Support for MLflow 2.0

[mb1069]

Hi,
Is kedro_mlflow going to have support for MLflow 2.0 and above in the near future?

Thank you for an awesome package!

[Galileo-Galilei]

Hi, I obviously do plan to support mlflow>=2.0, but I haven't decided yet what the best solution is:

• in an ideal world, I'd love to support mlflow~1.X AND mlflow ~2.X. This requires monkey patching a bunch of mlflow functions (especially the MlflowClient) and this is both a lot of work to setup and generates a lot of maintenance burden (lots of tests are breaking, I need a CI to test both scenario, I need to monkey patch all test.s, there is always some very weird side effects which can happen. by modifiying the inner working of mlflow..). Unfortunately, this scenario is very unlikely unless I have some help to maintain this package. Feel free to volunteer 😄
• The most realistic scenario is that I will drop mlflow~1.X support at some point and start to support mlflow 2.0. My problem here is that I think that most people have not migrated to mlflow~2.X yet (I can't verify it from user research, but if people come and upvote your question I will reconsider this).

➡️ The current scenario I have in mind is to move to mlflow 2 after moving to kedro ==0.19. My intuition is that it could be done in rougly 1 year, say early 2024.

If many people complain about it, I may create a branch to enable installing a development version of kedro_mlflow from github instead of PyPI which would be compatible with mlflow 2.

Have you already migrated to mlflow 2 in your org?

[Galileo-Galilei]

Hi @mb1069 , @edhenry, @melaanya,

can you give it a try to by installing this PR: pip install git+https://github.com/Galileo-Galilei/kedro-mlflow.git@mlflow2? I think I will be able to support both mlflow 1 & 2 and I can release it very soon but I'd like to have your feeback first.

[mb1069]

Hi @Galileo-Galilei , I've run your PR locally and it hasn't broken our workflow / deployment! Looking forward to the release.