Hi,

I understand the purpose of the project to build a docker with not only mlflow but also sql, s3, ... dependencies, but I'm unsure about
why the Dockerfile is like it is

I understand the multi-stage docker purpose, but here why can't we just install the dependencies like the following, I don't understnad the lib installation part not the Install build-essential to compile extensions comment ?

FROM python:3.9.16-slim

WORKDIR /mlflow/

COPY pyproject.toml poetry.toml poetry.lock /mlflow/

RUN python -m pip install --upgrade pip
RUN pip install poetry wheel &&  poetry install --no-root --no-dev

ENV PATH=/mlflow/.venv/bin:$PATH
ENV PYTHONUNBUFFERED 1

CMD mlflow server --backend-store-uri sqlite:///:memory --default-artifact-root ./mlruns --host=0.0.0.0 --port=5000

Thanks

Need to upgrade MLFlow to greater than 2.2.1

https://protectai.com/blog/ai-zero-day-an-mlflow-vulnerability-that-exposes-your-entire-ml-system-and-how-to-prevent-it

Describe the bug

When we try to use mlflow docker image on arm64 architecture with Postgres connection, the driver returns the following error.

(psycopg2.OperationalError) SCRAM authentication requires libpq version 10 or above

To Reproduce

Steps to reproduce the behavior:

1. Run the image on arm64 machine with PostgreSQL connection settings.
2. Look up to starting logs
3. See error

Expected behavior

It must start to connect PostgreSQL database.

** Tested system:**

• Architecture: arm64 (Raspberry Pi)
• Cluster: Kubernetes 1.23.5
• Mlflow Docker Image Version: 1.27.0.9

Additional context

You can find more information from this issue.