Repository navigation
Expand file tree
/
Copy pathrow_number.rs
More file actions
359 lines (311 loc) · 13.2 KB
/
Copy pathrow_number.rs
File metadata and controls
359 lines (311 loc) · 13.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: See the NOTICE file distributed with this work for additional information regarding copyright ownership
// SPDX-FileContributor: Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements
use crate::arrow::array_reader::ArrayReader;
use crate::errors::{ParquetError, Result};
use crate::file::metadata::{ParquetMetaData, RowGroupMetaData};
use arrow_array::{ArrayRef, Int64Array};
use arrow_schema::DataType;
use std::any::Any;
use std::collections::{HashMap, VecDeque};
use std::ops::Range;
use std::sync::Arc;
/// Tracks row numbers within a Parquet file and emits them as an `Int64Array`.
pub(crate) struct RowNumberReader {
/// Pre-computed row ranges that are not read yet.
///
/// This reader only keeps track of the ranges of row numbers for each row group. The range is
/// not materialized into a full array until it's needed.
remaining_row_ranges: VecDeque<Range<i64>>,
/// Row ranges read but not emitted.
///
/// These are either full or partial (split) row ranges taken from `remaining_row_ranges`.
buffered_row_ranges: Vec<Range<i64>>,
}
impl RowNumberReader {
pub(crate) fn try_new<'a>(
parquet_metadata: &'a ParquetMetaData,
row_groups: impl Iterator<Item = &'a RowGroupMetaData>,
) -> Result<Self> {
// Pass 1: Build a map from ordinal to first_row_index
// This is O(M) where M is the total number of row groups in the file
let mut ordinal_to_offset: HashMap<i32, i64> = HashMap::new();
let mut first_row_index: i64 = 0;
let mut missing_ordinals: usize = 0;
for rg in parquet_metadata.row_groups() {
if let Some(ordinal) = rg.ordinal() {
ordinal_to_offset.insert(ordinal, first_row_index);
} else {
missing_ordinals += 1;
}
first_row_index += rg.num_rows();
}
// Mixed ordinals: refuse to compute row numbers for the whole file,
// even if every *selected* row group carries an ordinal. Otherwise the
// same file would yield row numbers or an error depending on which row
// groups a query's pruning happens to select.
if missing_ordinals > 0 && !ordinal_to_offset.is_empty() {
return Err(ParquetError::General(format!(
"Cannot compute row numbers: file has inconsistent row-group \
ordinals ({} of {} row groups are missing the ordinal field)",
missing_ordinals,
parquet_metadata.num_row_groups(),
)));
}
// Pass 2: Build ranges in the order specified by the row_groups iterator
// This is O(N) where N is the number of selected row groups
// This preserves the user's requested order instead of sorting by ordinal
let ranges: VecDeque<_> = row_groups
.map(|rg| {
let ordinal = rg.ordinal().ok_or_else(|| {
ParquetError::General(
"Row group missing ordinal field, required to compute row numbers"
.to_string(),
)
})?;
let offset = ordinal_to_offset.get(&ordinal).ok_or_else(|| {
ParquetError::General(format!(
"Row group with ordinal {ordinal} not found in metadata"
))
})?;
Ok(*offset..*offset + rg.num_rows())
})
.collect::<Result<_>>()?;
Ok(Self {
buffered_row_ranges: Vec::new(),
remaining_row_ranges: ranges,
})
}
/// Take up to `count` rows from the first range, splitting it if needed.
///
/// Returns `None` if no ranges remain.
fn take_range(&mut self, count: usize) -> Option<Range<i64>> {
let first = self.remaining_row_ranges.front_mut()?;
if (first.end - first.start) <= count as i64 {
// take out the full range
self.remaining_row_ranges.pop_front()
} else {
// first range has more rows than we need.
// so we split the range and put the remaining back.
let split = first.start + count as i64;
let taken = first.start..split;
first.start = split;
Some(taken)
}
}
}
impl ArrayReader for RowNumberReader {
fn read_records(&mut self, batch_size: usize) -> Result<usize> {
let mut remaining = batch_size;
while remaining > 0 {
let Some(range) = self.take_range(remaining) else {
break;
};
remaining -= (range.end - range.start) as usize;
self.buffered_row_ranges.push(range);
}
Ok(batch_size - remaining)
}
fn skip_records(&mut self, num_records: usize) -> Result<usize> {
let mut remaining = num_records;
while remaining > 0 {
let Some(range) = self.take_range(remaining) else {
break;
};
remaining -= (range.end - range.start) as usize;
}
Ok(num_records - remaining)
}
fn as_any(&self) -> &dyn Any {
self
}
fn get_data_type(&self) -> &DataType {
&DataType::Int64
}
fn consume_batch(&mut self) -> Result<ArrayRef> {
let total_rows: i64 = self
.buffered_row_ranges
.iter()
.map(|range| range.end - range.start)
.sum();
let mut result = Vec::with_capacity(total_rows as usize);
for range in self.buffered_row_ranges.drain(..) {
result.extend(range);
}
Ok(Arc::new(Int64Array::from(result)))
}
fn get_def_levels(&self) -> Option<&[i16]> {
None
}
fn get_rep_levels(&self) -> Option<&[i16]> {
None
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::basic::Type as PhysicalType;
use crate::file::metadata::{
ColumnChunkMetaData, FileMetaData, ParquetMetaData, RowGroupMetaData,
};
use crate::schema::types::{SchemaDescriptor, Type as SchemaType};
use std::sync::Arc;
fn create_test_schema() -> Arc<SchemaDescriptor> {
let schema = SchemaType::group_type_builder("schema")
.with_fields(vec![Arc::new(
SchemaType::primitive_type_builder("test_col", PhysicalType::INT32)
.build()
.unwrap(),
)])
.build()
.unwrap();
Arc::new(SchemaDescriptor::new(Arc::new(schema)))
}
fn create_test_parquet_metadata(row_groups: Vec<(i32, i64)>) -> ParquetMetaData {
create_test_parquet_metadata_opt(
row_groups
.into_iter()
.map(|(ordinal, num_rows)| (Some(ordinal), num_rows))
.collect(),
)
}
fn create_test_parquet_metadata_opt(row_groups: Vec<(Option<i32>, i64)>) -> ParquetMetaData {
let schema_descr = create_test_schema();
let mut row_group_metas = vec![];
for (ordinal, num_rows) in row_groups {
let columns: Vec<_> = schema_descr
.columns()
.iter()
.map(|col| ColumnChunkMetaData::builder(col.clone()).build().unwrap())
.collect();
let mut builder = RowGroupMetaData::builder(schema_descr.clone())
.set_num_rows(num_rows)
.set_total_byte_size(100)
.set_column_metadata(columns);
if let Some(ordinal) = ordinal {
builder = builder.set_ordinal(ordinal);
}
row_group_metas.push(builder.build().unwrap());
}
let total_rows: i64 = row_group_metas.iter().map(|rg| rg.num_rows()).sum();
let file_metadata = FileMetaData::new(
1, // version
total_rows, // num_rows
None, // created_by
None, // key_value_metadata
schema_descr, // schema_descr
None, // column_orders
);
ParquetMetaData::new(file_metadata, row_group_metas)
}
fn consume_row_numbers(reader: &mut RowNumberReader) -> Vec<i64> {
let array = reader.consume_batch().unwrap();
array
.as_any()
.downcast_ref::<Int64Array>()
.unwrap()
.values()
.to_vec()
}
fn reader_for_all(metadata: &ParquetMetaData) -> RowNumberReader {
let all_rgs: Vec<_> = metadata.row_groups().iter().collect();
RowNumberReader::try_new(metadata, all_rgs.into_iter()).unwrap()
}
#[test]
fn test_row_number_reader_reverse_order() {
// Create metadata with 3 row groups, each with 2 rows
let metadata = create_test_parquet_metadata(vec![
(0, 2), // Row group 0: ordinal=0, rows 0-1
(1, 2), // Row group 1: ordinal=1, rows 2-3
(2, 2), // Row group 2: ordinal=2, rows 4-5
]);
// Select only row groups with ordinals 2 and 0 (in that order)
// This means we want row group 2 first, then row group 0, skipping row group 1
let selected_row_groups: Vec<_> = vec![
&metadata.row_groups()[2], // ordinal 2
&metadata.row_groups()[0], // ordinal 0
];
let mut reader =
RowNumberReader::try_new(&metadata, selected_row_groups.into_iter()).unwrap();
// Read all row numbers
let num_read = reader.read_records(6).unwrap();
assert_eq!(num_read, 4); // Should read 4 rows total (2 from each selected group)
// Expected: row group 2 first (rows 4-5), then row group 0 (rows 0-1)
assert_eq!(consume_row_numbers(&mut reader), vec![4, 5, 0, 1]);
}
#[test]
fn test_range_splitting_across_batches() {
// One row group with 10 rows
let metadata = create_test_parquet_metadata(vec![(0, 10)]);
let mut reader = reader_for_all(&metadata);
assert_eq!(reader.read_records(3).unwrap(), 3);
assert_eq!(consume_row_numbers(&mut reader), vec![0, 1, 2]);
assert_eq!(reader.read_records(3).unwrap(), 3);
assert_eq!(consume_row_numbers(&mut reader), vec![3, 4, 5]);
assert_eq!(reader.read_records(3).unwrap(), 3);
assert_eq!(consume_row_numbers(&mut reader), vec![6, 7, 8]);
// Only 1 row left, requesting 3
assert_eq!(reader.read_records(3).unwrap(), 1);
assert_eq!(consume_row_numbers(&mut reader), vec![9]);
}
#[test]
fn test_interleaved_skip_and_read() {
// Row group 0: rows 0..5
// Row group 1: rows 5..10
let metadata = create_test_parquet_metadata(vec![(0, 5), (1, 5)]);
let mut reader = reader_for_all(&metadata);
assert_eq!(reader.skip_records(2).unwrap(), 2); // skip [0,1]
assert_eq!(reader.read_records(2).unwrap(), 2); // read [2,3]
assert_eq!(reader.skip_records(3).unwrap(), 3); // skip [4] then [5,6]
assert_eq!(reader.read_records(3).unwrap(), 3); // read [7,8,9]
assert_eq!(consume_row_numbers(&mut reader), vec![2, 3, 7, 8, 9]);
}
#[test]
fn test_skip_then_read() {
// One row group with 10 rows
let metadata = create_test_parquet_metadata(vec![(0, 10)]);
let mut reader = reader_for_all(&metadata);
assert_eq!(reader.skip_records(3).unwrap(), 3);
assert_eq!(reader.read_records(4).unwrap(), 4);
assert_eq!(consume_row_numbers(&mut reader), vec![3, 4, 5, 6]);
}
/// A file with *mixed* row-group ordinals must never produce row numbers —
/// even when every selected row group carries an ordinal. Otherwise the
/// same file would succeed or fail depending on which row groups a query's
/// pruning selects.
#[test]
fn test_mixed_ordinals_always_error() {
let metadata = create_test_parquet_metadata_opt(vec![
(Some(0), 2), // has ordinal
(None, 2), // missing
(Some(2), 2), // has ordinal
]);
// Select only row groups WITH ordinals — must still fail.
let selected = vec![&metadata.row_groups()[0], &metadata.row_groups()[2]];
let Err(err) = RowNumberReader::try_new(&metadata, selected.into_iter()) else {
panic!("mixed ordinals with ordinal-only selection must fail");
};
assert!(
err.to_string().contains("inconsistent row-group ordinals"),
"unexpected error: {err}"
);
// Selecting a row group without an ordinal fails too.
let selected = vec![&metadata.row_groups()[1]];
assert!(RowNumberReader::try_new(&metadata, selected.into_iter()).is_err());
}
/// A file where *no* row group carries an ordinal fails for any selection
/// (fresh decode sequential-fills, so this only occurs for
/// programmatically-built metadata).
#[test]
fn test_no_ordinals_error() {
let metadata = create_test_parquet_metadata_opt(vec![(None, 2), (None, 2)]);
let selected = vec![&metadata.row_groups()[0]];
let Err(err) = RowNumberReader::try_new(&metadata, selected.into_iter()) else {
panic!("row numbers without any ordinals must fail");
};
assert!(
err.to_string().contains("missing ordinal field"),
"unexpected error: {err}"
);
}
}