ScyllaDB. I was replacing a high-write Postgres instance (10% creations, 90% updates), and was testing out a drop-in replacement that could "just scale." CRDB did wonderfully, but ultimately I decided to rearchitect the problem to fit more of a CRDT model (albeit a crappy version) and just throw an endless stream of events at the DB which get reconstructed when the data is pulled out.

The model ended up working fantastically well, and I can't say enough about how much I like ScyllaDB so far. I will say I wish there was an easier backup method, but I understand that's a big ask on a distributed DB and automating our backups wasn't that difficult.