I'm a software engineer and a bit new to databricks.  My goal is to create a model serving endpoint, that interfaces with several ML models. Traditionally this would look like:

API--> Service --> Data

Now using databricks, my understanding is that it will look like

Models Serving Endpoint --> Service Model --> ML Model

From a best practices perspective what is the best way to deploy? A single dab that bundles the resources to a single cluster? Multiple deployed models/clusters in more of a micro service fashion? 

Also is the service model even necessary? 

I can see benefits to each method. I'm certain there are aspects I'm overlooking.  I'd love to hear how others are deploying