AzureML Pipeline Checklist

As part of a project, I created a couple of Azure ML pipelines. There were a lot of learnings. I want this cheat sheet to serve as a good reference point. This checklist by any means is not complete and may differ from other projects’ implementations. I wanted this checklist as a quick note to remind myself of all the common things that are applicable in future projects that I/ others do.

The cheat sheet is divided into 3 sections: Development, Production and Cleanup

Development

• Have 2 workspaces: dev and prod
• Version control in place
• Have the Python development environment in place
• No hard-coded passwords/ API keys/ secrets in the code
• Automatic shutdown of development compute instance each day
• Set a budget and email notifications on cost
• Use YAML files for configuration

Production

• Use service principal instead of using users credentials
• Enable a schedule or trigger-based pipeline
• Automatic scaling down of computing target
• Enable CI/ CD
• Enable monitoring solutions to track metrics, input and prediction data

Cleanup

• Disable/ delete schedules and pipelines that are not required anymore
• Lifecycle management policies in place to delete the log files
• Remove access of people not required in the project

This list is just a dump of all things I could remember. I will keep on expanding points but in a new blog post.

I hope you find it useful. Please let me know your experience if you are reading this and what else can be added to this cheat sheet.

Thank you for reading.