We have ETL pipelines that write millions of records daily from OLAP to OLTP databases using REST APIs. We’re exploring db options that would provide very high write throughput, that doesn’t degrade overtime, and ensures reads from our apis are not overloaded by the high write throughput.
We’re exploring ScyllaDB, but our current schema is document based (denormalized) and not normalized. Thoughts?
As you require high write throughput, your use case is a good fit for ScyllaDB.
With regards to denormalization and data modeling in NoSQL vs. SQL in general, I recommend the Data Modeling course on ScyllaDB University
and specifically the Denormalization lesson.
Another feature that might be relevant for your use case is Workload Prioritization. It allows users to run OLAP and OLTP workloads on ScyllaDB without sacrificing latency or throughput.