-
Notifications
You must be signed in to change notification settings - Fork 660
/
mod.rs
295 lines (266 loc) · 9.73 KB
/
mod.rs
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
// 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.
//! Bloom filter implementation specific to Parquet, as described
//! in the [spec](https://github.com/apache/parquet-format/blob/master/BloomFilter.md)
use crate::data_type::AsBytes;
use crate::errors::ParquetError;
use crate::file::metadata::ColumnChunkMetaData;
use crate::file::reader::ChunkReader;
use crate::format::{
BloomFilterAlgorithm, BloomFilterCompression, BloomFilterHash, BloomFilterHeader,
};
use bytes::{Buf, Bytes};
use std::hash::Hasher;
use std::sync::Arc;
use thrift::protocol::{TCompactInputProtocol, TSerializable};
use twox_hash::XxHash64;
/// Salt as defined in the [spec](https://github.com/apache/parquet-format/blob/master/BloomFilter.md#technical-approach)
const SALT: [u32; 8] = [
0x47b6137b_u32,
0x44974d91_u32,
0x8824ad5b_u32,
0xa2b7289d_u32,
0x705495c7_u32,
0x2df1424b_u32,
0x9efc4947_u32,
0x5c6bfb31_u32,
];
/// Each block is 256 bits, broken up into eight contiguous "words", each consisting of 32 bits.
/// Each word is thought of as an array of bits; each bit is either "set" or "not set".
type Block = [u32; 8];
/// takes as its argument a single unsigned 32-bit integer and returns a block in which each
/// word has exactly one bit set.
fn mask(x: u32) -> Block {
let mut result = [0_u32; 8];
for i in 0..8 {
// wrapping instead of checking for overflow
let y = x.wrapping_mul(SALT[i]);
let y = y >> 27;
result[i] = 1 << y;
}
result
}
/// setting every bit in the block that was also set in the result from mask
fn block_insert(block: &mut Block, hash: u32) {
let mask = mask(hash);
for i in 0..8 {
block[i] |= mask[i];
}
}
/// returns true when every bit that is set in the result of mask is also set in the block.
fn block_check(block: &Block, hash: u32) -> bool {
let mask = mask(hash);
for i in 0..8 {
if block[i] & mask[i] == 0 {
return false;
}
}
true
}
/// A split block Bloom filter
pub struct Sbbf(Vec<Block>);
const SBBF_HEADER_SIZE_ESTIMATE: usize = 20;
/// given an initial offset, and a [ChunkReader], try to read out a bloom filter header and return
/// both the header and the offset after it (for bitset).
fn chunk_read_bloom_filter_header_and_offset<R: ChunkReader>(
offset: u64,
reader: Arc<R>,
) -> Result<(BloomFilterHeader, u64), ParquetError> {
let buffer = reader.get_bytes(offset as u64, SBBF_HEADER_SIZE_ESTIMATE)?;
let (header, length) = read_bloom_filter_header_and_length(buffer)?;
Ok((header, offset + length))
}
/// given a [Bytes] buffer, try to read out a bloom filter header and return both the header and
/// length of the header.
#[inline]
fn read_bloom_filter_header_and_length(
buffer: Bytes,
) -> Result<(BloomFilterHeader, u64), ParquetError> {
let total_length = buffer.len();
let mut buf_reader = buffer.reader();
let mut prot = TCompactInputProtocol::new(&mut buf_reader);
let header = BloomFilterHeader::read_from_in_protocol(&mut prot).map_err(|e| {
ParquetError::General(format!("Could not read bloom filter header: {}", e))
})?;
Ok((
header,
(total_length - buf_reader.into_inner().remaining()) as u64,
))
}
impl Sbbf {
fn new(bitset: &[u8]) -> Self {
let data = bitset
.chunks_exact(4 * 8)
.map(|chunk| {
let mut block = [0_u32; 8];
for (i, word) in chunk.chunks_exact(4).enumerate() {
block[i] = u32::from_le_bytes(word.try_into().unwrap());
}
block
})
.collect::<Vec<Block>>();
Self(data)
}
pub fn read_from_column_chunk<R: ChunkReader>(
column_metadata: &ColumnChunkMetaData,
reader: Arc<R>,
) -> Result<Option<Self>, ParquetError> {
let offset: u64 = if let Some(offset) = column_metadata.bloom_filter_offset() {
offset.try_into().map_err(|_| {
ParquetError::General("Bloom filter offset is invalid".to_string())
})?
} else {
return Ok(None);
};
let (header, bitset_offset) =
chunk_read_bloom_filter_header_and_offset(offset, reader.clone())?;
match header.algorithm {
BloomFilterAlgorithm::BLOCK(_) => {
// this match exists to future proof the singleton algorithm enum
}
}
match header.compression {
BloomFilterCompression::UNCOMPRESSED(_) => {
// this match exists to future proof the singleton compression enum
}
}
match header.hash {
BloomFilterHash::XXHASH(_) => {
// this match exists to future proof the singleton hash enum
}
}
// length in bytes
let length: usize = header.num_bytes.try_into().map_err(|_| {
ParquetError::General("Bloom filter length is invalid".to_string())
})?;
let bitset = reader.get_bytes(bitset_offset, length)?;
Ok(Some(Self::new(&bitset)))
}
#[inline]
fn hash_to_block_index(&self, hash: u64) -> usize {
// unchecked_mul is unstable, but in reality this is safe, we'd just use saturating mul
// but it will not saturate
(((hash >> 32).saturating_mul(self.0.len() as u64)) >> 32) as usize
}
/// Insert an [AsBytes] value into the filter
pub fn insert<T: AsBytes>(&mut self, value: T) {
self.insert_hash(hash_as_bytes(value));
}
/// Insert a hash into the filter
fn insert_hash(&mut self, hash: u64) {
let block_index = self.hash_to_block_index(hash);
let block = &mut self.0[block_index];
block_insert(block, hash as u32);
}
/// Check if an [AsBytes] value is probably present or definitely absent in the filter
pub fn check<T: AsBytes>(&self, value: T) -> bool {
self.check_hash(hash_as_bytes(value))
}
/// Check if a hash is in the filter. May return
/// true for values that was never inserted ("false positive")
/// but will always return false if a hash has not been inserted.
fn check_hash(&self, hash: u64) -> bool {
let block_index = self.hash_to_block_index(hash);
let block = &self.0[block_index];
block_check(block, hash as u32)
}
}
// per spec we use xxHash with seed=0
const SEED: u64 = 0;
#[inline]
fn hash_as_bytes<A: AsBytes>(value: A) -> u64 {
let mut hasher = XxHash64::with_seed(SEED);
hasher.write(value.as_bytes());
hasher.finish()
}
#[cfg(test)]
mod tests {
use super::*;
use crate::format::{
BloomFilterAlgorithm, BloomFilterCompression, SplitBlockAlgorithm, Uncompressed,
XxHash,
};
#[test]
fn test_hash_bytes() {
assert_eq!(hash_as_bytes(""), 17241709254077376921);
}
#[test]
fn test_mask_set_quick_check() {
for i in 0..1_000_000 {
let result = mask(i);
assert!(result.iter().all(|&x| x.count_ones() == 1));
}
}
#[test]
fn test_block_insert_and_check() {
for i in 0..1_000_000 {
let mut block = [0_u32; 8];
block_insert(&mut block, i);
assert!(block_check(&block, i));
}
}
#[test]
fn test_sbbf_insert_and_check() {
let mut sbbf = Sbbf(vec![[0_u32; 8]; 1_000]);
for i in 0..1_000_000 {
sbbf.insert(i);
assert!(sbbf.check(i));
}
}
#[test]
fn test_with_fixture() {
// bloom filter produced by parquet-mr/spark for a column of i64 f"a{i}" for i in 0..10
let bitset: &[u8] = &[
200, 1, 80, 20, 64, 68, 8, 109, 6, 37, 4, 67, 144, 80, 96, 32, 8, 132, 43,
33, 0, 5, 99, 65, 2, 0, 224, 44, 64, 78, 96, 4,
];
let sbbf = Sbbf::new(bitset);
for a in 0..10i64 {
let value = format!("a{}", a);
assert!(sbbf.check(value.as_str()));
}
}
/// test the assumption that bloom filter header size should not exceed SBBF_HEADER_SIZE_ESTIMATE
/// essentially we are testing that the struct is packed with 4 i32 fields, each can be 1-5 bytes
/// so altogether it'll be 20 bytes at most.
#[test]
fn test_bloom_filter_header_size_assumption() {
let buffer: &[u8; 16] =
&[21, 64, 28, 28, 0, 0, 28, 28, 0, 0, 28, 28, 0, 0, 0, 99];
let (
BloomFilterHeader {
algorithm,
compression,
hash,
num_bytes,
},
read_length,
) = read_bloom_filter_header_and_length(Bytes::copy_from_slice(buffer)).unwrap();
assert_eq!(read_length, 15);
assert_eq!(
algorithm,
BloomFilterAlgorithm::BLOCK(SplitBlockAlgorithm {})
);
assert_eq!(
compression,
BloomFilterCompression::UNCOMPRESSED(Uncompressed {})
);
assert_eq!(hash, BloomFilterHash::XXHASH(XxHash {}));
assert_eq!(num_bytes, 32_i32);
assert_eq!(20, SBBF_HEADER_SIZE_ESTIMATE);
}
}