WenyXu commented on issue #3725: URL: https://github.com/apache/incubator-opendal/issues/3725#issuecomment-1845283350
> Got it: > > https://github.com/apache/arrow-rs/blob/1534cc196cb5fca09a093fd7886b6d9bfa9831bd/parquet/src/arrow/async_reader/mod.rs#L164-L178 > > ```rust > impl<T: AsyncRead + AsyncSeek + Unpin + Send> AsyncFileReader for T { > fn get_bytes(&mut self, range: Range<usize>) -> BoxFuture<'_, Result<Bytes>> { > async move { > self.seek(SeekFrom::Start(range.start as u64)).await?; > > let to_read = range.end - range.start; > let mut buffer = Vec::with_capacity(to_read); > let read = self.take(to_read as u64).read_to_end(&mut buffer).await?; > if read != to_read { > return Err(eof_err!("expected to read {} bytes, got {}", to_read, read)); > } > > Ok(buffer.into()) > } > .boxed() > } > } > ``` > > parquet will call `seek` everytime to read a range of file which may lead to many read operations on s3 side. > > Let me find a way to optimize this case. > > Another way is to implement `parquet::arrow::AsyncFileReader` trait by opendal. Yes, I also noticed this. But it's kind of confusing why the `seek` will lead to many read operations on the s3 side. -- 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]
