From bd4572170928b4641ab759bf2da12577d7fce9b4 Mon Sep 17 00:00:00 2001 From: Adrian Garcia Badaracco <1755071+adriangb@users.noreply.github.com> Date: Tue, 22 Sep 2026 20:11:06 -0500 Subject: [PATCH] Fix async/push decoder read when only some columns have an offset index When the page index has an offset index for some projected column chunks but not for others, and the read has a `RowSelection` (or a `RowFilter`), the async reader and the push decoder fail with "Invalid column index N, column was not fetched". The sync reader is not affected. `InMemoryRowGroup::fetch_ranges` did not push a `page_start_offsets` entry for a column without an offset index. `fill_column_chunks` reads these entries by position, so it gave the page offsets to the wrong columns and did not fill the last column. Now `fetch_ranges` pushes `None` for such a column, and `fill_column_chunks` stores the full chunk as `ColumnChunkData::Dense`. Co-Authored-By: Claude Opus 5.5 --- parquet/src/arrow/in_memory_row_group.rs | 48 ++-- .../arrow/push_decoder/reader_builder/data.rs | 2 +- parquet/tests/arrow_reader/mod.rs | 1 + .../arrow_reader/partial_offset_index.rs | 216 ++++++++++++++++++ 4 files changed, 249 insertions(+), 18 deletions(-) create mode 100644 parquet/tests/arrow_reader/partial_offset_index.rs diff --git a/parquet/src/arrow/in_memory_row_group.rs b/parquet/src/arrow/in_memory_row_group.rs index 00917e396b49..e10a5e2f2a4d 100644 --- a/parquet/src/arrow/in_memory_row_group.rs +++ b/parquet/src/arrow/in_memory_row_group.rs @@ -43,8 +43,9 @@ pub(crate) struct InMemoryRowGroup<'a> { pub(crate) struct FetchRanges { /// The byte ranges to fetch pub(crate) ranges: Vec>, - /// If `Some`, the start offsets of each page for each column chunk - pub(crate) page_start_offsets: Option>>, + /// If `Some`, the start offsets of each page for each column chunk, or + /// `None` for a column chunk without an offset index (fetched in full) + pub(crate) page_start_offsets: Option>>>, } impl InMemoryRowGroup<'_> { @@ -70,7 +71,7 @@ impl InMemoryRowGroup<'_> { // If we have a `RowSelection` and an `OffsetIndex` then only fetch // pages required for the `RowSelection` // Consider preallocating outer vec: https://github.com/apache/arrow-rs/issues/8667 - let mut page_start_offsets: Vec> = vec![]; + let mut page_start_offsets: Vec>> = vec![]; let ranges = self .column_chunks @@ -88,6 +89,7 @@ impl InMemoryRowGroup<'_> { let Some(offset_idx) = page_index.offset_index(idx) else { // No offset index for this column, fetch the entire column ranges.push(start..start + len); + page_start_offsets.push(None); return ranges; }; @@ -106,7 +108,7 @@ impl InMemoryRowGroup<'_> { } else { ranges.extend(selection.scan_ranges(offset_idx.page_locations())); } - page_start_offsets.push(ranges.iter().map(|range| range.start).collect()); + page_start_offsets.push(Some(ranges.iter().map(|range| range.start).collect())); ranges }) @@ -141,7 +143,7 @@ impl InMemoryRowGroup<'_> { pub(crate) fn fill_column_chunks( &mut self, projection: &ProjectionMask, - page_start_offsets: Option>>, + page_start_offsets: Option>>>, chunk_data: I, ) where I: IntoIterator, @@ -158,20 +160,32 @@ impl InMemoryRowGroup<'_> { continue; } - if let Some(offsets) = page_start_offsets.next() { - let mut chunks = Vec::with_capacity(offsets.len()); - for _ in 0..offsets.len() { - chunks.push(chunk_data.next().unwrap()); + match page_start_offsets.next() { + // No offset index: `fetch_ranges` requested the whole chunk + Some(None) => { + if let Some(data) = chunk_data.next() { + *chunk = Some(Arc::new(ColumnChunkData::Dense { + offset: metadata.column(idx).byte_range().0 as usize, + data, + })); + } } + Some(Some(offsets)) => { + let mut chunks = Vec::with_capacity(offsets.len()); + for _ in 0..offsets.len() { + chunks.push(chunk_data.next().unwrap()); + } - *chunk = Some(Arc::new(ColumnChunkData::Sparse { - length: metadata.column(idx).byte_range().1 as usize, - data: offsets - .into_iter() - .map(|x| x as usize) - .zip(chunks) - .collect(), - })) + *chunk = Some(Arc::new(ColumnChunkData::Sparse { + length: metadata.column(idx).byte_range().1 as usize, + data: offsets + .into_iter() + .map(|x| x as usize) + .zip(chunks) + .collect(), + })) + } + None => {} } } } else { diff --git a/parquet/src/arrow/push_decoder/reader_builder/data.rs b/parquet/src/arrow/push_decoder/reader_builder/data.rs index 3beaa1250455..ab628bf28581 100644 --- a/parquet/src/arrow/push_decoder/reader_builder/data.rs +++ b/parquet/src/arrow/push_decoder/reader_builder/data.rs @@ -39,7 +39,7 @@ pub(super) struct DataRequest { ranges: Vec>, /// Optional page start offsets for each requested range. This is used /// to create the relevant InMemoryRowGroup - page_start_offsets: Option>>, + page_start_offsets: Option>>>, } impl DataRequest { diff --git a/parquet/tests/arrow_reader/mod.rs b/parquet/tests/arrow_reader/mod.rs index c7c77e0a444c..f16cdf2b30ee 100644 --- a/parquet/tests/arrow_reader/mod.rs +++ b/parquet/tests/arrow_reader/mod.rs @@ -47,6 +47,7 @@ mod invalid_utf8; mod io; mod large_string_overflow; mod parquet_testing; +mod partial_offset_index; #[cfg(feature = "async")] mod predicate_cache; mod row_filter; diff --git a/parquet/tests/arrow_reader/partial_offset_index.rs b/parquet/tests/arrow_reader/partial_offset_index.rs new file mode 100644 index 000000000000..72489da5c64a --- /dev/null +++ b/parquet/tests/arrow_reader/partial_offset_index.rs @@ -0,0 +1,216 @@ +// 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. + +//! Tests for reading files in which only some column chunks have an offset index + +use arrow::compute::kernels::cmp::{gt_eq, lt}; +use arrow::compute::{and, concat_batches}; +use arrow_array::{ArrayRef, Int32Array, RecordBatch}; +use bytes::Bytes; +use parquet::DecodeResult; +use parquet::arrow::arrow_reader::{ + ArrowPredicateFn, ArrowReaderBuilder, ArrowReaderMetadata, ArrowReaderOptions, + ParquetRecordBatchReaderBuilder, RowFilter, RowSelection, RowSelector, +}; +use parquet::arrow::push_decoder::ParquetPushDecoderBuilder; +use parquet::arrow::{ArrowWriter, ProjectionMask}; +use parquet::column::writer::ColumnCloseResult; +use parquet::errors::Result; +use parquet::file::metadata::{PageIndexPolicy, ParquetMetaDataReader}; +use parquet::file::properties::WriterProperties; +use parquet::file::writer::SerializedFileWriter; +use std::sync::Arc; + +#[test] +fn test_no_offset_index_first_column() { + check(|_row_group, column| column != 0); +} + +#[test] +fn test_no_offset_index_middle_column() { + check(|_row_group, column| column != 1); +} + +#[test] +fn test_no_offset_index_last_column() { + check(|_row_group, column| column != 2); +} + +#[test] +fn test_no_offset_index_row_group() { + check(|row_group, _column| row_group != 1); +} + +/// 400 rows in columns "a" (0..400), "b" (400..800) and "c" (800..1200) +fn test_batch() -> RecordBatch { + let column = |start| Arc::new(Int32Array::from_iter_values(start..start + 400)) as ArrayRef; + RecordBatch::try_from_iter([("a", column(0)), ("b", column(400)), ("c", column(800))]).unwrap() +} + +/// Returns [`test_batch`] as a file with 2 row groups of 200 rows and data +/// pages of 50 rows. A column chunk has a page index only if +/// `has_index(row_group, column)` is `true`. +fn test_file(has_index: impl Fn(usize, usize) -> bool) -> Bytes { + let batch = test_batch(); + let props = WriterProperties::builder() + .set_max_row_group_row_count(Some(200)) + .set_data_page_row_count_limit(50) + .set_write_batch_size(50) + .build(); + let mut buf = Vec::new(); + let mut writer = ArrowWriter::try_new(&mut buf, batch.schema(), Some(props)).unwrap(); + writer.write(&batch).unwrap(); + writer.close().unwrap(); + let source = Bytes::from(buf); + + // Copy the column chunks to a new file, without the page index of some chunks + let metadata = ParquetMetaDataReader::new() + .with_page_index_policy(PageIndexPolicy::Required) + .parse_and_finish(&source) + .unwrap(); + let page_index = metadata.page_index().unwrap(); + let schema = metadata.file_metadata().schema_descr().root_schema_ptr(); + let mut buf = Vec::new(); + let mut writer = SerializedFileWriter::new(&mut buf, schema, Default::default()).unwrap(); + for (row_group, row_group_metadata) in metadata.row_groups().iter().enumerate() { + let mut row_group_writer = writer.next_row_group().unwrap(); + for (column, column_metadata) in row_group_metadata.columns().iter().enumerate() { + let keep = has_index(row_group, column); + let close = ColumnCloseResult { + bytes_written: column_metadata.compressed_size() as u64, + rows_written: row_group_metadata.num_rows() as u64, + metadata: column_metadata.clone(), + bloom_filter: None, + column_index: page_index + .column_index(row_group, column) + .filter(|_| keep) + .cloned(), + offset_index: page_index + .offset_index(row_group, column) + .filter(|_| keep) + .cloned(), + }; + row_group_writer.append_column(&source, close).unwrap(); + } + row_group_writer.close().unwrap(); + } + writer.close().unwrap(); + Bytes::from(buf) +} + +/// Makes the reader select rows 150..250 +#[derive(Debug, Clone, Copy)] +enum Trigger { + RowSelection, + RowFilter, +} + +fn options() -> ArrowReaderOptions { + ArrowReaderOptions::new().with_page_index_policy(PageIndexPolicy::Optional) +} + +fn configure(builder: ArrowReaderBuilder, trigger: Trigger) -> ArrowReaderBuilder { + match trigger { + Trigger::RowSelection => builder.with_row_selection(RowSelection::from(vec![ + RowSelector::skip(150), + RowSelector::select(100), + RowSelector::skip(150), + ])), + Trigger::RowFilter => { + let mask = ProjectionMask::roots(builder.parquet_schema(), [0]); + let predicate = ArrowPredicateFn::new(mask, |batch: RecordBatch| { + let a = batch.column(0); + and( + >_eq(a, &Int32Array::new_scalar(150))?, + <(a, &Int32Array::new_scalar(250))?, + ) + }); + builder.with_row_filter(RowFilter::new(vec![Box::new(predicate)])) + } + } +} + +fn read_sync(file: &Bytes, trigger: Trigger) -> Result> { + let builder = ParquetRecordBatchReaderBuilder::try_new_with_options(file.clone(), options())?; + Ok(configure(builder, trigger) + .build()? + .collect::>()?) +} + +#[cfg(feature = "async")] +fn read_async(file: &Bytes, trigger: Trigger) -> Result> { + use futures::TryStreamExt; + use parquet::arrow::ParquetRecordBatchStreamBuilder; + + let runtime = tokio::runtime::Builder::new_current_thread() + .build() + .unwrap(); + runtime.block_on(async { + let input = std::io::Cursor::new(file.clone()); + let builder = ParquetRecordBatchStreamBuilder::new_with_options(input, options()).await?; + configure(builder, trigger).build()?.try_collect().await + }) +} + +fn read_push_decoder(file: &Bytes, trigger: Trigger) -> Result> { + let metadata = ArrowReaderMetadata::load(file, options())?; + let builder = ParquetPushDecoderBuilder::new_with_metadata(metadata); + let mut decoder = configure(builder, trigger).build()?; + let mut batches = vec![]; + loop { + match decoder.try_decode()? { + DecodeResult::NeedsData(ranges) => { + let data = ranges + .iter() + .map(|r| file.slice(r.start as usize..r.end as usize)) + .collect(); + decoder.push_ranges(ranges, data)?; + } + DecodeResult::Data(batch) => batches.push(batch), + DecodeResult::Finished => return Ok(batches), + } + } +} + +/// Reads rows 150..250 of a file where only the column chunks for which +/// `has_index(row_group, column)` is `true` have an offset index +fn check(has_index: impl Fn(usize, usize) -> bool) { + let file = test_file(&has_index); + let metadata = ArrowReaderMetadata::load(&file, options()).unwrap(); + let page_index = metadata.metadata().page_index().unwrap(); + for row_group in 0..2 { + for column in 0..3 { + let actual = page_index.offset_index(row_group, column).is_some(); + assert_eq!(actual, has_index(row_group, column)); + } + } + + let expected = test_batch().slice(150, 100); + for trigger in [Trigger::RowSelection, Trigger::RowFilter] { + let results = [ + ("sync", read_sync(&file, trigger)), + ("push decoder", read_push_decoder(&file, trigger)), + #[cfg(feature = "async")] + ("async", read_async(&file, trigger)), + ]; + for (reader, result) in results { + let batches = result.unwrap_or_else(|e| panic!("{reader} reader, {trigger:?}: {e}")); + let actual = concat_batches(&expected.schema(), &batches).unwrap(); + assert_eq!(actual, expected, "{reader} reader, {trigger:?}"); + } + } +}