westonpace commented on pull request #11616: URL: https://github.com/apache/arrow/pull/11616#issuecomment-1013430728
> Ah, so we fall back to the current implementation if the file is zero-copy (i.e. a buffer) or we're reading all columns? That sounds reasonable. Yes, that seems to remove the major slowdowns. There is still about a 15-20% slowdown on some of the async pure-buffer operations because we use a bit more Future but this seems acceptable to me now. > It'd be good to revisit this in the context of also adding column/batch statistics for filter pushdown as well. Though at that point we'd basically be Parquet/ORC without any of the fancy data encodings (which might be exactly what we're after). Yes, sounds about right. -- 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]
