Deploying multiple sagemaker resources (model, endpoint, config)

[rc19]

Hi there, I am new to terraform and have a novice question.

I am using this package and have a requirement of deploying multiple sagemaker models (not multiple instances of the same model but completely different configs, purpose, etc)

As I understand I can always create multiple modules in my main.tf each with single endpoint configs (model, endpoint, config).

Question1: Is the above the right way?
Question2: If not, is there a way that I only create one module and can somehow have different configs and loop through them for instantiation using this package?
Question 3: If not, should I be writing my own resources to support this use case?

TIA

[SebastianUA]

Hi @rc19,

Thank you very much for the message. Let me answer your questions. I'm supporting this module only on my end.
Sometimes, I don't have enough time to add new features.

At this moment, you need to reuse the approach that you described and run the needed resources in your main.tf file.
If you want to use a loop - you have to develop functionality for this case. We must understand if it will be useful or not because you need to remember the structure of the service (model, endpoint, config, etc.).

If you want, you can develop a high-level diagram to show me what would you like to solve and I will try to help you with it.

When I have time, I'll be able to refact this TF module and add new futures.

Hope, you can understand me.

[rc19]

Makes complete sense. Thank you for writing this beast of a library. It definitely takes considerable effort to keep maintaining it.

Right now I am going with the approach I described as my use case also is small and I will worry about scalability later.

I am still learning about terraform (its been only 2 days 😅 ) so I don't know what all is possible but this is my use case

Currently I have 2 models that I have deployed fro my service. This number can grow to n(may be 10 or even 100 depdnding on scale). Now the difference is that these n models are all different "class of models" and not just different instances of the same model.

We must understand if it will be useful or not because you need to remember the structure of the service

Completely agree I thought if we could have another meta class that holds the config/endpoint/model for one "family of models" than we can create a array of such meta class objects such that every object can have its own config and then inside each such object it's the same way as this module works

Hope it makes sense. I didn't think about how useful that is gonna be if the code becomes too complex.