sunchao opened a new issue, #11275: URL: https://github.com/apache/arrow-rs/issues/11275
### Is your feature request related to a problem or challenge? `ParquetMetaDataPushDecoder` discovers the page-index byte span by walking every row group's column metadata. It performs this walk even when both index policies are `Skip`, and repeats it each time `try_decode()` returns to page-index loading while the requested bytes are still unavailable. For metadata with many column chunks, these traversals repeat work whose result has not changed. ### Describe the solution you'd like Return immediately when both index policies are `Skip`. Otherwise, retain the discovered range while waiting for its bytes, still checking buffer availability on every decode attempt. Invalidate the retained range when either index policy changes and consume it before decoding or returning an error. The requested ranges, decoded metadata, and public APIs should remain unchanged. Regression coverage should include repeated polls, clearing buffered ranges, and policy changes while a range is pending. ### Describe alternatives you've considered Continue recomputing the range on every call. The proposed change retains only one optional range per decoder. ### Additional context OpenAI Codex assisted with the implementation port, tests, review, and this issue description. -- 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]
