[GitHub] [lucene] mayya-sharipova commented on a diff in pull request #792: LUCENE-10502: Use IndexedDISI to store docIds and DirectMonotonicWriter/Reader to handle ordToDoc

[GitBox <gi...@apache.org>]

mayya-sharipova commented on code in PR #792:
URL: https://github.com/apache/lucene/pull/792#discussion_r851509289


##########
lucene/core/src/java/org/apache/lucene/codecs/lucene92/Lucene92HnswVectorsWriter.java:
##########
@@ -0,0 +1,328 @@
+/*
+ * 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.lucene92;
+
+import static org.apache.lucene.codecs.lucene92.Lucene92HnswVectorsFormat.DIRECT_MONOTONIC_BLOCK_SHIFT;
+import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS;
+
+import java.io.IOException;
+import java.util.Arrays;
+import org.apache.lucene.codecs.CodecUtil;
+import org.apache.lucene.codecs.KnnVectorsReader;
+import org.apache.lucene.codecs.KnnVectorsWriter;
+import org.apache.lucene.codecs.lucene90.IndexedDISI;
+import org.apache.lucene.index.DocsWithFieldSet;
+import org.apache.lucene.index.FieldInfo;
+import org.apache.lucene.index.IndexFileNames;
+import org.apache.lucene.index.RandomAccessVectorValuesProducer;
+import org.apache.lucene.index.SegmentWriteState;
+import org.apache.lucene.index.VectorSimilarityFunction;
+import org.apache.lucene.index.VectorValues;
+import org.apache.lucene.search.DocIdSetIterator;
+import org.apache.lucene.store.IndexInput;
+import org.apache.lucene.store.IndexOutput;
+import org.apache.lucene.util.BytesRef;
+import org.apache.lucene.util.IOUtils;
+import org.apache.lucene.util.hnsw.HnswGraph.NodesIterator;
+import org.apache.lucene.util.hnsw.HnswGraphBuilder;
+import org.apache.lucene.util.hnsw.NeighborArray;
+import org.apache.lucene.util.hnsw.OnHeapHnswGraph;
+import org.apache.lucene.util.packed.DirectMonotonicWriter;
+
+/**
+ * Writes vector values and knn graphs to index segments.
+ *
+ * @lucene.experimental
+ */
+public final class Lucene92HnswVectorsWriter extends KnnVectorsWriter {
+
+  private final SegmentWriteState segmentWriteState;
+  private final IndexOutput meta, vectorData, vectorIndex;
+  private final int maxDoc;
+
+  private final int maxConn;
+  private final int beamWidth;
+  private boolean finished;
+
+  Lucene92HnswVectorsWriter(SegmentWriteState state, int maxConn, int beamWidth)
+      throws IOException {
+    this.maxConn = maxConn;
+    this.beamWidth = beamWidth;
+
+    assert state.fieldInfos.hasVectorValues();
+    segmentWriteState = state;
+
+    String metaFileName =
+        IndexFileNames.segmentFileName(
+            state.segmentInfo.name, state.segmentSuffix, Lucene92HnswVectorsFormat.META_EXTENSION);
+
+    String vectorDataFileName =
+        IndexFileNames.segmentFileName(
+            state.segmentInfo.name,
+            state.segmentSuffix,
+            Lucene92HnswVectorsFormat.VECTOR_DATA_EXTENSION);
+
+    String indexDataFileName =
+        IndexFileNames.segmentFileName(
+            state.segmentInfo.name,
+            state.segmentSuffix,
+            Lucene92HnswVectorsFormat.VECTOR_INDEX_EXTENSION);
+
+    boolean success = false;
+    try {
+      meta = state.directory.createOutput(metaFileName, state.context);
+      vectorData = state.directory.createOutput(vectorDataFileName, state.context);
+      vectorIndex = state.directory.createOutput(indexDataFileName, state.context);
+
+      CodecUtil.writeIndexHeader(
+          meta,
+          Lucene92HnswVectorsFormat.META_CODEC_NAME,
+          Lucene92HnswVectorsFormat.VERSION_CURRENT,
+          state.segmentInfo.getId(),
+          state.segmentSuffix);
+      CodecUtil.writeIndexHeader(
+          vectorData,
+          Lucene92HnswVectorsFormat.VECTOR_DATA_CODEC_NAME,
+          Lucene92HnswVectorsFormat.VERSION_CURRENT,
+          state.segmentInfo.getId(),
+          state.segmentSuffix);
+      CodecUtil.writeIndexHeader(
+          vectorIndex,
+          Lucene92HnswVectorsFormat.VECTOR_INDEX_CODEC_NAME,
+          Lucene92HnswVectorsFormat.VERSION_CURRENT,
+          state.segmentInfo.getId(),
+          state.segmentSuffix);
+      maxDoc = state.segmentInfo.maxDoc();
+      success = true;
+    } finally {
+      if (success == false) {
+        IOUtils.closeWhileHandlingException(this);
+      }
+    }
+  }
+
+  @Override
+  public void writeField(FieldInfo fieldInfo, KnnVectorsReader knnVectorsReader)
+      throws IOException {
+    long vectorDataOffset = vectorData.alignFilePointer(Float.BYTES);
+    VectorValues vectors = knnVectorsReader.getVectorValues(fieldInfo.name);
+
+    IndexOutput tempVectorData =
+        segmentWriteState.directory.createTempOutput(
+            vectorData.getName(), "temp", segmentWriteState.context);
+    IndexInput vectorDataInput = null;
+    boolean success = false;
+    try {
+      // write the vector data to a temporary file
+      DocsWithFieldSet docsWithField = writeVectorData(tempVectorData, vectors);
+      CodecUtil.writeFooter(tempVectorData);
+      IOUtils.close(tempVectorData);
+
+      // copy the temporary file vectors to the actual data file
+      vectorDataInput =
+          segmentWriteState.directory.openInput(
+              tempVectorData.getName(), segmentWriteState.context);
+      vectorData.copyBytes(vectorDataInput, vectorDataInput.length() - CodecUtil.footerLength());
+      CodecUtil.retrieveChecksum(vectorDataInput);
+      long vectorDataLength = vectorData.getFilePointer() - vectorDataOffset;
+
+      long vectorIndexOffset = vectorIndex.getFilePointer();
+      // build the graph using the temporary vector data
+      // we pass null for ordToDoc mapping, for the graph construction doesn't need to know docIds
+      // TODO: separate random access vector values from DocIdSetIterator?
+      Lucene92HnswVectorsReader.OffHeapVectorValues offHeapVectors =
+          new Lucene92HnswVectorsReader.OffHeapVectorValues(
+              vectors.dimension(), docsWithField.cardinality(), null, null, vectorDataInput);
+      OnHeapHnswGraph graph =
+          offHeapVectors.size() == 0
+              ? null
+              : writeGraph(offHeapVectors, fieldInfo.getVectorSimilarityFunction());
+      long vectorIndexLength = vectorIndex.getFilePointer() - vectorIndexOffset;
+      writeMeta(
+          fieldInfo,
+          vectorDataOffset,
+          vectorDataLength,
+          vectorIndexOffset,
+          vectorIndexLength,
+          docsWithField,
+          graph);
+      success = true;
+    } finally {
+      IOUtils.close(vectorDataInput);
+      if (success) {
+        segmentWriteState.directory.deleteFile(tempVectorData.getName());
+      } else {
+        IOUtils.closeWhileHandlingException(tempVectorData);
+        IOUtils.deleteFilesIgnoringExceptions(
+            segmentWriteState.directory, tempVectorData.getName());
+      }
+    }
+  }
+
+  /**
+   * Writes the vector values to the output and returns a set of documents that contains vectors.
+   */
+  private static DocsWithFieldSet writeVectorData(IndexOutput output, VectorValues vectors)
+      throws IOException {
+    DocsWithFieldSet docsWithField = new DocsWithFieldSet();
+    for (int docV = vectors.nextDoc(); docV != NO_MORE_DOCS; docV = vectors.nextDoc()) {
+      // write vector
+      BytesRef binaryValue = vectors.binaryValue();
+      assert binaryValue.length == vectors.dimension() * Float.BYTES;
+      output.writeBytes(binaryValue.bytes, binaryValue.offset, binaryValue.length);
+      docsWithField.add(docV);
+    }
+    return docsWithField;
+  }
+
+  private void writeMeta(
+      FieldInfo field,
+      long vectorDataOffset,
+      long vectorDataLength,
+      long vectorIndexOffset,
+      long vectorIndexLength,
+      DocsWithFieldSet docsWithField,
+      OnHeapHnswGraph graph)
+      throws IOException {
+    meta.writeInt(field.number);
+    meta.writeInt(field.getVectorSimilarityFunction().ordinal());
+    meta.writeVLong(vectorDataOffset);
+    meta.writeVLong(vectorDataLength);
+    meta.writeVLong(vectorIndexOffset);
+    meta.writeVLong(vectorIndexLength);
+    meta.writeInt(field.getVectorDimension());
+
+    // write docIDs
+    int count = docsWithField.cardinality();
+    meta.writeInt(count);
+    if (count == 0) {
+      meta.writeLong(-2); // docsWithFieldOffset
+      meta.writeLong(0L); // docsWithFieldLength
+      meta.writeShort((short) -1); // jumpTableEntryCount
+      meta.writeByte((byte) -1); // denseRankPower
+    } else if (count == maxDoc) {
+      meta.writeLong(-1); // docsWithFieldOffset
+      meta.writeLong(0L); // docsWithFieldLength
+      meta.writeShort((short) -1); // jumpTableEntryCount
+      meta.writeByte((byte) -1); // denseRankPower
+    } else {

Review Comment:
   @LuXugang  It looks to me that we write doc IDs twice to vectorData: first in `IndexedDISI.writeBitSet` and then when writing DirectMonotonicWriter?  Can you explain why do we need to do it twice?  Why can't we just use a single method?
   
   I think, If we decide to go with `DirectMonotonicWriter`, we don't need to write doc IDs with `IndexedDISI.writeBitSet`.



-- 
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-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org