mayya-sharipova commented on code in PR #12582: URL: https://github.com/apache/lucene/pull/12582#discussion_r1365887225
########## lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99HnswVectorsReader.java: ########## @@ -0,0 +1,628 @@ +/* + * 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. + */ + +package org.apache.lucene.codecs.lucene99; + +import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS; + +import java.io.IOException; +import java.util.Arrays; +import java.util.HashMap; +import java.util.Map; +import org.apache.lucene.codecs.CodecUtil; +import org.apache.lucene.codecs.HnswGraphProvider; +import org.apache.lucene.codecs.KnnVectorsReader; +import org.apache.lucene.codecs.lucene95.OffHeapByteVectorValues; +import org.apache.lucene.codecs.lucene95.OffHeapFloatVectorValues; +import org.apache.lucene.codecs.lucene95.OrdToDocDISIReaderConfiguration; +import org.apache.lucene.index.ByteVectorValues; +import org.apache.lucene.index.CorruptIndexException; +import org.apache.lucene.index.FieldInfo; +import org.apache.lucene.index.FieldInfos; +import org.apache.lucene.index.FloatVectorValues; +import org.apache.lucene.index.IndexFileNames; +import org.apache.lucene.index.SegmentReadState; +import org.apache.lucene.index.VectorEncoding; +import org.apache.lucene.index.VectorSimilarityFunction; +import org.apache.lucene.search.KnnCollector; +import org.apache.lucene.store.ChecksumIndexInput; +import org.apache.lucene.store.DataInput; +import org.apache.lucene.store.IndexInput; +import org.apache.lucene.store.RandomAccessInput; +import org.apache.lucene.util.Accountable; +import org.apache.lucene.util.Bits; +import org.apache.lucene.util.IOUtils; +import org.apache.lucene.util.RamUsageEstimator; +import org.apache.lucene.util.ScalarQuantizer; +import org.apache.lucene.util.hnsw.HnswGraph; +import org.apache.lucene.util.hnsw.HnswGraphSearcher; +import org.apache.lucene.util.hnsw.OrdinalTranslatedKnnCollector; +import org.apache.lucene.util.hnsw.RandomVectorScorer; +import org.apache.lucene.util.packed.DirectMonotonicReader; + +/** + * Reads vectors from the index segments along with index data structures supporting KNN search. + * + * @lucene.experimental + */ +public final class Lucene99HnswVectorsReader extends KnnVectorsReader + implements QuantizedVectorsReader, HnswGraphProvider { + + private static final long SHALLOW_SIZE = + RamUsageEstimator.shallowSizeOfInstance(Lucene99HnswVectorsFormat.class); + + private final FieldInfos fieldInfos; + private final Map<String, FieldEntry> fields = new HashMap<>(); + private final IndexInput vectorData; + private final IndexInput vectorIndex; + private final IndexInput quantizedVectorData; + private final Lucene99ScalarQuantizedVectorsReader quantizedVectorsReader; + + Lucene99HnswVectorsReader(SegmentReadState state) throws IOException { + this.fieldInfos = state.fieldInfos; + int versionMeta = readMetadata(state); + boolean success = false; + try { + vectorData = + openDataInput( + state, + versionMeta, + Lucene99HnswVectorsFormat.VECTOR_DATA_EXTENSION, + Lucene99HnswVectorsFormat.VECTOR_DATA_CODEC_NAME); + vectorIndex = + openDataInput( + state, + versionMeta, + Lucene99HnswVectorsFormat.VECTOR_INDEX_EXTENSION, + Lucene99HnswVectorsFormat.VECTOR_INDEX_CODEC_NAME); + if (fields.values().stream().anyMatch(FieldEntry::hasQuantizedVectors)) { + quantizedVectorData = + openDataInput( + state, + versionMeta, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_EXTENSION, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_CODEC_NAME); + quantizedVectorsReader = new Lucene99ScalarQuantizedVectorsReader(quantizedVectorData); + } else { + quantizedVectorData = null; + quantizedVectorsReader = null; + } + success = true; + } finally { + if (success == false) { + IOUtils.closeWhileHandlingException(this); + } + } + } + + private int readMetadata(SegmentReadState state) throws IOException { + String metaFileName = + IndexFileNames.segmentFileName( + state.segmentInfo.name, state.segmentSuffix, Lucene99HnswVectorsFormat.META_EXTENSION); + int versionMeta = -1; + try (ChecksumIndexInput meta = state.directory.openChecksumInput(metaFileName)) { + Throwable priorE = null; + try { + versionMeta = + CodecUtil.checkIndexHeader( + meta, + Lucene99HnswVectorsFormat.META_CODEC_NAME, + Lucene99HnswVectorsFormat.VERSION_START, + Lucene99HnswVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + readFields(meta, state.fieldInfos); + } catch (Throwable exception) { + priorE = exception; + } finally { + CodecUtil.checkFooter(meta, priorE); + } + } + return versionMeta; + } + + private static IndexInput openDataInput( + SegmentReadState state, int versionMeta, String fileExtension, String codecName) + throws IOException { + String fileName = + IndexFileNames.segmentFileName(state.segmentInfo.name, state.segmentSuffix, fileExtension); + IndexInput in = state.directory.openInput(fileName, state.context); + boolean success = false; + try { + int versionVectorData = + CodecUtil.checkIndexHeader( + in, + codecName, + Lucene99HnswVectorsFormat.VERSION_START, + Lucene99HnswVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + if (versionMeta != versionVectorData) { + throw new CorruptIndexException( + "Format versions mismatch: meta=" + + versionMeta + + ", " + + codecName + + "=" + + versionVectorData, + in); + } + CodecUtil.retrieveChecksum(in); + success = true; + return in; + } finally { + if (success == false) { + IOUtils.closeWhileHandlingException(in); + } + } + } + + private void readFields(ChecksumIndexInput meta, FieldInfos infos) throws IOException { + for (int fieldNumber = meta.readInt(); fieldNumber != -1; fieldNumber = meta.readInt()) { + FieldInfo info = infos.fieldInfo(fieldNumber); + if (info == null) { + throw new CorruptIndexException("Invalid field number: " + fieldNumber, meta); + } + FieldEntry fieldEntry = readField(meta); + validateFieldEntry(info, fieldEntry); + fields.put(info.name, fieldEntry); + } + } + + private void validateFieldEntry(FieldInfo info, FieldEntry fieldEntry) { + int dimension = info.getVectorDimension(); + if (dimension != fieldEntry.dimension) { + throw new IllegalStateException( + "Inconsistent vector dimension for field=\"" + + info.name + + "\"; " + + dimension + + " != " + + fieldEntry.dimension); + } + + int byteSize = + switch (info.getVectorEncoding()) { + case BYTE -> Byte.BYTES; + case FLOAT32 -> Float.BYTES; + }; + long vectorBytes = Math.multiplyExact((long) dimension, byteSize); + long numBytes = Math.multiplyExact(vectorBytes, fieldEntry.size); + if (numBytes != fieldEntry.vectorDataLength) { + throw new IllegalStateException( + "Vector data length " + + fieldEntry.vectorDataLength + + " not matching size=" + + fieldEntry.size + + " * dim=" + + dimension + + " * byteSize=" + + byteSize + + " = " + + numBytes); + } + if (fieldEntry.hasQuantizedVectors()) { + Lucene99ScalarQuantizedVectorsReader.validateFieldEntry( + info, fieldEntry.dimension, fieldEntry.size, fieldEntry.quantizedVectorDataLength); + } + } + + private VectorSimilarityFunction readSimilarityFunction(DataInput input) throws IOException { + int similarityFunctionId = input.readInt(); + if (similarityFunctionId < 0 + || similarityFunctionId >= VectorSimilarityFunction.values().length) { + throw new CorruptIndexException( + "Invalid similarity function id: " + similarityFunctionId, input); + } + return VectorSimilarityFunction.values()[similarityFunctionId]; + } + + private VectorEncoding readVectorEncoding(DataInput input) throws IOException { + int encodingId = input.readInt(); + if (encodingId < 0 || encodingId >= VectorEncoding.values().length) { + throw new CorruptIndexException("Invalid vector encoding id: " + encodingId, input); + } + return VectorEncoding.values()[encodingId]; + } + + private FieldEntry readField(IndexInput input) throws IOException { + VectorEncoding vectorEncoding = readVectorEncoding(input); + VectorSimilarityFunction similarityFunction = readSimilarityFunction(input); + return new FieldEntry(input, vectorEncoding, similarityFunction); + } + + @Override + public long ramBytesUsed() { + return Lucene99HnswVectorsReader.SHALLOW_SIZE + + RamUsageEstimator.sizeOfMap( + fields, RamUsageEstimator.shallowSizeOfInstance(FieldEntry.class)); + } + + @Override + public void checkIntegrity() throws IOException { + CodecUtil.checksumEntireFile(vectorData); + CodecUtil.checksumEntireFile(vectorIndex); + if (quantizedVectorsReader != null) { + quantizedVectorsReader.checkIntegrity(); + } + } + + @Override + public FloatVectorValues getFloatVectorValues(String field) throws IOException { + FieldEntry fieldEntry = fields.get(field); + if (fieldEntry.vectorEncoding != VectorEncoding.FLOAT32) { + throw new IllegalArgumentException( + "field=\"" + + field + + "\" is encoded as: " + + fieldEntry.vectorEncoding + + " expected: " + + VectorEncoding.FLOAT32); + } + return OffHeapFloatVectorValues.load( + fieldEntry.ordToDoc, + fieldEntry.vectorEncoding, + fieldEntry.dimension, + fieldEntry.vectorDataOffset, + fieldEntry.vectorDataLength, + vectorData); + } + + @Override + public ByteVectorValues getByteVectorValues(String field) throws IOException { + FieldEntry fieldEntry = fields.get(field); + if (fieldEntry.vectorEncoding != VectorEncoding.BYTE) { + throw new IllegalArgumentException( + "field=\"" + + field + + "\" is encoded as: " + + fieldEntry.vectorEncoding + + " expected: " + + VectorEncoding.FLOAT32); + } + return OffHeapByteVectorValues.load( + fieldEntry.ordToDoc, + fieldEntry.vectorEncoding, + fieldEntry.dimension, + fieldEntry.vectorDataOffset, + fieldEntry.vectorDataLength, + vectorData); + } + + @Override + public void search(String field, float[] target, KnnCollector knnCollector, Bits acceptDocs) + throws IOException { + FieldEntry fieldEntry = fields.get(field); + + if (fieldEntry.size() == 0 + || knnCollector.k() == 0 + || fieldEntry.vectorEncoding != VectorEncoding.FLOAT32) { + return; + } + if (quantizedVectorsReader != null) { Review Comment: Should we instead check if `if (fieldEntry.hasQuantizedVectors())`? This allow to use raw vector values for search if this particular field is not quantized. ########## lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99HnswVectorsReader.java: ########## @@ -0,0 +1,628 @@ +/* + * 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. + */ + +package org.apache.lucene.codecs.lucene99; + +import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS; + +import java.io.IOException; +import java.util.Arrays; +import java.util.HashMap; +import java.util.Map; +import org.apache.lucene.codecs.CodecUtil; +import org.apache.lucene.codecs.HnswGraphProvider; +import org.apache.lucene.codecs.KnnVectorsReader; +import org.apache.lucene.codecs.lucene95.OffHeapByteVectorValues; +import org.apache.lucene.codecs.lucene95.OffHeapFloatVectorValues; +import org.apache.lucene.codecs.lucene95.OrdToDocDISIReaderConfiguration; +import org.apache.lucene.index.ByteVectorValues; +import org.apache.lucene.index.CorruptIndexException; +import org.apache.lucene.index.FieldInfo; +import org.apache.lucene.index.FieldInfos; +import org.apache.lucene.index.FloatVectorValues; +import org.apache.lucene.index.IndexFileNames; +import org.apache.lucene.index.SegmentReadState; +import org.apache.lucene.index.VectorEncoding; +import org.apache.lucene.index.VectorSimilarityFunction; +import org.apache.lucene.search.KnnCollector; +import org.apache.lucene.store.ChecksumIndexInput; +import org.apache.lucene.store.DataInput; +import org.apache.lucene.store.IndexInput; +import org.apache.lucene.store.RandomAccessInput; +import org.apache.lucene.util.Accountable; +import org.apache.lucene.util.Bits; +import org.apache.lucene.util.IOUtils; +import org.apache.lucene.util.RamUsageEstimator; +import org.apache.lucene.util.ScalarQuantizer; +import org.apache.lucene.util.hnsw.HnswGraph; +import org.apache.lucene.util.hnsw.HnswGraphSearcher; +import org.apache.lucene.util.hnsw.OrdinalTranslatedKnnCollector; +import org.apache.lucene.util.hnsw.RandomVectorScorer; +import org.apache.lucene.util.packed.DirectMonotonicReader; + +/** + * Reads vectors from the index segments along with index data structures supporting KNN search. + * + * @lucene.experimental + */ +public final class Lucene99HnswVectorsReader extends KnnVectorsReader + implements QuantizedVectorsReader, HnswGraphProvider { + + private static final long SHALLOW_SIZE = + RamUsageEstimator.shallowSizeOfInstance(Lucene99HnswVectorsFormat.class); + + private final FieldInfos fieldInfos; + private final Map<String, FieldEntry> fields = new HashMap<>(); + private final IndexInput vectorData; + private final IndexInput vectorIndex; + private final IndexInput quantizedVectorData; + private final Lucene99ScalarQuantizedVectorsReader quantizedVectorsReader; + + Lucene99HnswVectorsReader(SegmentReadState state) throws IOException { + this.fieldInfos = state.fieldInfos; + int versionMeta = readMetadata(state); + boolean success = false; + try { + vectorData = + openDataInput( + state, + versionMeta, + Lucene99HnswVectorsFormat.VECTOR_DATA_EXTENSION, + Lucene99HnswVectorsFormat.VECTOR_DATA_CODEC_NAME); + vectorIndex = + openDataInput( + state, + versionMeta, + Lucene99HnswVectorsFormat.VECTOR_INDEX_EXTENSION, + Lucene99HnswVectorsFormat.VECTOR_INDEX_CODEC_NAME); + if (fields.values().stream().anyMatch(FieldEntry::hasQuantizedVectors)) { + quantizedVectorData = + openDataInput( + state, + versionMeta, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_EXTENSION, + Lucene99ScalarQuantizedVectorsFormat.QUANTIZED_VECTOR_DATA_CODEC_NAME); + quantizedVectorsReader = new Lucene99ScalarQuantizedVectorsReader(quantizedVectorData); + } else { + quantizedVectorData = null; + quantizedVectorsReader = null; + } + success = true; + } finally { + if (success == false) { + IOUtils.closeWhileHandlingException(this); + } + } + } + + private int readMetadata(SegmentReadState state) throws IOException { + String metaFileName = + IndexFileNames.segmentFileName( + state.segmentInfo.name, state.segmentSuffix, Lucene99HnswVectorsFormat.META_EXTENSION); + int versionMeta = -1; + try (ChecksumIndexInput meta = state.directory.openChecksumInput(metaFileName)) { + Throwable priorE = null; + try { + versionMeta = + CodecUtil.checkIndexHeader( + meta, + Lucene99HnswVectorsFormat.META_CODEC_NAME, + Lucene99HnswVectorsFormat.VERSION_START, + Lucene99HnswVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + readFields(meta, state.fieldInfos); + } catch (Throwable exception) { + priorE = exception; + } finally { + CodecUtil.checkFooter(meta, priorE); + } + } + return versionMeta; + } + + private static IndexInput openDataInput( + SegmentReadState state, int versionMeta, String fileExtension, String codecName) + throws IOException { + String fileName = + IndexFileNames.segmentFileName(state.segmentInfo.name, state.segmentSuffix, fileExtension); + IndexInput in = state.directory.openInput(fileName, state.context); + boolean success = false; + try { + int versionVectorData = + CodecUtil.checkIndexHeader( + in, + codecName, + Lucene99HnswVectorsFormat.VERSION_START, + Lucene99HnswVectorsFormat.VERSION_CURRENT, + state.segmentInfo.getId(), + state.segmentSuffix); + if (versionMeta != versionVectorData) { + throw new CorruptIndexException( + "Format versions mismatch: meta=" + + versionMeta + + ", " + + codecName + + "=" + + versionVectorData, + in); + } + CodecUtil.retrieveChecksum(in); + success = true; + return in; + } finally { + if (success == false) { + IOUtils.closeWhileHandlingException(in); + } + } + } + + private void readFields(ChecksumIndexInput meta, FieldInfos infos) throws IOException { + for (int fieldNumber = meta.readInt(); fieldNumber != -1; fieldNumber = meta.readInt()) { + FieldInfo info = infos.fieldInfo(fieldNumber); + if (info == null) { + throw new CorruptIndexException("Invalid field number: " + fieldNumber, meta); + } + FieldEntry fieldEntry = readField(meta); + validateFieldEntry(info, fieldEntry); + fields.put(info.name, fieldEntry); + } + } + + private void validateFieldEntry(FieldInfo info, FieldEntry fieldEntry) { + int dimension = info.getVectorDimension(); + if (dimension != fieldEntry.dimension) { + throw new IllegalStateException( + "Inconsistent vector dimension for field=\"" + + info.name + + "\"; " + + dimension + + " != " + + fieldEntry.dimension); + } + + int byteSize = + switch (info.getVectorEncoding()) { + case BYTE -> Byte.BYTES; + case FLOAT32 -> Float.BYTES; + }; + long vectorBytes = Math.multiplyExact((long) dimension, byteSize); + long numBytes = Math.multiplyExact(vectorBytes, fieldEntry.size); + if (numBytes != fieldEntry.vectorDataLength) { + throw new IllegalStateException( + "Vector data length " + + fieldEntry.vectorDataLength + + " not matching size=" + + fieldEntry.size + + " * dim=" + + dimension + + " * byteSize=" + + byteSize + + " = " + + numBytes); + } + if (fieldEntry.hasQuantizedVectors()) { + Lucene99ScalarQuantizedVectorsReader.validateFieldEntry( + info, fieldEntry.dimension, fieldEntry.size, fieldEntry.quantizedVectorDataLength); + } + } + + private VectorSimilarityFunction readSimilarityFunction(DataInput input) throws IOException { + int similarityFunctionId = input.readInt(); + if (similarityFunctionId < 0 + || similarityFunctionId >= VectorSimilarityFunction.values().length) { + throw new CorruptIndexException( + "Invalid similarity function id: " + similarityFunctionId, input); + } + return VectorSimilarityFunction.values()[similarityFunctionId]; + } + + private VectorEncoding readVectorEncoding(DataInput input) throws IOException { + int encodingId = input.readInt(); + if (encodingId < 0 || encodingId >= VectorEncoding.values().length) { + throw new CorruptIndexException("Invalid vector encoding id: " + encodingId, input); + } + return VectorEncoding.values()[encodingId]; + } + + private FieldEntry readField(IndexInput input) throws IOException { + VectorEncoding vectorEncoding = readVectorEncoding(input); + VectorSimilarityFunction similarityFunction = readSimilarityFunction(input); + return new FieldEntry(input, vectorEncoding, similarityFunction); + } + + @Override + public long ramBytesUsed() { + return Lucene99HnswVectorsReader.SHALLOW_SIZE + + RamUsageEstimator.sizeOfMap( + fields, RamUsageEstimator.shallowSizeOfInstance(FieldEntry.class)); + } + + @Override + public void checkIntegrity() throws IOException { + CodecUtil.checksumEntireFile(vectorData); + CodecUtil.checksumEntireFile(vectorIndex); + if (quantizedVectorsReader != null) { + quantizedVectorsReader.checkIntegrity(); + } + } + + @Override + public FloatVectorValues getFloatVectorValues(String field) throws IOException { + FieldEntry fieldEntry = fields.get(field); + if (fieldEntry.vectorEncoding != VectorEncoding.FLOAT32) { + throw new IllegalArgumentException( + "field=\"" + + field + + "\" is encoded as: " + + fieldEntry.vectorEncoding + + " expected: " + + VectorEncoding.FLOAT32); + } + return OffHeapFloatVectorValues.load( + fieldEntry.ordToDoc, + fieldEntry.vectorEncoding, + fieldEntry.dimension, + fieldEntry.vectorDataOffset, + fieldEntry.vectorDataLength, + vectorData); + } + + @Override + public ByteVectorValues getByteVectorValues(String field) throws IOException { + FieldEntry fieldEntry = fields.get(field); + if (fieldEntry.vectorEncoding != VectorEncoding.BYTE) { + throw new IllegalArgumentException( + "field=\"" + + field + + "\" is encoded as: " + + fieldEntry.vectorEncoding + + " expected: " + + VectorEncoding.FLOAT32); + } + return OffHeapByteVectorValues.load( + fieldEntry.ordToDoc, + fieldEntry.vectorEncoding, + fieldEntry.dimension, + fieldEntry.vectorDataOffset, + fieldEntry.vectorDataLength, + vectorData); + } + + @Override + public void search(String field, float[] target, KnnCollector knnCollector, Bits acceptDocs) + throws IOException { + FieldEntry fieldEntry = fields.get(field); + + if (fieldEntry.size() == 0 + || knnCollector.k() == 0 + || fieldEntry.vectorEncoding != VectorEncoding.FLOAT32) { + return; + } + if (quantizedVectorsReader != null) { Review Comment: Should we instead check if `if (fieldEntry.hasQuantizedVectors())`? This allows to use raw vector values for search if this particular field is not quantized. -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
