2010YOUY01 commented on code in PR #14149: URL: https://github.com/apache/datafusion/pull/14149#discussion_r1918087137
########## datafusion/physical-plan/src/sorts/row_serde.rs: ########## @@ -0,0 +1,407 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +use arrow::row::RowConverter; +use arrow::row::Rows; +use datafusion_common::error::DataFusionError; +use datafusion_common::Result; +use std::fs::File; +use std::future::Future; +use std::io::BufWriter; +use std::io::Write; +use std::io::{BufReader, Read, Seek, SeekFrom}; +use std::path::Path; + +use std::sync::Arc; + +use tokio::sync::mpsc::Sender; + +use crate::stream::ReceiverStreamBuilder; +/// used for spill Rows +pub struct RowStreamBuilder { + inner: ReceiverStreamBuilder<Rows>, +} + +impl RowStreamBuilder { + pub fn new(capacity: usize) -> Self { + Self { + inner: ReceiverStreamBuilder::new(capacity), + } + } + + pub fn tx(&self) -> Sender<Result<Rows, DataFusionError>> { + self.inner.tx() + } + + pub fn spawn<F>(&mut self, task: F) + where + F: Future<Output = Result<(), DataFusionError>>, + F: Send + 'static, + { + self.inner.spawn(task) + } + + pub fn spawn_blocking<F>(&mut self, f: F) + where + F: FnOnce() -> Result<(), DataFusionError>, + F: Send + 'static, + { + self.inner.spawn_blocking(f) + } +} + +#[derive(Clone, Copy, Debug)] +pub enum CompressionType { + UNCOMPRESSED, + #[cfg(feature = "compress")] + GZIP, +} + +/// +----------------+------------------+----------------+------------------+ +/// | Block1 Data | Block1 Metadata | Block2 Data | Block2 Metadata | ... +/// +----------------+------------------+----------------+------------------+ +/// | FileMetadata | MetadataLength | +/// +----------------+------------------+ +#[derive(Debug)] +pub struct RowWriter { + writer: BufWriter<File>, + block_offsets: Vec<u64>, + current_offset: u64, + compression: CompressionType, +} + +impl RowWriter { + pub fn new( + path: &Path, + compression: Option<CompressionType>, + ) -> Result<Self, DataFusionError> { + let file = File::create(path).map_err(|e| { + DataFusionError::Execution(format!("Failed to create file at {path:?}: {e}")) Review Comment: IO Error? ########## datafusion/physical-plan/src/sorts/row_serde.rs: ########## @@ -0,0 +1,407 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +use arrow::row::RowConverter; +use arrow::row::Rows; +use datafusion_common::error::DataFusionError; +use datafusion_common::Result; +use std::fs::File; +use std::future::Future; +use std::io::BufWriter; +use std::io::Write; +use std::io::{BufReader, Read, Seek, SeekFrom}; +use std::path::Path; + +use std::sync::Arc; + +use tokio::sync::mpsc::Sender; + +use crate::stream::ReceiverStreamBuilder; +/// used for spill Rows +pub struct RowStreamBuilder { + inner: ReceiverStreamBuilder<Rows>, +} + +impl RowStreamBuilder { + pub fn new(capacity: usize) -> Self { + Self { + inner: ReceiverStreamBuilder::new(capacity), + } + } + + pub fn tx(&self) -> Sender<Result<Rows, DataFusionError>> { + self.inner.tx() + } + + pub fn spawn<F>(&mut self, task: F) + where + F: Future<Output = Result<(), DataFusionError>>, + F: Send + 'static, + { + self.inner.spawn(task) + } + + pub fn spawn_blocking<F>(&mut self, f: F) + where + F: FnOnce() -> Result<(), DataFusionError>, + F: Send + 'static, + { + self.inner.spawn_blocking(f) + } +} + +#[derive(Clone, Copy, Debug)] +pub enum CompressionType { + UNCOMPRESSED, + #[cfg(feature = "compress")] + GZIP, +} + +/// +----------------+------------------+----------------+------------------+ +/// | Block1 Data | Block1 Metadata | Block2 Data | Block2 Metadata | ... +/// +----------------+------------------+----------------+------------------+ +/// | FileMetadata | MetadataLength | +/// +----------------+------------------+ +#[derive(Debug)] +pub struct RowWriter { + writer: BufWriter<File>, + block_offsets: Vec<u64>, + current_offset: u64, + compression: CompressionType, +} + +impl RowWriter { + pub fn new( + path: &Path, + compression: Option<CompressionType>, + ) -> Result<Self, DataFusionError> { + let file = File::create(path).map_err(|e| { + DataFusionError::Execution(format!("Failed to create file at {path:?}: {e}")) + })?; + + Ok(Self { + writer: BufWriter::new(file), + block_offsets: Vec::new(), + current_offset: 0, + compression: compression.unwrap_or(CompressionType::UNCOMPRESSED), + }) + } + + pub fn write_rows(&mut self, rows: &Rows) -> Result<(), DataFusionError> { + self.block_offsets.push(self.current_offset); + let (row_data, row_offsets) = self.prepare_row_data(rows)?; + let compressed_data = self.compress(&row_data)?; + + self.writer.write_all(&compressed_data)?; + + self.write_block_metadata(&row_offsets)?; + + self.current_offset += + (compressed_data.len() + self.metadata_size(&row_offsets)) as u64; + + Ok(()) + } + + fn prepare_row_data( + &self, + rows: &Rows, + ) -> Result<(Vec<u8>, Vec<u32>), DataFusionError> { + let mut row_offsets = Vec::with_capacity(rows.num_rows()); + let mut current_offset = 0u32; + let mut row_data = Vec::new(); + + for i in 0..rows.num_rows() { + row_offsets.push(current_offset); + let row = rows.row(i).data(); + row_data.extend_from_slice(row); + current_offset += row.len() as u32; + } + + Ok((row_data, row_offsets)) + } + + fn write_block_metadata( + &mut self, + row_offsets: &[u32], + ) -> Result<(), DataFusionError> { + for &offset in row_offsets { + self.writer.write_all(&offset.to_le_bytes())?; + } + self.writer + .write_all(&(row_offsets.len() as u32).to_le_bytes())?; + Ok(()) + } + + fn metadata_size(&self, row_offsets: &[u32]) -> usize { + 4 + // row count + row_offsets.len() * 4 // row offsets + } + + pub fn finish(mut self) -> Result<(), DataFusionError> { Review Comment: I think we need some mechanism to prevent writing again into a finished `RowsWriter` ``` writer.write_rows(batch1); writer.write_rows(batch2); writer.finsh(); writer.write_rows(batch3); // should fail ``` ########## datafusion/physical-plan/src/sorts/row_serde.rs: ########## @@ -0,0 +1,407 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +use arrow::row::RowConverter; +use arrow::row::Rows; +use datafusion_common::error::DataFusionError; +use datafusion_common::Result; +use std::fs::File; +use std::future::Future; +use std::io::BufWriter; +use std::io::Write; +use std::io::{BufReader, Read, Seek, SeekFrom}; +use std::path::Path; + +use std::sync::Arc; + +use tokio::sync::mpsc::Sender; + +use crate::stream::ReceiverStreamBuilder; +/// used for spill Rows +pub struct RowStreamBuilder { + inner: ReceiverStreamBuilder<Rows>, +} + +impl RowStreamBuilder { + pub fn new(capacity: usize) -> Self { + Self { + inner: ReceiverStreamBuilder::new(capacity), + } + } + + pub fn tx(&self) -> Sender<Result<Rows, DataFusionError>> { + self.inner.tx() + } + + pub fn spawn<F>(&mut self, task: F) + where + F: Future<Output = Result<(), DataFusionError>>, + F: Send + 'static, + { + self.inner.spawn(task) + } + + pub fn spawn_blocking<F>(&mut self, f: F) + where + F: FnOnce() -> Result<(), DataFusionError>, + F: Send + 'static, + { + self.inner.spawn_blocking(f) + } +} + +#[derive(Clone, Copy, Debug)] +pub enum CompressionType { + UNCOMPRESSED, + #[cfg(feature = "compress")] + GZIP, +} + +/// +----------------+------------------+----------------+------------------+ +/// | Block1 Data | Block1 Metadata | Block2 Data | Block2 Metadata | ... +/// +----------------+------------------+----------------+------------------+ +/// | FileMetadata | MetadataLength | +/// +----------------+------------------+ +#[derive(Debug)] +pub struct RowWriter { + writer: BufWriter<File>, + block_offsets: Vec<u64>, + current_offset: u64, + compression: CompressionType, +} + +impl RowWriter { + pub fn new( + path: &Path, + compression: Option<CompressionType>, + ) -> Result<Self, DataFusionError> { + let file = File::create(path).map_err(|e| { + DataFusionError::Execution(format!("Failed to create file at {path:?}: {e}")) + })?; + + Ok(Self { + writer: BufWriter::new(file), + block_offsets: Vec::new(), + current_offset: 0, + compression: compression.unwrap_or(CompressionType::UNCOMPRESSED), + }) + } + + pub fn write_rows(&mut self, rows: &Rows) -> Result<(), DataFusionError> { + self.block_offsets.push(self.current_offset); + let (row_data, row_offsets) = self.prepare_row_data(rows)?; + let compressed_data = self.compress(&row_data)?; + + self.writer.write_all(&compressed_data)?; + + self.write_block_metadata(&row_offsets)?; + + self.current_offset += + (compressed_data.len() + self.metadata_size(&row_offsets)) as u64; + + Ok(()) + } + + fn prepare_row_data( + &self, + rows: &Rows, + ) -> Result<(Vec<u8>, Vec<u32>), DataFusionError> { + let mut row_offsets = Vec::with_capacity(rows.num_rows()); + let mut current_offset = 0u32; + let mut row_data = Vec::new(); + + for i in 0..rows.num_rows() { Review Comment: I think we can directly copy all rows' data at once, instead of one by one. But perhaps we can leave those performance-related improvements to a follow on PR, and set up a benchmark first. I suspect we can also miss some other unnecessary mem copies 🤔 ########## datafusion/physical-plan/src/sorts/row_serde.rs: ########## @@ -0,0 +1,407 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +use arrow::row::RowConverter; +use arrow::row::Rows; +use datafusion_common::error::DataFusionError; +use datafusion_common::Result; +use std::fs::File; +use std::future::Future; +use std::io::BufWriter; +use std::io::Write; +use std::io::{BufReader, Read, Seek, SeekFrom}; +use std::path::Path; + +use std::sync::Arc; + +use tokio::sync::mpsc::Sender; + +use crate::stream::ReceiverStreamBuilder; +/// used for spill Rows +pub struct RowStreamBuilder { + inner: ReceiverStreamBuilder<Rows>, +} + +impl RowStreamBuilder { + pub fn new(capacity: usize) -> Self { + Self { + inner: ReceiverStreamBuilder::new(capacity), + } + } + + pub fn tx(&self) -> Sender<Result<Rows, DataFusionError>> { + self.inner.tx() + } + + pub fn spawn<F>(&mut self, task: F) + where + F: Future<Output = Result<(), DataFusionError>>, + F: Send + 'static, + { + self.inner.spawn(task) + } + + pub fn spawn_blocking<F>(&mut self, f: F) + where + F: FnOnce() -> Result<(), DataFusionError>, + F: Send + 'static, + { + self.inner.spawn_blocking(f) + } +} + +#[derive(Clone, Copy, Debug)] +pub enum CompressionType { + UNCOMPRESSED, + #[cfg(feature = "compress")] + GZIP, +} + +/// +----------------+------------------+----------------+------------------+ +/// | Block1 Data | Block1 Metadata | Block2 Data | Block2 Metadata | ... +/// +----------------+------------------+----------------+------------------+ +/// | FileMetadata | MetadataLength | +/// +----------------+------------------+ +#[derive(Debug)] +pub struct RowWriter { + writer: BufWriter<File>, + block_offsets: Vec<u64>, + current_offset: u64, + compression: CompressionType, +} + +impl RowWriter { + pub fn new( + path: &Path, + compression: Option<CompressionType>, + ) -> Result<Self, DataFusionError> { + let file = File::create(path).map_err(|e| { + DataFusionError::Execution(format!("Failed to create file at {path:?}: {e}")) + })?; + + Ok(Self { + writer: BufWriter::new(file), + block_offsets: Vec::new(), + current_offset: 0, + compression: compression.unwrap_or(CompressionType::UNCOMPRESSED), + }) + } + + pub fn write_rows(&mut self, rows: &Rows) -> Result<(), DataFusionError> { + self.block_offsets.push(self.current_offset); + let (row_data, row_offsets) = self.prepare_row_data(rows)?; + let compressed_data = self.compress(&row_data)?; + + self.writer.write_all(&compressed_data)?; + + self.write_block_metadata(&row_offsets)?; + + self.current_offset += + (compressed_data.len() + self.metadata_size(&row_offsets)) as u64; + + Ok(()) + } + + fn prepare_row_data( + &self, + rows: &Rows, + ) -> Result<(Vec<u8>, Vec<u32>), DataFusionError> { + let mut row_offsets = Vec::with_capacity(rows.num_rows()); + let mut current_offset = 0u32; + let mut row_data = Vec::new(); + + for i in 0..rows.num_rows() { + row_offsets.push(current_offset); + let row = rows.row(i).data(); + row_data.extend_from_slice(row); + current_offset += row.len() as u32; + } + + Ok((row_data, row_offsets)) + } + + fn write_block_metadata( + &mut self, + row_offsets: &[u32], + ) -> Result<(), DataFusionError> { + for &offset in row_offsets { + self.writer.write_all(&offset.to_le_bytes())?; + } + self.writer + .write_all(&(row_offsets.len() as u32).to_le_bytes())?; + Ok(()) + } + + fn metadata_size(&self, row_offsets: &[u32]) -> usize { + 4 + // row count + row_offsets.len() * 4 // row offsets + } + + pub fn finish(mut self) -> Result<(), DataFusionError> { + let metadata = self.prepare_file_metadata()?; + let compressed_metadata = self.compress(&metadata)?; + + self.writer.write_all(&compressed_metadata)?; + self.writer + .write_all(&(compressed_metadata.len() as u32).to_le_bytes())?; + self.writer.flush()?; + + Ok(()) + } + + fn prepare_file_metadata(&self) -> Result<Vec<u8>, DataFusionError> { + let mut metadata = Vec::new(); + metadata.extend_from_slice(&(self.block_offsets.len() as u32).to_le_bytes()); + for &offset in &self.block_offsets { + metadata.extend_from_slice(&(offset as u32).to_le_bytes()); + } + Ok(metadata) + } + + fn compress(&self, data: &[u8]) -> Result<Vec<u8>, DataFusionError> { + match self.compression { + CompressionType::UNCOMPRESSED => Ok(data.to_vec()), + #[cfg(feature = "compress")] + CompressionType::GZIP => { + let mut encoder = flate2::write::GzEncoder::new( + Vec::new(), + flate2::Compression::default(), + ); + encoder.write_all(data)?; + Ok(encoder.finish()?) + } + } + } +} + +pub struct RowReader { + reader: BufReader<File>, + block_offsets: Vec<u64>, + current_block: usize, + compression: CompressionType, + converter: Arc<RowConverter>, +} + +impl RowReader { + pub fn new( + path: &Path, + compression: Option<CompressionType>, + converter: Arc<RowConverter>, + ) -> Result<Self, DataFusionError> { + let mut reader = BufReader::new(File::open(path)?); + let compression = compression.unwrap_or(CompressionType::UNCOMPRESSED); + let block_offsets = Self::read_file_metadata(&mut reader, compression)?; + + Ok(Self { + reader, + block_offsets, + current_block: 0, + compression, + converter, + }) + } + + fn read_file_metadata( + reader: &mut BufReader<File>, + compression: CompressionType, + ) -> Result<Vec<u64>, DataFusionError> { + reader.seek(SeekFrom::End(-4))?; + let mut len_buf = [0u8; 4]; + reader.read_exact(&mut len_buf)?; + let metadata_len = u32::from_le_bytes(len_buf) as u64; + + reader.seek(SeekFrom::End(-(metadata_len as i64 + 4)))?; + let mut metadata_buf = vec![0u8; metadata_len as usize]; + reader.read_exact(&mut metadata_buf)?; + let metadata = Self::decompress(&metadata_buf, compression)?; + + let block_offsets = metadata[4..] + .chunks_exact(4) + .map(|chunk| u32::from_le_bytes(chunk.try_into().unwrap()) as u64) + .collect(); + + Ok(block_offsets) + } + + fn read_block(&mut self, block_idx: usize) -> Result<Rows, DataFusionError> { + let start_offset = self.block_offsets[block_idx]; + let end_offset = if block_idx + 1 < self.block_offsets.len() { + self.block_offsets[block_idx + 1] + } else { + let metadata_pos = self.reader.seek(SeekFrom::End(-4))?; + let mut len_buf = [0u8; 4]; + self.reader.read_exact(&mut len_buf)?; + let metadata_len = u32::from_le_bytes(len_buf); + metadata_pos - metadata_len as u64 + }; + + self.reader.seek(SeekFrom::Start(end_offset - 4))?; + let mut len_buf = [0u8; 4]; + self.reader.read_exact(&mut len_buf)?; + let row_count = u32::from_le_bytes(len_buf) as usize; + + let offsets_size = row_count * 4; + self.reader + .seek(SeekFrom::Start(end_offset - 4 - offsets_size as u64))?; + let mut offsets = Vec::with_capacity(row_count); + for _ in 0..row_count { + let mut offset_buf = [0u8; 4]; + self.reader.read_exact(&mut offset_buf)?; + offsets.push(u32::from_le_bytes(offset_buf) as usize); + } + + let data_size = (end_offset - start_offset - 4 - offsets_size as u64) as usize; + self.reader.seek(SeekFrom::Start(start_offset))?; + let mut buffer = vec![0u8; data_size]; + self.reader.read_exact(&mut buffer)?; + let data = Self::decompress(&buffer, self.compression)?; + + self.build_rows(&data, &offsets) + } + + fn build_rows( + &self, + data: &[u8], + row_offsets: &[usize], + ) -> Result<Rows, DataFusionError> { + let mut rows = self.converter.empty_rows(row_offsets.len(), data.len()); + + for i in 0..row_offsets.len() { + let start = row_offsets[i]; + let end = if i + 1 < row_offsets.len() { + row_offsets[i + 1] + } else { + data.len() + }; + + rows.push(self.converter.parser().parse(&data[start..end])); + } + + Ok(rows) + } + + fn decompress( + data: &[u8], + compression: CompressionType, + ) -> Result<Vec<u8>, DataFusionError> { + match compression { + CompressionType::UNCOMPRESSED => Ok(data.to_vec()), + #[cfg(feature = "compress")] + CompressionType::GZIP => { + let mut decoder = flate2::read::GzDecoder::new(data); + let mut decoded_data = Vec::new(); + decoder.read_to_end(&mut decoded_data)?; + Ok(decoded_data) + } + } + } +} + +impl Iterator for RowReader { + type Item = Result<Rows, DataFusionError>; + + fn next(&mut self) -> Option<Self::Item> { + if self.current_block >= self.block_offsets.len() { + return None; + } + + let result = self.read_block(self.current_block); + self.current_block += 1; + Some(result) + } +} +#[cfg(test)] +mod tests { + use arrow::{ + compute::concat_batches, + row::{RowConverter, SortField}, + }; + use arrow_array::{ArrayRef, Int32Array, RecordBatch}; + use arrow_schema::{DataType, Field, Schema}; + use datafusion_common::DataFusionError; + use std::sync::Arc; + use tempfile::NamedTempFile; + + use crate::sorts::row_serde::{RowReader, RowWriter}; Review Comment: Maybe we can enumerate more tests for edge cases: - call `write_rows()` multiple times - Write only one row - Write one row for multiple times - Include variable length field like string -- 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: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
