InsaneReader(LeafReader in, String insaneField) { super(in); this.insaneField = insaneField; ArrayList<FieldInfo> filteredInfos = new ArrayList<>(); for (FieldInfo fi : in.getFieldInfos()) { if (fi.name.equals(insaneField)) { filteredInfos.add( new FieldInfo( fi.name, fi.number, fi.hasVectors(), fi.omitsNorms(), fi.hasPayloads(), fi.getIndexOptions(), DocValuesType.NONE, -1, Collections.emptyMap(), fi.getPointDimensionCount(), fi.getPointNumBytes())); } else { filteredInfos.add(fi); } } fieldInfos = new FieldInfos(filteredInfos.toArray(new FieldInfo[filteredInfos.size()])); }
@Override public void writeField(FieldInfo fieldInfo, PointsReader reader) throws IOException { PointValues values = reader.getValues(fieldInfo.name); boolean singleValuePerDoc = values.size() == values.getDocCount(); try (BKDWriter writer = new BKDWriter( writeState.segmentInfo.maxDoc(), writeState.directory, writeState.segmentInfo.name, fieldInfo.getPointDimensionCount(), fieldInfo.getPointNumBytes(), maxPointsInLeafNode, maxMBSortInHeap, values.size(), singleValuePerDoc)) { if (values instanceof MutablePointValues) { final long fp = writer.writeField(dataOut, fieldInfo.name, (MutablePointValues) values); if (fp != -1) { indexFPs.put(fieldInfo.name, fp); } return; } values.intersect( new IntersectVisitor() { @Override public void visit(int docID) { throw new IllegalStateException(); } public void visit(int docID, byte[] packedValue) throws IOException { writer.add(packedValue, docID); } @Override public Relation compare(byte[] minPackedValue, byte[] maxPackedValue) { return Relation.CELL_CROSSES_QUERY; } }); // We could have 0 points on merge since all docs with dimensional fields may be deleted: if (writer.getPointCount() > 0) { indexFPs.put(fieldInfo.name, writer.finish(dataOut)); } } }
@Override public void merge(MergeState mergeState) throws IOException { if (mergeState.segmentInfo.getIndexSort() != null) { // TODO: can we gain back some optos even if index is sorted? E.g. if sort results in large // chunks of contiguous docs from one sub // being copied over...? super.merge(mergeState); return; } for (PointsReader reader : mergeState.pointsReaders) { if (reader instanceof Lucene60PointsReader == false) { // We can only bulk merge when all to-be-merged segments use our format: super.merge(mergeState); return; } } for (PointsReader reader : mergeState.pointsReaders) { if (reader != null) { reader.checkIntegrity(); } } for (FieldInfo fieldInfo : mergeState.mergeFieldInfos) { if (fieldInfo.getPointDimensionCount() != 0) { if (fieldInfo.getPointDimensionCount() == 1) { boolean singleValuePerDoc = true; // Worst case total maximum size (if none of the points are deleted): long totMaxSize = 0; for (int i = 0; i < mergeState.pointsReaders.length; i++) { PointsReader reader = mergeState.pointsReaders[i]; if (reader != null) { FieldInfos readerFieldInfos = mergeState.fieldInfos[i]; FieldInfo readerFieldInfo = readerFieldInfos.fieldInfo(fieldInfo.name); if (readerFieldInfo != null && readerFieldInfo.getPointDimensionCount() > 0) { PointValues values = reader.getValues(fieldInfo.name); if (values != null) { totMaxSize += values.size(); singleValuePerDoc &= values.size() == values.getDocCount(); } } } } // System.out.println("MERGE: field=" + fieldInfo.name); // Optimize the 1D case to use BKDWriter.merge, which does a single merge sort of the // already sorted incoming segments, instead of trying to sort all points again as if // we were simply reindexing them: try (BKDWriter writer = new BKDWriter( writeState.segmentInfo.maxDoc(), writeState.directory, writeState.segmentInfo.name, fieldInfo.getPointDimensionCount(), fieldInfo.getPointNumBytes(), maxPointsInLeafNode, maxMBSortInHeap, totMaxSize, singleValuePerDoc)) { List<BKDReader> bkdReaders = new ArrayList<>(); List<MergeState.DocMap> docMaps = new ArrayList<>(); for (int i = 0; i < mergeState.pointsReaders.length; i++) { PointsReader reader = mergeState.pointsReaders[i]; if (reader != null) { // we confirmed this up above assert reader instanceof Lucene60PointsReader; Lucene60PointsReader reader60 = (Lucene60PointsReader) reader; // NOTE: we cannot just use the merged fieldInfo.number (instead of resolving to // this // reader's FieldInfo as we do below) because field numbers can easily be different // when addIndexes(Directory...) copies over segments from another index: FieldInfos readerFieldInfos = mergeState.fieldInfos[i]; FieldInfo readerFieldInfo = readerFieldInfos.fieldInfo(fieldInfo.name); if (readerFieldInfo != null && readerFieldInfo.getPointDimensionCount() > 0) { BKDReader bkdReader = reader60.readers.get(readerFieldInfo.number); if (bkdReader != null) { bkdReaders.add(bkdReader); docMaps.add(mergeState.docMaps[i]); } } } } long fp = writer.merge(dataOut, docMaps, bkdReaders); if (fp != -1) { indexFPs.put(fieldInfo.name, fp); } } } else { mergeOneField(mergeState, fieldInfo); } } } finish(); }