ML-Flow & ML-Skinny installed togehter - Fails on resolution for conda

[rxm7706]

Solution to issue cannot be found in the documentation.

• [X] I checked the documentation.

Issue

I was not expecting that ML-FLOW skinny & ML-FLOW could not be installed together using conda.

My assumption was that most packages that inherited mlflow as a dependency; would move to changing their dependency to mlfow-skinny (similar to matplotlib vs matplotlib-base) to avoid package bloat.

Using PiP & Poetry I was able to install mlflow & mlflow-skinny in the same environment - using conda I received this error

Could not solve for environment specs The following packages are incompatible ├─ mlflow 2.8.1 would require │ └─ mlflow-skinny <0a0 ,

It probably comes down to how we have defined run constraints in the conda recipe

  run_constrained:
    - mlflow{{ mlflow_other }} <0a0

stack-trace

├─ mlflow-skinny is installable with the potential options
│  ├─ mlflow-skinny [1.22.0|1.23.0|...|2.8.1], which can be installed;
│  ├─ mlflow-skinny [1.14.0|1.14.1|...|1.20.2] would require
│  │  └─ python >=3.6,<3.7.0a0 , which can be installed;
│  ├─ mlflow-skinny [1.14.0|1.14.1|...|1.30.0] would require
│  │  └─ python >=3.7,<3.8.0a0 , which can be installed;
│  ├─ mlflow-skinny [1.14.0|1.14.1|...|2.8.1] would require
│  │  └─ python >=3.8,<3.9.0a0 , which can be installed;
│  ├─ mlflow-skinny [1.14.0|1.14.1|...|2.8.1] would require
│  │  └─ python >=3.9,<3.10.0a0 , which can be installed;
│  └─ mlflow-skinny [2.3.2|2.4.0|...|2.8.1] would require
│     └─ python >=3.11,<3.12.0a0 , which can be installed;
└─ mlflow >=2.8.1  is installable with the potential options
   ├─ mlflow 2.8.1 would require
   │  └─ mlflow-skinny <0a0 , which conflicts with any installable versions previously reported;
   ├─ mlflow 2.8.1 would require
   │  └─ python >=3.11,<3.12.0a0 , which can be installed;
   ├─ mlflow 2.8.1 would require
   │  └─ python >=3.8,<3.9.0a0 , which can be installed;
   └─ mlflow 2.8.1 would require
      └─ python >=3.9,<3.10.0a0 , which can be installed.

Installed packages

"mlflow>=2.8.1" "mlflow-skinny>=2.8.1"

Environment info

linux python 3.10

[xhochy]

Both have the same files. Should we make mlflow depend on mlflow-skinny im general?

[rxm7706]

Both have the same files. Should we make mlflow depend on mlflow-skinny im general?

That's possible - I didn't think of that

I was thinking of trying

  run_constrained:
    - mlflow{{ mlflow_other }} <0a0

  run_constrained:
    - mlflow{{ mlflow_other }} == {version)

Looking at how source is organized ; your suggestion seems to be a cleaner approach. https://github.com/mlflow/mlflow/tree/master/requirements https://github.com/mlflow/mlflow/blob/master/setup.py

LMK which approach - you'd like me to try -- or if we should try both

[xhochy]

I would prefer the one where mlflow depends on mlflow-skinny. This doesn't match the pip based approach but is cleaner and something we can do in conda here which is probably not that simple with pip.

[rxm7706]

@xhochy - thank you, understood, I can take a shot at it next week.