Pcodec is awesome! 15% smaller and 2x faster than ClickHouse's DoubleDelta

Two weeks ago, I posted about how ClickHouse outcompressed Parquet on my IoT dataset. But just days ago, pcodec was merged into ClickHouse, and it significantly improves on compression ratio while also trading blows on decompression speed with the previous schema. It's a rather niche and sophisticated codec though, at least compared to the ones we had before in ClickHouse, so I thought I'd make a post where I try to explain the key conceptual ideas behind it. Enjoy!