Model data around the queries and filters users actually need, then test partitioning and ordering on a realistic sample. Measure ingestion and aggregation together before scaling the deployment. Plan retention and backup procedures explicitly, and avoid assuming that an analytical database behaves like a row-oriented transactional system for frequent single-record updates.
This is best suited to high-volume event and analytical reporting workloads. Before committing a larger project, account for these trade-offs: different design trade-offs from transactional databases; clusters require operational expertise.
