RyanJamesStewart commented on PR #22416: URL: https://github.com/apache/datafusion/pull/22416#issuecomment-4545459668
The residual Vec is the real exposure: after a long `split_off` chain (the 1M to 1024 case), the final remaining Vec keeps the original allocation, and capacity-based memory accounting will charge for the whole thing until it drops. The push-after-emit caller in `multi_group_by/bytes.rs` is what blocked an unconditional `shrink_to_fit` on the emitted prefix, but that constraint doesn't apply to the residual side. The right place to address it is on the residual at drop or finalize time, not on each emitted chunk. Happy to open a follow-up issue tracking that and link it back here if useful. -- 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]
