Documentation of expected Deployment Patterns

[anishshah97]

Description

There is no clear documentation about how one should approach deployment once using kedro-mlflow (especially with custom models). I would love to see how one integrates CI/CD or other Ops style tools to achieve code -> mlflow model -> deployed model workflows (as many issues are faced trying to use mlflow serve).

Context

Having a clear cut understanding of the deployment patterns one should (subjectively) follow when using kedro and kedro-mlflow makes it a lot easier to understand how to approach MLOps with kedro and kedro-mlflow especially with respect to custom models.

Possible Implementation

Updating the kedro-mlflow-tutorial with a basic example of an end to end workflow.

Possible Alternatives

Adding clear documentation to a README

[Galileo-Galilei]

This is a very good question and I have a lot to say about this. I'll update this answer by the end of the week.

[Galileo-Galilei]

I plan to update kedro-mlflow-tutorial with 2 demo tutorials to show how to deploy a kedro-mlflow model :

• in "batch mode" with airflow
• in "API mode" with docker
  This will be done only after migration to kedro==0.18.0 though, so it will not be available before a couple of months