zlacker

Deleting whole partitions is generally useful strategy. It's like the difference between single inserts and batch inserts (often huge performance difference, and much lower IO)

Since you mentioned Cassandra and TTL, I'll mention ClickHouse, very nice TTL options, splitting into smaller partitions and using "ttl_only_drop_parts=1" has prove itself in the production with big data ingestion rates.

Last, but not the least, I almost always prefer Postgres for data storage needs, one can trust it to be safe and fast enough. Only some specific situations warrant other solutions, but it's a long way until that point (if ever), and better not optimize too early.