Issues Policy acknowledgement <li class="task-li

Yuki, is there any specific reason why we set local here? </blockquot

Seems it's updated here <div class="Box-he

[BUG] latest mlflow version does not support conda environment docker about mlflow HOT 4 CLOSED

menno4000 commented on June 10, 2024

[BUG] latest mlflow version does not support conda environment docker

from mlflow.

Comments (4)

B-Step62 commented on June 10, 2024 1

Yuki, is there any specific reason why we set local here?

The original motivation was multiple requests saying the docker image doesn't need additional virtualization (virtualenv/conda). But I think we shouldn't override what user specifies, so the better way would be default to local and let users to pass either (note that if the base image is not python, we should default to virtualenv because direct python installation to ubuntu image is troublesome). How does that sound?

from mlflow.

serena-ruan commented on June 10, 2024

Seems it's updated here

mlflow/mlflow/pyfunc/backend.py

Lines 379 to 382 in c2d13b1

    
           # We don't need virtualenv or conda if base image is python 
        
           env_manager = ( 
        
               _EnvManager.LOCAL if base_image.startswith("python") else self._env_manager 
        
           )

@B-Step62 Yuki, is there any specific reason why we set local here? If it's not consistent with the env_manager user set then maybe we should emit a warning at least 🤔

from mlflow.

menno4000 commented on June 10, 2024

Passing the input would definitely be appreciated, though the newest python base image might not be inherently equipped to facilitate a conda environment yaml file. Looking at the code of models.docker_utils, the step of the conda installation itself are not applied to the python base images.

mlflow/mlflow/models/docker_utils.py

Lines 83 to 98 in c2d13b1

    
           if base_image.startswith("python:"): 
        
               setup_python_venv_steps = ( 
        
                   "RUN apt-get -y update && apt-get install -y --no-install-recommends nginx" 
        
               ) 
        
               setup_java_steps = "" 
        
               install_mlflow_steps = _pip_mlflow_install_step(output_dir, mlflow_home) 
        
           elif base_image == UBUNTU_BASE_IMAGE: 
        
               setup_python_venv_steps = ( 
        
                   "RUN apt-get -y update && DEBIAN_FRONTEND=noninteractive TZ=Etc/UTC apt-get install -y " 
        
                   "--no-install-recommends wget curl nginx ca-certificates bzip2 build-essential cmake " 
        
                   "git-core\n\n" 
        
               ) 
        
               setup_python_venv_steps += ( 
        
                   SETUP_MINICONDA if env_manager == em.CONDA else SETUP_PYENV_AND_VIRTUALENV 
        
               )

In any case, adding the conda installation back would be huge as it allows us for working with the generated dockerfiles in combination with the conda environment management.

from mlflow.

serena-ruan commented on June 10, 2024

Close as the PR is merged, pls feel free to test against the latest master :)

from mlflow.

Recommend Projects

[BUG] latest mlflow version does not support conda environment docker about mlflow HOT 4 CLOSED

Comments (4)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	# We don't need virtualenv or conda if base image is python
	env_manager = (
	_EnvManager.LOCAL if base_image.startswith("python") else self._env_manager
	)

	if base_image.startswith("python:"):
	setup_python_venv_steps = (
	"RUN apt-get -y update && apt-get install -y --no-install-recommends nginx"
	)
	setup_java_steps = ""
	install_mlflow_steps = _pip_mlflow_install_step(output_dir, mlflow_home)

	elif base_image == UBUNTU_BASE_IMAGE:
	setup_python_venv_steps = (
	"RUN apt-get -y update && DEBIAN_FRONTEND=noninteractive TZ=Etc/UTC apt-get install -y "
	"--no-install-recommends wget curl nginx ca-certificates bzip2 build-essential cmake "
	"git-core\n\n"
	)
	setup_python_venv_steps += (
	SETUP_MINICONDA if env_manager == em.CONDA else SETUP_PYENV_AND_VIRTUALENV
	)