Implementation:Bentoml BentoML EnvManager

Knowledge Sources	Bentoml_BentoML
Domains	Environment Management, Dependency Injection, Bento Execution
Last Updated	2026-02-13 15:00 GMT

Overview

Provides a high-level manager class for creating and managing isolated Python environments (currently Conda) for running BentoML bentos.

Description

The EnvManager class acts as a factory and lifecycle manager for isolated Python environments. It supports two modes of operation: ephemeral environments (created in a temporary directory and automatically cleaned up via weakref.finalize) and persistent environments (stored under the BentoML environment store directory). The constructor uses simple_di dependency injection (@inject) to resolve the env_store path from the BentoML container configuration.

The class provides two factory methods: from_bento() creates an EnvManager from an existing Bento object (deriving the environment name from the bento tag for persistent environments), and from_bentofile() is a placeholder for future support of creating environments directly from a bento info specification.

Currently only conda is supported as an environment type; other values raise NotImplementedError.

Usage

Use EnvManager when you need to run a bento inside an isolated conda environment, such as when the bento has conda-specific dependencies that should not be installed in the host Python environment. Use from_bento() for the most common workflow.

Code Reference

Source Location

• Repository: Bentoml_BentoML
• File: src/bentoml/_internal/env_manager/manager.py
• Lines: 1-73

Signature

class EnvManager:
    environment: Environment

    @inject
    def __init__(
        self,
        env_type: t.Literal["conda"],
        bento: Bento,
        is_ephemeral: bool = True,
        env_name: str | None = None,
        env_store: str = Provide[BentoMLContainer.env_store_dir],
    ): ...

    @classmethod
    def from_bento(
        cls,
        env_type: t.Literal["conda"],
        bento: Bento,
        is_ephemeral: bool = True,
    ) -> EnvManager: ...

    @classmethod
    def from_bentofile(
        cls, env_type: str, bento_info: BentoInfo, is_ephemeral: str
    ) -> EnvManager: ...

Import

from bentoml._internal.env_manager.manager import EnvManager

I/O Contract

Inputs

Name	Type	Required	Description
env_type	Literal["conda"]	Yes	Type of environment to create (currently only "conda")
bento	Bento	Yes	The BentoML Bento object with environment configuration
is_ephemeral	bool	No (default True)	Whether the environment is temporary (auto-cleaned) or persistent
env_name	str or None	No	Name for the environment; auto-generated if not provided
env_store	str	No (injected)	Base directory for persistent environment storage

Outputs

Name	Type	Description
EnvManager	EnvManager	An EnvManager instance with an initialized environment attribute
environment	Environment	The underlying Conda (or future) environment instance

Usage Examples

from bentoml._internal.env_manager.manager import EnvManager
from bentoml._internal.bento.bento import Bento

bento = Bento.from_path("/path/to/bento")

# Create an ephemeral conda environment from a bento
mgr = EnvManager.from_bento(env_type="conda", bento=bento, is_ephemeral=True)

# Run commands in the isolated environment
mgr.environment.run(["python", "main.py"])

# Create a persistent environment (stored under BENTOML_HOME/envs/)
mgr_persistent = EnvManager.from_bento(
    env_type="conda", bento=bento, is_ephemeral=False
)

Related Pages

• Implementation:Bentoml_BentoML_Environment_Types