v0.3.1 -- stable release with OrderedStream

Merge branch 'master' of https://github.com/marsupialtail/quokka

v0.2.9

• Various bug fixes

0.2.3

Added dependency on ldbpy for faster kernels.

Added dockerized setup for Docker on Ray

Added fast covariance, approx quantile, approx median relying on ldbpy

v0.2.1

Fixed a few issues with previous release

• better utils: create_new_cluster should work reliably, if not already
• try to remove dependencies on botocore and fsspec by using Pyarrow readers for everything.

v0.2.0

API is now stable.

• Expression syntax.
• Join ordering.
• Fixed numerous issues for better performance.
• Revamped utils to be more stable and connect to existing Ray cluster.

v0.1.5

• switched aggregations to DuckDB
• switched evaluate in sql_utils to emit polars expression instead of lambda
• added sorted semantics to data sources
• added preliminary support for window functions
• added tutorials
• fixed polars update issue, mandate > 0.16 since changed groupby semantics
• fixed numerous other bugs

0.1.4

Numerous updates including progress bars, backpressure and batch limits.

Also start from clean slate AMI.

v0.1.3

OSDI reproducible.

Note that compared to master the replay tasks are scheduled on to existing TaskManagers.

This is perhaps as performant at least according to benchmarks in the paper but probably not good going forward. ReplayTaskManager is a good thing.

v0.1.2

Should be mostly stable now -- large improvements in CSV and Parquet S3 reader stability.

v0.0.8

Should be first stable-ish release. Tutorials should work without any problems.