-
Notifications
You must be signed in to change notification settings - Fork 3.1k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Use bloom filter for evaluating dynamic filters on strings
Benchmark (filterSize) (inputDataSet) (inputNullChance) (nonNullsSelectivity) (nullsAllowed) Mode Cnt Before Score After Score Units BenchmarkDynamicPageFilter.filterPages 100 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 145.858 ± 4.541 590.506 ± 28.510 ops/s BenchmarkDynamicPageFilter.filterPages 1000 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 136.995 ± 2.395 596.036 ± 22.694 ops/s BenchmarkDynamicPageFilter.filterPages 10000 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 136.990 ± 5.284 594.118 ± 15.764 ops/s BenchmarkDynamicPageFilter.filterPages 100000 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 114.591 ± 7.307 587.445 ± 9.818 ops/s BenchmarkDynamicPageFilter.filterPages 1000000 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 43.234 ± 1.621 578.800 ± 15.694 ops/s BenchmarkDynamicPageFilter.filterPages 5000000 VARCHAR_RANDOM 0.05 0.2 false thrpt 20 40.018 ± 2.245 464.153 ± 20.914 ops/s
- Loading branch information
1 parent
a99d96e
commit 137c6e9
Showing
5 changed files
with
315 additions
and
27 deletions.
There are no files selected for viewing
192 changes: 192 additions & 0 deletions
192
core/trino-main/src/main/java/io/trino/sql/gen/columnar/BloomFilter.java
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,192 @@ | ||
| /* | ||
| * Licensed 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. | ||
| */ | ||
| package io.trino.sql.gen.columnar; | ||
|
|
||
| import com.google.common.annotations.VisibleForTesting; | ||
| import com.google.common.collect.ImmutableList; | ||
| import io.airlift.slice.Slice; | ||
| import io.airlift.slice.XxHash64; | ||
| import io.trino.operator.project.InputChannels; | ||
| import io.trino.spi.Page; | ||
| import io.trino.spi.block.VariableWidthBlock; | ||
| import io.trino.spi.connector.ConnectorSession; | ||
| import io.trino.spi.predicate.Domain; | ||
| import io.trino.spi.type.CharType; | ||
| import io.trino.spi.type.Type; | ||
| import io.trino.spi.type.VarbinaryType; | ||
| import io.trino.spi.type.VarcharType; | ||
|
|
||
| import java.util.List; | ||
| import java.util.function.Supplier; | ||
|
|
||
| public class BloomFilter | ||
| { | ||
| private BloomFilter() {} | ||
|
|
||
| public static boolean canUseBloomFilter(Domain domain) | ||
| { | ||
| Type type = domain.getType(); | ||
| if (type.getJavaType() != Slice.class) { | ||
| return false; | ||
| } | ||
| if (!(type instanceof VarcharType || type instanceof CharType || type instanceof VarbinaryType)) { | ||
| return false; | ||
| } | ||
| return !domain.isNone() && !domain.isAll() && domain.isNullableDiscreteSet(); | ||
| } | ||
|
|
||
| public static Supplier<FilterEvaluator> createBloomFilterEvaluator(Domain discreteDomain, int inputChannel) | ||
| { | ||
| return () -> new ColumnarFilterEvaluator( | ||
| new DictionaryAwareColumnarFilter( | ||
| new BloomColumnarFilter(discreteDomain.getNullableDiscreteSet(), inputChannel))); | ||
| } | ||
|
|
||
| private static final class BloomColumnarFilter | ||
| implements ColumnarFilter | ||
| { | ||
| private final SliceBloomFilter bloomFilter; | ||
| private final boolean isNullAllowed; | ||
| private final InputChannels inputChannels; | ||
|
|
||
| public BloomColumnarFilter(Domain.DiscreteSet discreteSet, int inputChannel) | ||
| { | ||
| this.isNullAllowed = discreteSet.containsNull(); | ||
| this.bloomFilter = new SliceBloomFilter((List<Slice>) (List<?>) discreteSet.getNonNullValues()); | ||
| this.inputChannels = new InputChannels(ImmutableList.of(inputChannel), ImmutableList.of(inputChannel)); | ||
| } | ||
|
|
||
| @Override | ||
| public int filterPositionsRange(ConnectorSession session, int[] outputPositions, int offset, int size, Page page) | ||
| { | ||
| VariableWidthBlock block = (VariableWidthBlock) page.getBlock(0); | ||
| int selectedPositionsCount = 0; | ||
| if (block.mayHaveNull()) { | ||
| for (int position = offset; position < offset + size; position++) { | ||
| boolean result = block.isNull(position) ? isNullAllowed : bloomFilter.test(block, position); | ||
| outputPositions[selectedPositionsCount] = position; | ||
| selectedPositionsCount += result ? 1 : 0; | ||
| } | ||
| return selectedPositionsCount; | ||
| } | ||
|
|
||
| for (int position = offset; position < offset + size; position++) { | ||
| outputPositions[selectedPositionsCount] = position; | ||
| selectedPositionsCount += bloomFilter.test(block, position) ? 1 : 0; | ||
| } | ||
| return selectedPositionsCount; | ||
| } | ||
|
|
||
| @Override | ||
| public int filterPositionsList(ConnectorSession session, int[] outputPositions, int[] activePositions, int offset, int size, Page page) | ||
| { | ||
| VariableWidthBlock block = (VariableWidthBlock) page.getBlock(0); | ||
| int selectedPositionsCount = 0; | ||
| if (block.mayHaveNull()) { | ||
| for (int index = offset; index < offset + size; index++) { | ||
| int position = activePositions[index]; | ||
| boolean result = block.isNull(position) ? isNullAllowed : bloomFilter.test(block, position); | ||
| outputPositions[selectedPositionsCount] = position; | ||
| selectedPositionsCount += result ? 1 : 0; | ||
| } | ||
| return selectedPositionsCount; | ||
| } | ||
|
|
||
| for (int index = offset; index < offset + size; index++) { | ||
| int position = activePositions[index]; | ||
| outputPositions[selectedPositionsCount] = position; | ||
| selectedPositionsCount += bloomFilter.test(block, position) ? 1 : 0; | ||
| } | ||
| return selectedPositionsCount; | ||
| } | ||
|
|
||
| @Override | ||
| public InputChannels getInputChannels() | ||
| { | ||
| return inputChannels; | ||
| } | ||
| } | ||
|
|
||
| public static final class SliceBloomFilter | ||
| { | ||
| private final long[] bloom; | ||
| private final int bloomSizeMask; | ||
|
|
||
| /** | ||
| * A Bloom filter for a set of Slice values. | ||
| * This is approx 2X faster than the Bloom filter implementations in ORC and parquet because | ||
| * it uses single hash function and uses that to set 3 bits within a 64 bit word. | ||
| * The memory footprint is up to (4 * values.size()) bytes, which is much smaller than maintaining a hash set of strings. | ||
| * | ||
| * @param values List of values used for filtering | ||
| */ | ||
| public SliceBloomFilter(List<Slice> values) | ||
| { | ||
| int bloomSize = getBloomFilterSize(values.size()); | ||
| bloom = new long[bloomSize]; | ||
| bloomSizeMask = bloomSize - 1; | ||
| for (Slice value : values) { | ||
| long hashCode = XxHash64.hash(value); | ||
| // Set 3 bits in a 64 bit word | ||
| bloom[bloomIndex(hashCode)] |= bloomMask(hashCode); | ||
| } | ||
| } | ||
|
|
||
| private static int getBloomFilterSize(int valuesCount) | ||
| { | ||
| // Linear hash table size is the highest power of two less than or equal to number of values * 4. This means that the | ||
| // table is under half full, e.g. 127 elements gets 256 slots. | ||
| int hashTableSize = Integer.highestOneBit(valuesCount * 4); | ||
| // We will allocate 8 bits in the bloom filter for every slot in a comparable hash table. | ||
| // The bloomSize is a count of longs, hence / 8. | ||
| return Math.max(1, hashTableSize / 8); | ||
| } | ||
|
|
||
| public boolean test(VariableWidthBlock block, int position) | ||
| { | ||
| return contains(block.getRawSlice(), block.getRawSliceOffset(position), block.getSliceLength(position)); | ||
| } | ||
|
|
||
| @VisibleForTesting | ||
| public boolean contains(Slice data) | ||
| { | ||
| return contains(data, 0, data.length()); | ||
| } | ||
|
|
||
| private boolean contains(Slice data, int offset, int length) | ||
| { | ||
| long hashCode = XxHash64.hash(data, offset, length); | ||
| long mask = bloomMask(hashCode); | ||
| return mask == (bloom[bloomIndex(hashCode)] & mask); | ||
| } | ||
|
|
||
| private int bloomIndex(long hashCode) | ||
| { | ||
| // Lower 21 bits are not used by bloomMask | ||
| // These are enough for the maximum size array that will be used here | ||
| return (int) (hashCode & bloomSizeMask); | ||
| } | ||
|
|
||
| private static long bloomMask(long hashCode) | ||
| { | ||
| // returned mask sets 3 bits based on portions of given hash | ||
| // Extract 38th to 43rd bits | ||
| return (1L << ((hashCode >> 21) & 63)) | ||
| // Extract 32nd to 37th bits | ||
| | (1L << ((hashCode >> 27) & 63)) | ||
| // Extract 26th to 31st bits | ||
| | (1L << ((hashCode >> 33) & 63)); | ||
| } | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.