Querying Parquet with Millisecond Latency

I would guess that it's mostly because of tooling and library availability. There has been very little cross language support for reading and writing Parquet files. Mostly it was concentrated in Java and C++. But since it fell under the Arrow umbrella, that situation has been improving.

My bet is that it will be much more widely adopted in the coming years. There are more tools and systems that accept and return Parquet files for bulk data (see most data warehouses). And there are libraries in most languages that are improving every day for working withe the format. Although, I'd obviously believe that as one of the creators of InfluxDB IOx :)

The format itself is, I think, a real pain to work with without libraries to hide its complexity. But the advantages are big enough that people are building those libraries to make it more broadly adoptable.