DMS to S3 endpoint with CDC : should I use partitioning ?

Question

I'm using DMS to replicate an RDS Aurora MySQL to S3 (FullLoad + CDC), for analytics purpose.
I use the option "partitioned = true" on the target S3 endpoint, because I'm afraid the CDC generates too many objects per prefix.
Unfortunately, the Glue Crawler seems completely lost when I activate partitioning, where as all works like a charm when I don't.

Is partitioning a bad idea in my situation ? (1 object per minute in a single prefix seems problematic anyway).

Should I rather activate the "GlueCatalogGeneration" option in the endpoint ?

Gonzalo Herreros · Accepted Answer

Don't worry about "too many prefixes on the same partition", nowaways s3 can handle that pretty well.  
What is a big issue is generating too many partitions, the purpose of partitions is help the reading reduce the data read but if the partitions are too granular, then it's counterproductive. On most cases hour or even day are the best granularity (depending on how much data, history and how users will use the table)

DMS to S3 endpoint with CDC : should I use partitioning ?

Add your answer

Relevant content