MLOPs image: Dataprep failed to start Engine when dotnetcore2 '*.a' files are missing

[abij]

I'm trying to reduce the MLOps image size, because its 4.35GB. I followed some steps to reduce docker image size. Clean conda, remove file that are not needed.

Reproduce:

To reproduce inside current image:

# Start bash the image
docker run -it --rm -v ~/.azure:/root/.azure/ -v ~/.azureml:/root/.azureml/ mcr.microsoft.com/mlops/python

Steps to cleanup, which should end-up in the Docker build of the image.

# The test: A working engine:
python -c 'import logging; logging.basicConfig(level="DEBUG"); from azureml.dataprep.api.engineapi import api; api.EngineAPI()'

# First cleanup step (~ 725Mb):
conda clean -tpiy

# Still working engine:
python -c 'import logging; logging.basicConfig(level="DEBUG"); from azureml.dataprep.api.engineapi import api; api.EngineAPI()'

# Second cleanup step (19392 files)
find /usr/local/envs/mlopspython_ci/ -follow -type f -name '*.pyc' -delete

# Still working engine:
python -c 'import logging; logging.basicConfig(level="DEBUG"); from azureml.dataprep.api.engineapi import api; api.EngineAPI()'

# 3th cleanup step
find /usr/local/envs/mlopspython_ci/ -follow -type f -name '*.a' -delete

# Broken engine:
python -c 'import logging; logging.basicConfig(level="DEBUG"); from azureml.dataprep.api.engineapi import api; api.EngineAPI()'

The required "*.a"-files are part of the site-packages/dotnetcore2/bin/shared/Microsoft.NETCore.App. I ended up excluding this path. My. final image is 1,8 Gb large.

Maybe the files are required by design, then the error message could be more clear. Or the auto-downloader should fetch them.

note: The MLOps images is build FROM conda/miniconda3, the git-repo behind it mentions that the repo is deprecated in favor of ContinuumIO.