Version 0.21 Goals

[joocer]

Version 0.20 was primarily a performance release. It contained a lot of small changes to improve performance, these were most noticeable on the ClickBench benchmark.

It wasn't without functional improvements, one of the significant capabilities added in this release was Iceberg support.

Version 0.21 will continue with performance improvements, with group by and parallel execution being hot topics. Python 3.13 support gives some opportunity for testing GIL-less performance, although this is likely to be production quality.

Some work on GROUP BY was done in 0.20, introducing incremental group by focusing on the core set of 5 aggregations (count, sum, min, max, avg), this provided some minor improvements in most situations.

Work on improving JOIN speed was also part of 0.20, introduction of a bloom filter to the join provided significant speed up but the application of this improvement is limited to very few scenarios.

[joocer]

A lot of work is going into 0.21 to improve typing of columns. This includes the introduction of element types for array columns.

This work is all about reliability, moving detection of failure scenarios forward to planning time, or providing information to reduce ambiguity during execution.