adriangb opened a new pull request, #25132: URL: https://github.com/apache/datafusion/pull/25132
**Do not review or merge — this PR exists only so the benchmark bot can reach the ref.** It measures option 1 of https://github.com/apache/datafusion/issues/25116 (record the groups a batch has rows for, keep the per-group scratch vectors) as the counterfactual for https://github.com/apache/datafusion/pull/25123, which additionally replaces those vectors with a counting sort. #25123 is a reproducible +19% on `clickbench_extended` Q14 and +6.5% on Q5, but a reproducible −10% on Q2, which groups by a 110-value column with four `COUNT(DISTINCT)` adapters. This branch is the half of that change that has the same sign at every cardinality, and the question is how much of the win it keeps. Will be closed once the numbers are in. 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
