Querying Parquet with Millisecond Latency

Almost no adoption of Parquet outside the Spark/Hadoop crowd, not even in the common Linux repos (Debian in particular); anyone have any idea why?