One thing about columnar store that annoys me, is that when you do want to retrieve the records for something you've found it's really slow. I wish there was something of a hybrid.
We have something similar to that at Singlestore (formerly memsql). We support secondary keys and compression schemes that are incremental so you can decompress a few rows out of our columnstore without needing to decompressing the data for other rows stored near by like a traditional columnstore (Redshift, snowflake, etc). We call this subsegment access [1].
I’m pretty sure Snowflake is hybrid like you’re describing. It certainly performs like it. Check out their original white paper for the architecture details
