Implementation:Bentoml BentoML Configuration Containers

Knowledge Sources	Bentoml_BentoML
Domains	Configuration, Dependency Injection, Infrastructure
Last Updated	2026-02-13 15:00 GMT

Overview

The BentoMLContainer is the central dependency injection container that provides configuration, store instances, tracing, metrics, and other shared resources throughout the BentoML framework.

Description

This module defines two core components:

BentoMLConfiguration: A class that loads, merges, and validates BentoML configuration from multiple sources in priority order:

1. Default configuration (from the versioned spec module)
2. Override defaults (programmatic)
3. Configuration file overrides (BENTOML_CONFIG file)
4. JSON configuration overrides
5. Environment variable overrides (BENTOML_CONFIG_OPTIONS)

Configuration values support environment variable expansion and are validated against a schema.

_BentoMLContainerClass / BentoMLContainer: A dataclass-based dependency injection container using the simple_di providers pattern. It provides singleton factory methods and static providers for:

• Storage paths: bentoml_home, bento_store_dir, model_store_dir, env_store_dir, tmp_bento_store_dir
• Store instances: bento_store, model_store, tmp_bento_store
• Server configuration: api_server_config, runners_config, http, grpc, ssl, cors
• Observability: tracer_provider (OpenTelemetry with Zipkin/Jaeger/OTLP exporters), metrics_client (Prometheus), duration_buckets
• Cloud integration: rest_api_client, bentocloud_client, cloud_config
• Runtime state: session_id, worker_index, development_mode, serialization_strategy

Usage

Used pervasively throughout BentoML as the central configuration and service locator. Other modules inject dependencies from BentoMLContainer using Provide[BentoMLContainer.xxx] decorators. The container is initialized at import time via load_config().

Code Reference

Source Location

• Repository: Bentoml_BentoML
• File: src/bentoml/_internal/configuration/containers.py
• Lines: 1-509

Signature

class BentoMLConfiguration:
    def __init__(self, override_config_file: str | None = None,
                 override_config_values: str | None = None,
                 override_defaults: dict[str, Any] | None = None,
                 override_config_json: dict[str, Any] | None = None,
                 *, validate_schema: bool = True, use_version: int = 1): ...
    def to_dict(self) -> providers.ConfigDictType: ...

@dataclass
class _BentoMLContainerClass:
    config = providers.Configuration()
    model_aliases = providers.Static({})

    # SingletonFactory providers
    bentoml_home: str
    bento_store: BentoStore
    model_store: ModelStore
    tracer_provider: TracerProvider
    metrics_client: PrometheusClient
    rest_api_client: RestApiClient
    bentocloud_client: BentoCloudClient
    # ... and more

BentoMLContainer = _BentoMLContainerClass()

Import

from bentoml._internal.configuration.containers import BentoMLContainer
from bentoml._internal.configuration.containers import BentoMLConfiguration

I/O Contract

Inputs

BentoMLConfiguration.__init__()

Name	Type	Required	Description
override_config_file	str or None	No	Path to a YAML configuration file to override defaults
override_config_values	str or None	No	Space-separated key=value config overrides (from env var)
override_defaults	dict or None	No	Programmatic dictionary of default overrides
override_config_json	dict or None	No	JSON dictionary of configuration overrides
validate_schema	bool	No	Whether to validate configuration against schema (default: True)
use_version	int	No	Configuration spec version to use (default: 1)

Outputs

Name	Type	Description
BentoMLContainer.bentoml_home	str	The root BentoML home directory path (default: ~/bentoml)
BentoMLContainer.bento_store	BentoStore	Singleton BentoStore for managing local bentos
BentoMLContainer.model_store	ModelStore	Singleton ModelStore for managing local models
BentoMLContainer.tracer_provider	TracerProvider	OpenTelemetry tracer provider configured per settings
BentoMLContainer.metrics_client	PrometheusClient	Prometheus metrics client
BentoMLContainer.rest_api_client	RestApiClient	REST API client for BentoCloud
BentoMLContainer.config	Configuration	The merged configuration dictionary

Usage Examples

from bentoml._internal.configuration.containers import BentoMLContainer

# Access configuration values
api_server_config = BentoMLContainer.api_server_config

# Get the model store
model_store = BentoMLContainer.model_store.get()

# Use in dependency injection
from simple_di import Provide, inject

@inject
def my_function(model_store=Provide[BentoMLContainer.model_store]):
    models = model_store.list()
    return models

# Access bentoml home directory
home = BentoMLContainer.bentoml_home.get()

Related Pages

• Implementation:Bentoml_BentoML_Cloud_ModelAPI
• Implementation:Bentoml_BentoML_YataiClient