I’m only familiar with speeding up Parquet - it looks like you’re mainly sorting or partitioning the data into different views so that you can choose the best v

I’m only familiar with speeding up Parquet - it looks like you’re mainly sorting or partitioning the data into different views so that you can choose the best view format at runtime based on your desired query or aggregation? We’ve seen this increase speeds by many orders of magnitude, so I wouldn’t be surprised that this creates meaningful speed-ups when done automatically (Which is cool!). Random aside, how do you handle the consistency problems that can occur when you have multiple views when doing deletes?