databricks / mlops-stacks

This repo provides a customizable stack for starting new ML projects on Databricks that follow production best-practices out of the box.
https://docs.databricks.com/en/dev-tools/bundles/mlops-stacks.html
Apache License 2.0
460 stars 156 forks source link

Fix model deployment #68

Closed mingyu89 closed 1 year ago

mingyu89 commented 1 year ago

The model deployment was broken due to model layout change. Found by @niall-turbitt https://docs.google.com/document/d/1ChQrfnRxN2CAhECX7xJZw_tmu8PoweI9AlLqgx67Tmk/edit#

The change is to update github action to correctly call deploy and fix the import of utils.


I'm not importing "{{cookiecutter.project_name}}.utils" because currently there's no guarantee that the project_name is also a valid package name.(For example, the "mlops-azure-cuj" is not a valid package name)

After I added the __init__ files, the relative import of "../.." works on my laptop but doesn't work for github actions(maybe because of the current work directory).

I have to use absolute path to import utils and it's working.


Tested on CUJ project. The CUJ deployment was also broken after the layout change yesterday and I didn't notice it. Now it's fixed. https://github.com/databricks/mlops-azure-cuj/actions/runs/4431460872/jobs/7774443532

This is the change made in CUJ project https://github.com/databricks/mlops-azure-cuj/pull/74/files/f0f4df8dec32780fc2a6bd5f317e637a13eac865..1034207c3c9547c825a8bfeaa4e717ad9da39d27