🐾What does retirement of CodeCommit mean for MLOps?🐾
🤓 Recently, AWS quietly deprecated several services, including CodeCommit service, which received highly emotional response from the community. But what is more practical, it leads to the question what does it mean for CodeCommit current users? And the question that instantly came to my mind was: how this change would affect MLOps setup with SageMaker?
Which part of MLOps will be affected?
There are several parts of MLOps setup with SageMaker, and I usually divide it to four parts. Two of them are related to model pipeline itself such as:
model training after commit to repository
model deployment after its approval in Model Registry
Only model training part can be affected with CodeCommit deprecation.
Also, we have some miscellaneous automations, which help during development process, for example automatic attachment of custom kernel images to SageMaker domain. Usually they are not related to choice of git tool.
And the last, but not least part is SageMaker project management, which will be affected mostly by CodeCommit deprecation.
What does it mean for SageMaker current users?
Current users will be able to continue using CodeCommit repositories which were already created, and create new repositories. So, this change wouldn’t affect users who already have MLOps setup with CodeCommit immediately. But, usually when you use some product, you expect it to be updated, bugs to be fixed eventually, etc. IMHO, users would start thinking about moving to other source code management tools. Moreover, it opens discussion about vendor-lock again, because now customers are varied about other AWS-native services that can be quietly deprecated and it can be an argument while choosing which product to use.
What does it mean for SageMaker new clients?
From my perspective, it wouldn't make any sense to use CodeBuild and CodePipeline services for model training anymore, because it will be much easier to use native CI/CD tools (such as GitLab CI in case you use GitLab repositories).
Moreover, it would be hard to setup SageMaker project management without CodeCommit. Currently, you can use EventBridge events for creation/deletion of branches to add/delete new Model Registry groups for each new branch. It makes development process easier for data scientists and eliminates possibility of human errors during resources creation. Other git tools such as GitLab currently have poor or no integration with EventBridge.
🎁 What is the future of code management on AWS?
I heard some rumours about AWS thinking on acquiring GitLab. No official statements for AWS, but it makes sense for me because Azure, their main competitor, owns GitHub, which powers their Azure DevOps system. So, it looks like AWS needs their code management tool as well. Let me know in comments what do you think about such possibility🤓
Thank you for reading, let’s chat 💬
💬 What do you think about CodeCommit deprecation?
💬 Do you use CodeCommit to store your ML code?
💬 Are you planning to migrate to other git tools?
I love hearing from readers 🫶🏻 Please feel free to drop comments, questions, and opinions below👇🏻