Suggestions for CDC driven data ingestion

Thanks for the reply Prashanth.  A couple of notes, while I research.  

In general, there always seems to be some minor to significant latency in spinning up batch jobs, often adding between 2 and 4 minutes to any notebook, pipeline execution, or Spark job. I was attempting to avoid this accumulated latency with something closer to a long-running job, but this may be better suited to Azure functions or service. 

Regarding points 2 & 3, my existing spark job is already doing this effectively caching in memory and batching table upserts/deletes every 2-5 minutes.    

On #4, there are config settings in Debezium that prevent full table-locks during snapshot, yet still provide acceptable concurrancy in a majority of cases. They just aren't made available within the generic EventStream setup in Fabric.  Probably a big miss in my opinion.