
• Lines: 1-155

Signature

class DefaultMonitor(MonitorBase["JSONSerializable"]):
    PRESERVED_COLUMNS = (COLUMN_TIME, COLUMN_RID, COLUMN_TID) = (
        "timestamp", "request_id", "trace_id",
    )

    def __init__(
        self,
        name: str,
        log_path: str,
        log_config_file: str | None = None,
        **_: t.Any,
    ) -> None: ...

    def _init_logger(self) -> None: ...

    def export_schema(self, columns_schema: dict[str, dict[str, str]]) -> None: ...

    def export_data(
        self,
        datas: dict[str, collections.deque[JSONSerializable]],
    ) -> None: ...

Import

from bentoml._internal.monitoring.default import DefaultMonitor

I/O Contract

Inputs

Name	Type	Required	Description
name	str	Yes	Name of the monitor instance, used for directory naming
log_path	str	Yes	Base directory path where log files will be written
log_config_file	str or None	No	Path to a custom YAML logging configuration file; uses built-in default if None
columns_schema	dict[str, dict[str, str]]	Yes (export_schema)	Schema definition mapping column names to metadata dicts
datas	dict[str, collections.deque[JSONSerializable]]	Yes (export_data)	Data records keyed by column name with deque of values

Outputs

Name	Type	Description
(side effect)	Log files	Schema written to <log_path>/<name>/schema/schema.<worker_id>.log
(side effect)	Log files	Data written to <log_path>/<name>/data/data.<worker_id>.log

Usage Examples

import collections
from bentoml._internal.monitoring.default import DefaultMonitor

# Create monitor instance
monitor = DefaultMonitor(
    name="iris_classifier_prediction",
    log_path="/var/log/bentoml/monitoring",
)

# Export schema
schema = {
    "sepal_length": {"name": "sepal_length", "role": "feature", "type": "numerical"},
    "prediction": {"name": "prediction", "role": "prediction", "type": "categorical"},
}
monitor.export_schema(schema)

# Export data
data = {
    "sepal_length": collections.deque([5.1, 4.9]),
    "prediction": collections.deque(["setosa", "versicolor"]),
}
monitor.export_data(data)

Related Pages

• Implementation:Bentoml_BentoML_OTLPMonitor