Our great sponsors
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
Looks like somebody requested it after reading your TIL. https://github.com/pola-rs/polars/issues/12493#issuecomment-...
It will be in the next release. (later today?)
Right, there's all sorts of metadata and often stats included in any parquet file: https://github.com/apache/parquet-format#file-format
The offsets of said metadata are well-defined (i.e. in the footer) so for S3 / blob storage so long as you can efficiently request a range of bytes you can pull the metadata without having to read all the data.
We can run a DuckDb instance (EC2/S3) closer to the data so that sorta helps too.
What I'm really excited about using DuckDB in a similar way to map-reduce. What if there was a way to take some SQL's logical plan and turn it into a physical plan that uses compute resources from a pool serverless DuckDB instances. Starting at the leafs of the graph (physical plan) pulling data from the source (parquet), and returning their completed work up the branches, until it is completed and ready to be used as the results.
I've seen a few examples of this already, but nothing that I would consider production ready. I have a hunch that someone is going to drop such a project on us shortly, and it's going to change a lot of things we have become use to in the data world.
https://github.com/BauplanLabs/quack-reduce